The Starter Play: How to Build a Marketing Attribution Report in 10 Minutes Using Claude + HubSpot (or your CRM)

Marketing attribution has been the thing everyone wants but it’s hard to generate. You know the conversation — “we need to understand what’s driving pipeline” — but it can take weeks to pull, clean and analyze the data to get a directionally correct answer.

Here’s the thing: if your data is in HubSpot (or any CRM that has a connector to Claude), you can have a working attribution report today that tells you what marketing activities are actually correlating with closed revenue.

Subscribe now

Before You Start

• Claude account with the HubSpot connector enabled (this is key — without the connector, Claude can’t read your data). If you are using a different CRM, just make sure that it is also connected to your Claude account and follow these same instructions).

• HubSpot access with closed/won deal data and at least some marketing activity tracked against contacts

• A rough idea of the time window you want to analyze (last 6-9 months is a good starting point)

• Deals should have contacts associated — Claude reads contact-level properties, not just deal records

Step 1: Connect Claude to HubSpot

In Claude, open a new conversation and enable the HubSpot connector from the tools panel. You don’t need to export anything or build a prompt file ahead of time. Claude reads your HubSpot data directly once the connector is active.

Step 2: Run the Attribution Prompt

Start with this prompt, adjusted for your time window:

“Look at every deal that closed won in the last 9 months. Analyze the activity of each contact associated with those deals. Determine what marketing activities generated closed revenue. Give me an executive summary with: total revenue, average deal size, time to close, and the top marketing touchpoints correlated with winning. Then break it down by deal tier if you can.”

Claude will work through your data and return a structured report — not just text. Expect an executive summary, a breakdown by marketing activity, and a recommendations section. Brant Morton at Rally got a Markov chain analysis on his first try, which he wasn’t even expecting.

Your output looks something like this:

Step 3: Iterate on the Report

Don’t stop at the first output. Keep prompting to go deeper. Try:

“Now show me first touch and last touch attribution separately.”

“Show me the most common activity sequences for your top 20% of deals by deal size.”

“What marketing activities appeared in more than 50% of closed won deals?”

Each follow-up refines the picture. The first report is a starting point, not the final answer.

Step 4: Read the Output Critically

Claude can only read HubSpot contact properties right now — not full activity timelines or custom objects. That means it’s working from fields like first conversion, most recent conversion, and MQL date. The insights are real, but they’re directional. If your report tells you “one-on-one sales emails are your top driver,” that’s probably true — and also not very actionable. Push Claude to surface the marketing signals that aren’t obvious.

Step 5: Screenshot and Document What Surprised You

Before you close the conversation, take screenshots of the sections that surfaced something unexpected. This is your working version. You’ll build on it.

What Good Looks Like

A solid first-pass report gives you: total pipeline influenced, average deal size by segment, the top 3-5 marketing touchpoints correlated with winning, and at least one insight you didn’t already know. Brant’s report surfaced that in-person dinners were showing up in enterprise deals while content downloads were driving smaller ones — a genuine signal that shaped their demand gen priorities.

Watch-Outs

Claude is reading your HubSpot data, which means it’s only as good as what’s actually logged. If your team isn’t consistent about recording activity — or if marketing touches aren’t tracked at the contact level — the report will reflect that. Garbage in, limited insight out.

Also: custom objects in HubSpot are not currently readable by Claude’s connector. If you’ve built custom conversion tracking, you won’t see it here. That’s what Version 2 is for.

The Takeaway

This would have taken a data analyst weeks to build. You did it in one conversation. The report isn’t perfect — but it’s real, it’s based on your actual data, and it gives you something to react to. That’s more than most teams have. Start here, pressure-test the output with your sales and marketing leads, and then decide what you want to go deeper on.

The Advanced Play

How to Build a Full Marketing Attribution App Using HubSpot, Supabase, and Lovable.

Claude’s HubSpot connector is a great starting point. But it has a ceiling: 200K tokens, no custom objects, no full activity timelines. If you want attribution across your entire history — every touchpoint, every channel, every contact — you need to pull the data out of HubSpot and build something on top of it.

The good news: you don’t need a product team to do it. Brant Morton at Rally built this with a $30/month Lovable subscription, a free Supabase table, and a few weekends. It’s running in production.

This is Version 2. Do Version 1 first — it’ll tell you what questions to go answer with this.

Before You Start

• HubSpot access with API permissions (you’ll be pulling data via automation, not the Claude connector)

• A Make.com or n8n account to set up the HubSpot → Supabase workflow

• A Supabase account (free tier works to start)

• A Lovable account ($30/month starter tier is enough)

• At least one quarter of closed/won deal data in HubSpot with marketing activities logged

• Basic comfort with no-code automation — you don’t need to write code, but you need to follow a workflow

Step 1: Set Up Your Supabase Data Table

Create a table in Supabase to receive marketing touchpoint data. You’ll want columns for: contact ID, company ID, event type, event date, campaign, channel, and deal association (if any). This is your attribution database. Everything else gets built on top of it.

Step 2: Build the HubSpot → Supabase Sync in Make or n8n

Create a workflow that fires every time a marketing interaction is logged in HubSpot — email delivery, form submission, ad interaction, webinar attendance, sales call — and writes it to your Supabase table. This is the key architectural decision: instead of asking Claude to re-read everything every time, you’re building a running log that never hits a token limit.

Brant pulled in: marketing email deliveries, ad interactions, sales emails, calls, form submissions, and webinars. That gave him 442,000 touchpoints — way beyond what Claude’s HubSpot connector can handle in a single session.

Step 3: Build the Lovable App on Top of Supabase

Open Lovable and connect it to your Supabase table. Start with a prompt like:

“Build a marketing attribution dashboard connected to my Supabase table [table name]. I want to see: total touchpoints by channel, attribution by model (first touch, last touch, linear, time decay, W-shaped), and a contact-level view that shows every interaction a contact had before a deal closed.”

Lovable will build the app. Iterate from there. Brant added a company rollup view (so you can see activity across all contacts at an account, not just the one on the deal), a paid ads page pulling LinkedIn and Google Ads data, and a “levers to pull” page that surfaces recommendations by attribution model.

Step 4: Add Attribution Models

The real power is running multiple attribution models against the same dataset. Start with first touch and last touch — those are easy. Then add linear and time decay. Markov chain is the most accurate but requires a Python calculation running in the background; Lovable can set it up but you’ll need to fine-tune the logic.

The point isn’t to pick one model as “right.” It’s to see where they agree and disagree — and investigate the gaps.

Step 5: Use It to Ask Better Questions

The dashboard is a question-generator, not a final answer. Use it to surface: which two or three channels show up in almost every closed/won deal, where marketing and sales activity overlap (and where they don’t), and which touchpoints appear in enterprise deals vs. smaller ones. Brant found that in-person dinners correlated with his largest deals — something that never would have surfaced in a standard HubSpot attribution report.

What Good Looks Like

A fully working version of this gives you a live dashboard with six attribution models, contact-level activity timelines, company-level rollups, and a paid ads ROI view — all from your own data, without a third-party attribution platform. Brant built Rally’s version in a weekend

Watch-Outs

Your data quality matters more here than in Version 1 — you’re running actual statistical models. If your team isn’t logging activities consistently, the models will surface noise as signal. Audit your HubSpot data hygiene before you build.

Also: Markov chain analysis requires some tuning. Don’t trust the first output of that model without cross-referencing it against your simpler models.

The Takeaway

Three months ago, standing up a marketing attribution system like this would have required a dedicated PM, a team of engineers, and months of roadmap time. Juliette Kopecky said it clearly: at HubSpot, they had an entire product team devoted to tools like this. Brant built the equivalent over a weekend. That’s not hyperbole — that’s what this moment actually is. The question isn’t whether you can afford to build this. It’s whether you can afford not to know what’s actually driving your revenue.

Special thanks to Brant Morton, GTM Ops, and Juliette Kopecky, CMO, both at Rally, for building this in public and sharing every detail.

Subscribe now

Turn your CRM export into a complete pipeline forecast and recommendations — in one prompt.

Play created by Jonathan Kvarfordt, Head of Marketing at Momentum.io. You can find more of his GTM AI plays at gtmaipodcast.com.

FOR: RevOps & Revenue Leaders | TIME SAVED: 6–8 Hours Per Week | OUTPUTS: 3 Deliverables

Subscribe now

01 — What It Does

Most revenue teams are still manually producing weekly pipeline forecasts — pulling from Salesforce, formatting spreadsheets, copying numbers into slides. The Pipeline Intelligence Engine play using Claude, replaces that process entirely. You give it your CRM data. It gives you everything you need for the forecast meeting.

The Claude skill runs three parallel analyses on your pipeline data and produces three files in one prompt:

Deliverable:

What’s Inside Pipeline Health Spreadsheet (.xlsx) Stage conversion rates, velocity metrics, deals at risk, week-over-week movement, and per-rep breakdown — all with auditable formulas.

Interactive Dashboard (.html) Funnel visualization, pipeline by rep vs. quota, coverage ratio gauge, and a sortable deals-at-risk table. Renders in any modern browser.

Coaching Priorities Document (.md/.docx) The 5 deals most likely to slip, rep performance highlights, and specific talk tracks for 1:1s — based on your actual data, not templates.

02 — Prerequisites

Before running the skill, confirm you have the following:

• CRM data access — A CSV or spreadsheet export from Salesforce, HubSpot, or any CRM. Alternatively, connect via CRM connector or MCP integration.

• Pipeline export — Include active opportunities with: deal name, stage, amount, close date, owner, created date, and last activity date. The more fields, the richer the output. If you have your CRM connected, it can pull directly from your CRM. If not, you will want to include this in your export.

• Quota context — Your current quarter’s quota target and rep roster. Add these as a reference file in your project folder — the skill gets significantly smarter with this context.

• Skill installed — The Pipeline Intelligence Engine skill folder must be uploaded to Claude Cowork. See Setup below. You can find the skill here.

Watch out: If your pipeline export is missing close dates or rep assignments, the skill will flag the gaps and work with what’s available — but stage velocity and rep analysis will be limited. Clean data going in means better coaching insights coming out.

Download Project Skill

03 — Setup

Set up takes about five minutes. You only need to do this once.

Install the Skill

1. Open Claude Cowork and go to Settings > Capabilities > Add Skill.

2. Upload the skill folder — the Play-01-Pipeline-Intelligence-Engine directory.

3. Confirm it appears in your active skills list.

Prepare Your Project Folder

Create a dedicated folder for your pipeline work. Place the following files inside:

• Your pipeline data export (CSV or .xlsx from your CRM)

• Your quota sheet or forecast template — optional but recommended

• A rep roster with names and quota targets — optional but recommended

Connect Your CRM (Optional)

If you prefer live data pulls over manual exports, connect Salesforce or HubSpot via the built-in connector or MCP integration in Cowork settings. Once connected, the skill can pull pipeline data directly — no export required.

04 — How to Use

Once the skill is installed and your data is ready, the process runs in five steps.

Step 1 — Point Cowork at Your Folder

Select your pipeline project folder in Claude Cowork. This tells the skill where to read data from and where to save the three output files.

Step 2 — Trigger the Skill

Use any of these prompts to start the analysis:

“Run the Pipeline Intelligence Engine on my pipeline data”

“Analyze my pipeline”

“Create a pipeline health spreadsheet, interactive dashboard, and coaching priorities doc from this export”

Step 3 — Answer the Setup Questions

The skill will ask a short set of questions on the first run. Have these ready:

• What CRM or data source are you using?

• What is your current fiscal quarter and quota target?

• Do you want analysis by rep, by team, or both?

• What is your average sales cycle length?

• Any specific deals or segments you want highlighted?

You can skip any questions you have already covered in your prompt or project folder context.

Step 4 — Review Your Three Deliverables

The skill generates all three files in parallel and saves them to your project folder with today’s date in the filename:

• pipeline-health-[date].xlsx

• pipeline-dashboard-[date].html

• coaching-priorities-[date].md or .docx

Review each file. If any section needs refinement — different chart format, additional rep breakdowns, more specific coaching talk tracks — ask for edits in natural language. The skill will update the relevant file.

Step 5 — Make It Recurring

Once you are satisfied with the output, say:

“Save this as a weekly workflow” or “Package this as a plugin my whole RevOps team can use”

The skill will configure the workflow to run automatically — typically every Monday morning. Your team can trigger it the same way without needing to re-enter context each week.

05 — Example Use Cases

Weekly Pipeline Review

Export your Salesforce pipeline on Monday morning, drop it in the folder, and trigger the skill. In 5 minutes, you have your health spreadsheet for the forecast call, an interactive dashboard for the CRO, and coaching talking points for every manager’s 1:1. No manual formatting, no Vlookups.

Sales Leader Forecast Prep

Before your weekly forecast meeting, run the skill on the latest pipeline pull. The coaching priorities doc tells you exactly which deals to probe and what questions to ask each rep. You walk in knowing where the risk is — not finding out in the meeting.

Board-Ready Pipeline Metrics

The board wants pipeline health data. Run the skill and hand them the dashboard — funnel conversion, coverage ratios, and velocity from your actual data, not a static slide from last month.

New RevOps Hire Onboarding

Share this skill with a new RevOps team member. They generate the same reports the senior team produces on day one. No training required on which spreadsheets to pull or how to calculate pipeline velocity.

Multi-Segment Analysis

Run it on enterprise pipeline one day, mid-market the next. Compare the dashboards side by side to see where conversion rates differ and which segment needs different coaching.

06 — Best Practices

Add Context to Your Folder

The richer your input, the richer your output. Before triggering the skill, drop in your quota sheet, rep roster, and last quarter’s pipeline data for comparison. A 10-minute setup saves you an hour of back-and-forth on the outputs.

Verify Formulas on First Run

All calculations in the pipeline health spreadsheet include auditable formulas. On your first run, spot-check two or three numbers against your CRM directly. Once you trust the math, you can move faster every week.

Iterate in Natural Language

If a chart is not quite right, or you want the deals-at-risk table sorted differently, just say so. The skill responds to plain-English requests: “Sort the deals at risk by amount descending” or “Add a tab that breaks out pipeline by close month.”

Set Up the Recurring Workflow

The skill saves 6–8 hours per week — but only if you actually run it every week. Setting up the recurring workflow removes the activation energy. Schedule it for Monday morning, before your first meeting. The outputs will be waiting for you.

Handle Edge Cases Proactively

If you are working with fewer than 20 deals, the skill shifts automatically from statistical analysis to deal-by-deal review. For multi-segment pipelines, decide upfront whether you want combined or separate analysis — combined is faster, separate is more precise for coaching.

Watch out: Never use placeholder or synthetic data with this skill. It is designed to work from your actual pipeline — and will flag if it detects anything that looks like demo data. Coaching insights that don’t come from real deals aren’t useful coaching insights.

Master Checklist

Setup

• Install the skill — Upload the Pipeline Intelligence Engine folder to Claude Cowork via Settings > Capabilities > Add Skill.

• Prepare your folder — Add your CRM export, quota sheet, and rep roster.

• Connect your CRM — Optional — link Salesforce or HubSpot for live data pulls.

Each Run

• Export pipeline data — Pull the latest from your CRM, or let the connector do it.

• Trigger the skill — Say “Analyze my pipeline” or “Run the Pipeline Intelligence Engine.”

• Answer setup questions — CRM source, quota target, analysis scope, sales cycle length.

• Review three deliverables — Spreadsheet, dashboard, coaching doc — one pass before the forecast meeting.

• Iterate if needed — Ask for adjustments in plain language.

Ongoing

• Save as recurring workflow — Set it to run every Monday morning.

• Spot-check formulas — Verify calculations against source data on first run of each quarter.

• Read the coaching doc before 1:1s — Not during — give yourself time to internalize the talking points.

Subscribe now

Every year, go-to-market teams go through the same painful ritual: spreadsheets, arguments, guesswork, and a territory plan that nobody fully trusts. I sat down with the GTM team at Forter to understand how they used AI to completely reimagine their territory planning process. Forter is a global fraud prevention platform serving 500,000 merchants and 2B consumers, processing $500B GMV in 2025 alone.

I spoke with Ozge Ozcan, CRO; Parker Tipton, Head of Field Ops; Gabby Upson, RevOps; and Kris Alspach, Head of Revenue Operations, about the challenge they faced, what they built, and what it means for anyone trying to scale a GTM operation with a lean team.

The Problem: A Global Business Without a Defined ICP

When Ozge took the CRO role, and Forter appointed its new CMO, Jason Grunberg, they started to assess foundational components of the company’s GTM strategy:

Where are we actually winning?

And how can we better align our Sales, Customer, and Marketing teams to accelerate growth?

“It has been a while since we took a step back and asked what are our ICP win rates,” Ozge told me. “Where are we winning? Where are we experimenting? And where are we not winning, but we think we are — because of research bias?”

Kris, who joined in October, walked into the middle of annual planning and quickly recognized the gap. “This is a global business, and we haven’t defined the ICP,” he said. “Territory carving strategy didn’t exist before Parker and Gabby started this initiative.”

Parker added: “ICP was insider knowledge, not a system. Territory design was a manual process and trust within the organization had really eroded.”

The data situation was just as fraught. Forter’s thousands of accounts lived across disparate systems. Industry classifications had ballooned to 65+ variations. Revenue banding came from conflicting sources. And the enrichment tool they were relying on — SimilarWeb — was only covering a small fraction of all accounts, leaving a sizeable account gap with no data at all.

And the timing? They discovered all of this in the second half of the year, with annual planning deadlines looming. Territories needed to be delivered in three weeks.

What They Were Doing Before

Before this project, Gabby had owned the territory process for nearly three years. “I don’t know if I would even call it a process. Territories were ad hoc — when reps needed their books replenished, when someone was dissatisfied, when we had new headcount. It wasn’t a process.”

The workflow was entirely spreadsheet-based. Each time something changed, a new spreadsheet got created. There was no scoring model, no defensible assignment logic, no global consistency across Americas, EMEA, and APAC. Reps didn’t trust their territories, and the RevOps team spent enormous amounts of time manually rebalancing account lists without being able to clearly explain the reasoning behind decisions.

“It was always a new spreadsheet every single time one thing changed,” Gabby said. “Messy, challenging, always ad hoc.”

How They Built an AI-Powered Territory Planning Workflow

What Parker and Gabby built over the next several weeks was a fully AI-assisted ICP and territory design process. Here’s exactly how they did it.

Step 1: Pull All Account Data and Audit the Gaps

The first move was getting everything out of Salesforce and into a working environment to understand the true state of their data. This wasn’t glamorous work — it was pulling CSVs, running analysis in Google Sheets, and mapping what they actually had versus what they needed.

“The first phase really showed us where our data inequalities were and where the gaps were,” Parker explained. The gaps were significant: inconsistent industry classifications (65+ variations), unreliable revenue banding from multiple conflicting sources, and missing e-commerce GMV data — a critical metric for a fraud prevention platform focused on transactions.

Step 2: Enrich the Data with Clay

To close those gaps, the team brought in Clay, an AI-powered data enrichment platform. They integrated Clay directly with Salesforce, pulling their entire account base in to identify and fill the missing fields.

“Clay was a really big factor for us,” Parker said. “We pulled down our entire account base and used Clay’s data providers to fill the missing space — annual revenue, GMV data, employee counts — things that were all over the place or nonexistent.”

But Clay didn’t just clean the existing list. Once they had a clearer ICP definition (more on that below), they fed those criteria into Clay to find net-new accounts that matched the profile — adding several thousand accounts to their universe. “Now we have net new accounts that we did not have in our CRM that we can now target,” Parker said.

Step 3: Build the ICP Framework — with AI as a Sparring Partner

The ICP analysis itself was primarily a manual, collaborative effort between Parker, Gabby, finance, and GTM leadership. But they used general-purpose LLMs — ChatGPT, Claude, Gemini — throughout the process as what Parker called a “sparring partner.”

“We used AI for strategic reasoning — to pressure test what we were actually thinking,” he explained. “It could come up with different analysis or different thoughts that maybe we weren’t considering, or flag when we weren’t looking at the data in a specific way.”

The output was a tiered ICP: a primary target profile and a secondary profile, defined by industry, revenue band, GMV potential, and historical win rates. This was something the company had never had before in a codified form.

Step 4: Use a General LLM to Build the Territory Assignment Prompt

Rather than writing complex territory logic from scratch, Parker uploaded all of their ICP rules, quota coverage requirements, and territory classification criteria to an LLM and used a conversational back-and-forth — often via voice, not just typing — to synthesize everything into a single structured prompt.

“I try and use the conversation piece as much as possible — talking back and forth rather than typing,” Parker said. “I find it quicker, easier, and I can multitask. Even walking the dog, I’m still working.”

The prompt that came out was extensive — covering ACV/quota coverage thresholds, enterprise vs. strategic segmentation, ICP tier weighting, acceptable territory size ranges, and regional classifications for North America, EMEA, and APAC. “This would take a human being an incredibly long time to write,” Parker said. “Using tools like this, it takes me seconds.”

He also noted something that surprised the team: “I didn’t even have to explain what Carve was to the LLM. I just said, ‘We’re using a tool called Carve.’ It already knew exactly what it does and built the prompt for it.”

Step 5: Run Territory Assignment in Carve (by Gradient Works)

Carve, an AI-powered territory design tool from Gradient Works (and also a Stage 2 Portfolio company), is where everything came together. The team uploaded their enriched account CSV, fed in the prompt from Step 4, and let Carve do what would have taken days of manual work.

Here’s how Carve works in practice:

Upload your data. Bring in a CSV of your account universe (or connect directly via Salesforce integration). Define what each column means so the AI understands your schema.

Feed in your prompt. Paste in the territory logic you developed with your LLM in Step 4. Carve accepts plain natural language — no special formatting required. Gabby noted she wrote prompts “in all lowercase letters” and the tool responded correctly.

Run multiple scenarios simultaneously. You can test different territory configurations against the same data set — adjusting coverage thresholds, adding geographies, creating “to-be-hired” patches — and see how the outcomes differ without doing any manual work.

Ask exploratory questions about your data. Query conversationally: “How many apparel accounts do we have in Washington State?” This was especially valuable for answering ad hoc strategy questions from leadership without spinning up a new analysis each time.

Review AI-generated assignment explanations. Every account assignment comes with a clear rationale — “Assigned to [rep]: Geo = Enterprise East, Industry = Retail” — so you can answer rep questions and address challenges quickly and with confidence.

One dynamic the team found particularly interesting: as Parker and Gabby each ran separate projects within Carve, the tool got progressively smarter. “The questions it was asking from Gabby’s first build to my second build were much more robust and quicker,” Parker said. “It’s educating itself as we work together as a team.”

It also builds in human judgment checkpoints — surfacing ambiguous cases and asking the user to decide rather than auto-assigning — which is a meaningful trust feature. I especially like how it asks you — ‘I can make this decision, or do you want to?’ giving the human the extra reassurance and control.

Step 6: Export, Pressure Test, and Roll Out with Data-Backed Explanations

Once territories were built in Carve, the team exported results back to spreadsheets, added visuals, and brought them to leadership and the field. Because every assignment came with an AI-generated explanation, they were able to answer rep questions and address the inevitable reactions to territory changes far more efficiently than in past cycles.

“There is the analytics part of getting territories right, and then there is the emotional part,” Ozge reflected. “Having the data to explain the logic is incredibly difficult when you’ve spent days in a spreadsheet and lost the thread of why you made each decision. The team was able to answer questions much more rapidly with the AI tool.”

The Results

The numbers are striking for what a two-person RevOps team delivered:

• 65,000+ accounts enriched, scored, and assigned with defensible, explainable logic

• First-ever codified ICP at a company with sizable revenue and customer base

• First true territory model with geo-based logic, segmentation rules, and global consistency across Americas, EMEA, and APAC

• 15x quota coverage target achieved across all account books, with documented rationale for every assignment

• Delivered in approximately 6 weeks against a hard annual planning deadline

“For a two-person team like Gabby and myself, we can now do things at a level and scale that would usually take a 5, 6, or 7-person RevOps organization,” Parker said.

To put it in perspective, when I was leading teams of 600 people, territories and assignments would take us a couple of months and three people just to build the analysis and show the data to managers, so everyone felt comfortable that the opportunity was distributed fairly. Seeing this — it’s crazy.

What’s Next for Forter’s AI-Powered GTM Stack

The team isn’t stopping here. Their next phase involves two major initiatives.

Intent Data with Common Room. Parker described rolling out Common Room post-SKO to surface account-level intent signals for sellers, connected directly to SalesLoft for outreach and Salesforce as the system of record. Reps will see a composite “BIT” score — combining ICP fit and intent signals — so that when they log in each morning, the highest-priority accounts are surfaced immediately.

AI-Powered Win-Loss Analysis. Ozge’s next initiative is to run a deep win-loss analysis using existing data — without requiring field interviews. She acknowledged the prerequisite challenge: building a culture where customer call recordings in Gong or Zoom are treated as strategic competitive assets rather than surveillance.

“Customer calls and conversations are now my competitive advantage data points,” she said. “That’s a mindset shift for the field — and it’s the next hurdle we need to go through.”

Three Lessons for GTM Leaders

1. Clean data is a prerequisite, not an outcome. AI amplifies what you have. The team’s first investment was in data quality — and it unlocked everything that followed. If your data is broken, the AI output will be too.

2. A defined process beats a smarter tool. You need data that’s accurate and a workflow or process already outlined. AI doesn’t just come in and do all those things for you. The ICP framework gave Carve the rules it needed to make meaningful decisions.

3. Use LLMs as thinking partners, not just execution tools. Some of the most valuable AI use in this workflow wasn’t automation — it was using ChatGPT, Claude, or Gemini to pressure test assumptions, identify gaps in reasoning, and translate complex business logic into a clean, structured prompt. That’s a skill every RevOps leader should develop.

Subscribe now

Launching a new product is hard. Launching a new AI product, alongside new pricing and packaging, is even harder.

That was the challenge facing DataSnipper last year. In just one quarter, the company rolled out:

• Its first generative AI product, DocuMine

• A fully new pricing and packaging model

• Company-wide changes impacting sales, marketing, product, legal, and security

All while Carrie Bosworth, CRO, and Leah Levitte, Director of Enablement, were still relatively new in their roles.

Many teams would have solved this with volume: more training, more documentation, more tools. Instead, Carrie and Leah decided not to overwhelm the go-to-market team and make a different bet: use AI to give reps confidence at the exact moment they needed it.

Here’s the step-by-step playbook Carrie and Leah used, and how other teams can replicate it.

Step 1: Acknowledge the Real Problem (It’s Not “More Training”)

DataSnipper didn’t approach selling a brand-new product as a content gap.

By the time of launch, reps already had all the content they needed:

• Live trainings and kickoffs

• Manager enablement sessions

• Product certifications

• Long-form playbooks

• Call recordings

• Office hours

People don’t need more information,” said Leah. “They need help synthesizing it and making it relevant to their day-to-day.”

This is the first crucial insight: At scale, enablement breaks down not because reps lack information, but because they lack confidence in applying it.

Step 2: Redefine Enablement as “Training Wheels,” Not Mastery

Carrie framed the goal with a simple analogy: training wheels. Training wheels aren’t about perfection. They’re about reducing fear, enabling early wins, letting people practice without embarrassment, and building muscle memory over time.

The goal isn’t to replace rep judgment. It’s to get reps comfortable enough to show up confidently in front of customers, especially when the topic is something they are unfamiliar with, such as an AI product and new pricing.

This mindset shaped everything that followed.

Step 3: Start With One High-Impact Use Case

Rather than boil the ocean, DataSnipper focused on one product and one moment: Helping reps prepare for customer conversations about DocuMine.

That clarity and focus mattered. They didn’t try to reinvent the entire enablement stack, replace every workflow, or build an AI sales platform.

They focused on just one problem: “How do I prepare for this specific customer call with confidence?”

Step 4: Build a Custom GPT Around Your Actual GTM Reality

The core of the DocuMine enablement playbook was a custom internal GPT, trained specifically on DataSnipper’s GTM reality: product positioning, objection handling, pricing and packaging, vertical context, discovery questions, and points of view for first calls, demos, and negotiations.

What made it work wasn’t “AI.” It was specificity.

Instead of giving reps another 30-page playbook, Leah built a “call prep assistant” that responded in the exact format reps needed, for the exact situations reps faced.

Here’s how to replicate it step by step.

A. Start with the moment of uncertainty (not the entire enablement universe)

Don’t try to build an assistant for “everything sales might ask.” Pick one repeatable moment, like:

• “Help me prepare for a first call with a [vertical]”

• “I’m going into a pricing conversation, what are the landmines?”

• “I’m demoing to a CFO, what should I emphasize?”

Your goal is to reduce fear and increase preparedness right before customer conversations, when reps are actually going to use the tool.

B. Define the output format first (this is the secret)

Most internal GPTs fail because they return generic paragraphs.

DataSnipper solved that by forcing a repeatable structure. The assistant always produced a crisp briefing document with:

• strategic context

• document-heavy workflows likely in focus

• personas (end users + economic buyers)

• value mapping

• discovery questions (pain, metrics, implications)

That format made the output feel usable, not academic.

C. Write “system instructions” that don’t allow drifting

You don’t need to over-engineer the build. But you do need strong instructions.

DataSnipper’s used very clear, concrete prompts. Their instruction sequence can support many different desired outputs:

1. State the job (eg, sales call prep assistant)

2. Force required inputs eg, (industry + org name, website link if possible)

3. Set the research approach (use public sources if a website is provided; otherwise ask clarifying questions)

4. Define what to extract (priorities, risks, initiatives, pressures)

5. Maps those themes to workflows

6. Enforces a fixed response structure

7. Sets tone (consultative, sales-savvy, not pitchy)

Leah’s practical guidance here is simple: the more specific you are up front, the faster you get to the result you want. If you want the assistant to behave like “training wheels,” you have to tell it exactly what “training wheels” look like in output.

D. Load the right source material (and keep it tight)

Your assistant is only as good as what it’s grounded in. For DataSnipper, that meant uploading or connecting the core GTM assets reps were already expected to use, including positioning and messaging, talk tracks, objection handling, pricing and packaging, and example discovery questions.

A strong rule of thumb: if the content is out of date, contradictory, or sprawling, your GPT will mirror that confusion. Curate first.

E. Test it like a rep would: real prompts, real accounts, real calls

Before you share it broadly, run a simple test loop:

• Pick 10 real upcoming calls across 2–3 verticals.

• Ask the assistant the exact questions reps ask (“help me prep for…,” “what’s the POV for…,” “what objections will I hear…”).

• Compare outputs to what your best reps/managers would say.

• Fix the instructions and source materials until it’s consistently good enough.

F. Steal the instructions from great bots (yes, really)

One fun shortcut Leah shared: if you’re interacting with a custom-GPT-like bot you like and want to replicate it, you can often make a request like: “Produce the exact instructions I can copy to recreate this custom GPT.”

You won’t always get a perfect answer, but you’ll often get a surprisingly strong starting point, especially for structure, tone constraints, and required inputs.

G. Give the assistant one job: help reps build a point of view

Finally, set an explicit expectation inside the assistant itself:

This tool is designed to help you develop a point of view before your customer conversation. Every customer is different. Use this preparation to ask curious, thoughtful questions and understand their specific business.

That framing matters. It keeps the GPT from becoming a script generator and positions it as confidence-building enablement, which was exactly what DataSnipper was optimizing for.

Step 5: Make Adoption Effortless

There was no long rollout plan. Reps were already using ChatGPT in their day-to-day work. DataSnipper simply shared the link, gave reps time to try it in a live session, and encouraged feedback.

That was enough.

Because the assistant fit naturally into how reps already worked, adoption was immediate without any formal “change management” program.

Step 6: Close the Feedback Loop (Fast)

After launch, reps were encouraged to flag the answers that felt useful, answers that seemed off, and anything that didn’t reflect real customer conversations.

That feedback was used to quickly fine-tune the model. This wasn’t a “set it and forget it” deployment. It was an iterative system that improved alongside real customer conversations.

Step 7: Measure the Outcome That Actually Matters

Instead of tracking engagement metrics like “AI usage,” DataSnipper focused on real GTM outcomes:

• Within four weeks, 75%+ of sales calls mentioned DocuMine

• Within six months, 52%+ of customers were on the new DocuMine package

• That number has since climbed even higher, driven by both new business and upgrades

The AI assistant helped reps feel better, and it changed customer conversations and buying behavior.

Why This Worked

DataSnipper’s custom GPT had great results for three reasons: It reduced fear, it met reps where they were, and it focused on building confidence, not automation.

DataSnipper’s reps didn’t need to become overnight experts on the new DocuMine product, they just needed help preparing. And to that end, the tool was designed to answer real questions reps ask before real calls. As Carrie put it, “We still need really confident humans behind the conversation.”

How to Replicate This at Your Company

Companies of every size can copy this playbook.

Start small:

1. Pick one product or initiative

2. Identify the one moment where reps feel the most uncertainty

3. Build a lightweight AI assistant trained on your GTM reality

4. Make it easy to use

5. Iterate based on real feedback

The goal isn’t AI everywhere. The goal is better conversations, sooner.

Subscribe now

For years, go-to-market strategy assumed a relatively stable buyer entry point: search.

Founders invested in SEO, content teams optimized for keywords, and awareness was measured in rankings and clicks. That model quietly broke.

At the Stage 2 Capital Summit, Sydney Sloan, CMO Advisor to G2, laid out why we are now in a different era entirely. Buyers increasingly begin their journeys inside large language models, not search engines, where visibility is determined not by who ranks, but by who gets cited.

This shift has implications well beyond marketing. It changes how companies build awareness, how trust is established, and how early-stage GTM strategy must evolve in an AI-mediated buying environment.

What follows are five themes from Sydney’s talk that every founder, revenue leader, and marketer should be internalizing now

1. Buyers Are No Longer Starting with Search, They’re Starting with Answers

The most important insight from Sydney’s talk is also the simplest: buyer behavior has changed, and it changed fast.

“In August, 50% of buyers said they start their search on an LLM,” she said. “In April, that was 29%. In four months, that changed 71%.”

This is not a marginal shift. Buyers are no longer discovering products by typing keywords into Google and clicking through content. They are asking LLMs complex, contextual questions and expecting synthesized guidance.

From a GTM perspective, this matters because awareness is no longer driven by traffic alone. If your company does not appear in the answers buyers are receiving, you may never enter the consideration set at all.

2. The Objective Has Shifted from Winning the Click to Winning the Answer

Classic SEO was built around a clear goal: rank highly and earn the click. This is no longer sufficient.

“We’re moving from this era of ‘I need to win that click’ to actually winning the answer,” Sydney explained.

Answer engine optimization (AEO) reflects this shift. Rather than optimizing for individual keywords, companies must optimize for the questions buyers are asking, and for being included in the responses LLMs generate.

This reframes content strategy from a traffic exercise into a trust and relevance exercise. The question is no longer “How do we rank?” but “When buyers ask real questions, do we show up as part of the answer?”

3. AEO Is a Real Category, and It’s Forming at Breakneck Speed

Sydney was explicit in distinguishing between closely related terms: “AEO is about ensuring your product shows up when a question is being asked. GEO is about generating content.”

That distinction matters, especially given how quickly AEO has become its own category.

“In April there were seven vendors; now there are over 100,” Sydney said.

The speed of category formation signals both urgency and noise. For most companies, the opportunity is not to build tooling, but to understand how AEO fits into their broader GTM motion.

Founders should treat AEO the same way they once treated SEO: not as a tactic, but as infrastructure for awareness in the dominant discovery channel of the moment.

4. LLMs Reward Consensus, Not Isolated Optimization

One of the most operationally important insights in Sydney’s talk is how LLMs actually determine what they surface.

“LLMs work on consensus,” Sydney said.

LLMs scan across many sources to determine the most consistent, credible answer. That means visibility is no longer driven by a single high-ranking page, but by coherence across your entire digital footprint.

“You want to ensure your content is consistent,” she said. “This includes how you describe yourself on your website, in partner portals, on G2.”

From a GTM standpoint, this pushes companies toward a more integrated view of brand, content, reviews, and distribution. Inconsistent positioning across channels does not just confuse buyers; it reduces the likelihood that LLMs will surface you at all.

5. Awareness in the AI Era Is Built on Breadth, Consistency, and Human Language

Finally, Sydney emphasized that AEO is not about gaming a new algorithm. It is about answering real buyer questions clearly and consistently, in human language, across many surfaces.

She encouraged teams to build content around personas and jobs-to-be-done, even when that content does not directly promote the product.

“Even if it’s not what you do, that’s how you build your brand,” Sydney said.

This mirrors what we see repeatedly in successful GTM strategies: durable awareness comes from being genuinely helpful before it comes from being persuasive.

What This Means for Go-To-Market Leaders

AEO is not just a marketing concern. It sits upstream of pipeline, influencing which vendors buyers ever learn about in the first place.

As AI becomes the default interface for research and discovery, go-to-market teams must rethink how awareness is earned. Winning in this environment requires:

• Understanding the real questions buyers are asking

• Ensuring consistent positioning across owned and third-party channels

• Investing in breadth, credibility, and distribution, not just rankings

As Sydney put it:

“There is no playbook. You just have to stay in the playground.”

For founders and operators building in this moment, AEO is quickly becoming a core GTM capability, not because it is new, but because buyer behavior has already moved on.

Subscribe now

Every company wants deeper insight into why they win (and lose) deals. But actually getting that insight is a different story…and a consistent question we ask in the board room. Where are we winning and why? What can we learn that we can apply to the process or product? And can we leverage AI to get better insights faster?

I asked the CMO of Rally, Juliette Kopecky, to work on a project with me to test using AI to build this process in a way that is reliable, repeatable and actually useful. This process can also be used to understand why you lose deals, too, and ultimately combine those learnings to forecast more accurately.

Below is a recap - what wasn’t working, what Rally did with AI, and what they learned so you can implement this in your own organization.

The GTM Challenge: Win-Loss Is Critical… But Hard to Do Well

Win-loss analysis has always been a high-leverage GTM motion. It tells you why customers buy, where your product differentiates, and what is working in your process.

But despite its importance, the traditional process is riddled with problems:

1. The reasons behind wins and losses are complex.

No two deals are alike. Context, champions, competitors, timing, procurement, pricing—so many factors influence outcomes that it’s difficult for humans to abstract meaningful patterns at scale.

2. There is bias on both sides.

Buyers don’t always tell you the full truth and sellers have blind spots and emotional investment in the deal.

3. It is extremely time-consuming.

One-on-one interviews take hours to schedule, conduct, transcribe, and analyze. Most companies only do a handful per quarter—far too few to spot meaningful trends.

What Rally Tried: A Three-Phase AI-Augmented Win Analysis Workflow

To tackle the challenge, Rally broke the project into three phases—each building on the last.

Phase 1: Can AI replicate the win interview using sales calls alone?

Instead of relying on scheduled buyer interviews, Rally asked:

If we feed AI the sales call recordings, can it reconstruct what a proper win-loss interview might uncover?

They processed large volumes of sales conversations from all their won deals in a quarter and evaluated whether AI could extract insights similar to human-led interviews. Rally uses Gong to record their calls, so they tested Gong’s built in AI analysis to summarize the calls. They uploaded their own list of prompts so that the summaries were consistent from call to call.

What they learned: AI is surprisingly strong at pulling structured information out of unstructured call data. It can summarize, classify, and surface themes quickly—producing a baseline “interview-like” assessment without ever speaking to the customer.

Phase 2: Can AI identify trends across deals?

Once AI could summarize individual sales calls, the next question was scale:

Could AI detect patterns across dozens or hundreds of deals?

This phase involved aggregating insights across call transcripts—things like:

• What customer pain points resonated the most?

• What product features did customers want?

• How did the product compare to their current solution and/or competitor offerings?

• What are key steps in the process for buyers? What roles need to be involved?

Rally uploaded these summaries to build their own GPT in ChatGPT so they could run the insight analysis across all of the call summaries. They did this manually, but realized they could also automate it in the future given that Gong has an API into ChatGPT.

What they learned:
This is where AI shined. The model could sift through massive volumes of interactions and surface repeatable patterns much faster than any manual process. Rally could take these learnings and apply them back to the sales process so that they could ensure that they were following all the right steps to have the best outcome. For example, the majority of the won deals were focused on doing research with their own users, improving governance and compliance in their research practice, and speeding up the research recruitment process, so they have prioritized these use cases in discovery.

Phase 3: Can AI predict outcomes and recommend actions?

The ultimate ambition:

If AI can understand the past, can it help shape the future?

This phase can test whether AI could not only analyze trends but also predict deal risk, flag competitive steps, suggest next steps, and recommend product improvements. Rally is testing this phase now to see if they can run the first 2-3 calls once a deal reaches a certain stage to determine the likelihood of closing. To do this, you can build another custom GPT as a Forecast and load in the several attributes and prompts you want it to detect in the call. It can give you the results, as well as a forecast.

The Outcome: A Hybrid Approach Is the Winning Formula

After running all three phases, Rally reached a clear conclusion:

AI is incredible for helping us with:

• Speed

• Breadth

• Pattern recognition

• Synthesizing large amounts of data

But AI alone is not a replacement for human-led conversations.

The richness, nuance, and authenticity of a live customer interview still matter—especially for complex buying journeys.

So what’s the right approach?

AI as the starting place for analysis, humans for depth.

AI dramatically improves the win-loss process by:

• giving teams the ability to analyze more data, more often

• preparing interviewers with a stronger context

• filling insight gaps when a buyer won’t take a call

• generating trendlines that humans simply cannot see

This hybrid model creates better conversations, better enablement, and better GTM decision-making, all while saving teams enormous time.

What This Means for GTM Teams

Win-loss analysis used to be something companies did quarterly because of the time commitment. With AI, it can be continuous. And when paired with human depth and judgment, it becomes one of the most powerful GTM tools available. This project showed that the future of GTM isn’t AI or humans—it’s AI-powered humans making better, faster, more confident decisions.

My challenge to you: Run this analysis for both deals you won and deals you lost on your Q4 data. Include two slides in your Q1 Board Deck:

1. Why we win deals

   • What our CRM says

   • What AI says

   • What we learned

2. Why we lose deals

   • What our CRM says

   • What our AI says

   • What we learned

If you want to share, I would love to hear about it!

This year, nearly every founder I spoke with was experimenting with AI in their go-to-market motion.

Very few were doing it the same way.

Across the GTM AI articles I’ve written this year, a handful of clear patterns emerged. Not hype cycles or shiny tools, but practical shifts in how strong teams are building, operating, and scaling GTM.

Here are five that stood out.

Subscribe now

1. GTM Is No Longer a Static Playbook

The most effective teams aren’t “setting the playbook and scaling it.”

They’re building GTM systems that adapt in real time by adjusting targeting, sequencing, and prioritization as signals change.

Teams like MedScout stood out this year, not because they automated everything, but because they used AI to continuously pressure-test their assumptions: which accounts mattered, which signals were meaningful, and where humans should stay involved.

The takeaway for founders is simple: Your GTM advantage comes from learning faster than your competitors, not documenting better slides.

2. The CRM Is Fading Into the Background

CRMs are still important, mostly as systems of record.

The work itself is increasingly happening elsewhere, with AI handling data capture, enrichment, routing, and follow-up in the background.

The strongest teams I saw this year weren’t asking, “How do we get reps to update the CRM?”
They were asking, “Why is a human doing this at all?”

It’s becoming table stakes to use a company like Momentum to get insights into your GTM strategy and pipeline by capturing all the data at the point of action, which alleviates the rep having to input it into the CRM. In this case, the data is being automatically transitioned to the CRM and the CRM has become a more accurate database.

For companies that want to take it a step further, Day AI, one of our Stage 2 portfolio companies, has created an AI-native CRM that provides instant pipeline visibility without manual input, proactive follow-up reminders, automatic meeting context, and daily sales briefings. Unlike traditional CRMs, Day AI is designed to effectively fade into the background, reducing the burden of admin while empowering companies with the information they need.

This shift alone is giving sales teams more time for selling, marketing, and decision-making all without having to add headcount.

3. The Front Door of the Funnel Has Moved

Buyers are no longer discovering companies in simple, predictable ways, like Google ads and organic search.

Between LLM-driven answers, AI search interfaces, and non-linear inbound paths, discovery is fragmented. It’s also a lot harder to measure with traditional tools.

Founders who adapted fastest treated inbound, SEO, and conversion not as discrete channels, but as one integrated problem: How do buyers find us, understand us, and decide to engage before they ever talk to sales?

Momentum is a great example of a company that is tackling this challenge head on in a number of creative ways, including a robust content marketing program.

Those who didn’t are already feeling the gap.

4. AI Works Best When Strategy Is Already Clear

One of the most consistent signals this year: AI amplifies clarity.

Teams with a sharp ICP, strong GTM judgment, and intentional sequencing saw real leverage from AI. Teams without those things found that AI can just create confusion.

Companies like QuotaPath show what’s possible when AI is used to support well-defined decisions (not replace them).

AI doesn’t fix fuzzy strategy. It exposes it.

5. Lean Teams Are Now Running Enterprise-Grade GTM

Perhaps the biggest shift this year: what small teams can execute.

With off-the-shelf AI tools, early-stage companies are now running outbound, inbound, and campaign motions that once required full SDR teams and dedicated ops functions.

Companies like Tofu are building technology where sophistication is no longer gated by headcount. And companies like Rally are creating enterprise-grade workflows with very small teams and just a few tools.

The bar has moved — and that’s good news for founders willing to design AI-native workflows early.

Looking Ahead

The teams that stood out this year weren’t chasing trends.

They were asking better questions about where humans add value, where AI can make an impact, and how GTM systems should evolve as markets change.

This mindset is what I expect to define the next phase of GTM as we go into 2026. Those that aren’t embracing that roles and systems are changing will be left behind.

Subscribe now

Subscribe now

If you’re a marketing team of one at a SaaS or tech company, you know the daily math: big goals, small bandwidth. You’re expected to plan, launch, and measure multi-channel campaigns for different personas while juggling product launches, content calendars, and a founder’s next idea.

The good news is that AI is making it possible for a small team to market like a big one. The challenge is knowing how to use it and where it genuinely moves the needle.

I recently sat down with Elaine Zelby, Co-Founder and CRO of Tofu, a Stage 2 portfolio company, to explore how marketers are scaling campaigns in the AI era. Elaine’s approach is refreshingly grounded: use AI to multiply execution, not to replace strategy.

1. From “One-Size-Fits-All” to “One-Size-Fits-One”

“Everything is about relevance and context,” Elaine said. “Historically, marketing was one-size-fits-all. But really, one size fits one. That’s what marketing has always aspired to be, and now it can.”

For years, personalization meant little more than a {First Name} token in an email. Today, AI can tailor every asset, including emails, ads, decks, even landing pages, to an individual’s industry, behavior, and history with your brand.

In Tofu, a feature called Deep Research scans a company’s full catalog of marketing materials (case studies, webinars, call transcripts) to surface insights instantly. Imagine typing: “Find all case studies that focus on performance. List the customer, industry, quotes, and three highlights.”

Within seconds, it compiles the results. For a solo marketer, that’s days of research condensed into minutes, and a foundation for relevance that goes far beyond copy tweaks.

Even without a dedicated platform like Tofu, marketers with a shoestring budget can replicate this type of automated research using tools such as ChatGPT’s file-analysis and search capabilities.

Let’s say you have a folder of case studies, sales call transcripts, and webinar summaries, a typical mix for a lean SaaS marketing team. You could upload those files directly into ChatGPT and then prompt it:

“Search across all my uploaded materials and find every case study that mentions performance improvements as a key outcome. For each one, list:

• The customer name and industry

• The problem they solved

• Any quantitative results (e.g., time saved, conversion lift, uptime)

• Three notable quotes or takeaways.

In a few seconds, ChatGPT would extract that data across dozens (or hundreds) of documents and turn unstructured archives into a structured research report.

You could then follow up with iterative prompts to narrow your insights:

• Of those performance-related case studies, which industries show the largest ROI gains?

• Summarize how our messaging around performance has evolved between 2023 and 2025.

• Pull exact customer quotes we could reuse in ad copy or sales decks.

Founder takeaway: Personalization isn’t about pandering; it’s about precision. The marketer who builds campaigns around context wins the next era of GTM.

2. AI Is Not a Strategy, It’s a Multiplier

Elaine’s biggest warning: “AI isn’t a migrator or a magic button; it doesn’t deliver strategy.”

AI can execute, but it can’t decide why your message matters. The human still defines the story; AI just turns it into 27 assets in the time it used to take to write one.

For example, Tofu can take a single messaging doc and instantly create:

• A blog post

• Three LinkedIn posts for different personas

• A Google Ad set

• A pitch deck

• A personalized email sequence

Each variant respects your tone, structure, and brand rules. The marketer defines the direction; AI scales it across channels and formats.

Tofu can learn the tone of a specific person, for example, an executive’s LinkedIn voice. It analyzes sample posts (word choice, sentence rhythm, emoji use) and builds a reusable tone profile. The result: every social post sounds like that person actually wrote it, not like “AI wrote it for them.”

For cash-strapped founders, or marketers who want to test AI as a marketing multiplier, you can use this playbook with a free tool:

Step 1: Upload Your Core Messaging Document

Start with your value proposition doc, launch brief, or positioning statement. This could be a Google Doc that articulates:

• Your core audience segments (e.g., DevOps engineers, marketing leaders, procurement)

• Your problem statement and solution

• Key differentiators and proof points

Upload that file into an AI tool and give it a clear command such as:

Use this messaging document to create the following assets for our upcoming product launch:

• A 700-word blog post that explains the launch in plain language for existing customers.

• Three LinkedIn posts: one targeting executives, one targeting technical users, and one for partners.

• A Google Ad set with 10 responsive headlines and 4 descriptions.

• A 6-slide pitch deck introducing the product to prospects.

Step 2: Generate Each Asset Type

Then, you can see how the AI tool will structure and output each deliverable in sequence, complete with section headers, tone adjustments, and platform formatting.

You can refine on the fly:

“Make the executive LinkedIn post more thought-leadership oriented.”
“Simplify the technical post to a 9th-grade reading level.”
“Reformat the pitch deck into a 5-slide investor-style deck with a single data point per slide.”

You can even ask for brand voice consistency:

“Match the tone to our existing posts: authoritative but conversational. No jargon.”

If you’ve previously uploaded sample posts or a brand guide, most AI tools can analyze those and emulate the same structure and language style (similar to TOFU’s “Tone Creation” feature).

Step 3: Package and Reuse

Once you’re happy with the results, ask the tools to bundle everything:

“Create a table summarizing these assets, their target persona, and recommended publication channel. Then export all text to a single Google Doc.”

You can also have it generate image prompts for your designer or an image-generation model:

“For each LinkedIn post, suggest a concept for a supporting image.”

Founder takeaway: Keep humans in charge of ideas; let AI handle the heavy lifting.

3. Start Small: Prove Impact Before You Scale

AI adoption shouldn’t start with a full-stack overhaul. Elaine recommends a proof-of-concept mindset. “You should always have a free trial before you buy,” she said. “Deeply identify the lowest-hanging fruit.”

Maybe that’s a closed-lost re-engagement campaign, one of the most avoided opportunities in SaaS. Marketers know they should run re-engagement campaigns, but if you have to choose between finding new leads and trying to engage people who have already said no, the new leads usually win out. But now that you can spin up a new campaign in an hour, marketers can do both.

For Elaine, closed-lost re-engagement is a great test case for Tofu’s Personalized Nurture flow:

1. Ingest call recordings to identify loss reasons

2. Generate a four-email nurture sequence that addresses those objections

3. Personalize the content by persona and company

4. Launch it automatically through HubSpot or Marketo

   

What used to take a week of coordination can now be done during a 30-minute demo call.

Another small-start idea: Database reactivation. Identify “high-fit, low-activity” leads, and run an AI-driven re-engagement sequence that references their last interaction.

Founder takeaway: The goal isn’t to automate everything. It’s to free yourself to focus on strategy by automating the 80% that repeats.

4. Expand Your Definition of ‘Campaign’

Executives often think of a “marketing campaign” as a series of LinkedIn posts or a blog article. Marketers have always known that a true campaign is a connected ecosystem of complementary assets.

Regardless of your preexisting notion of “campaign,” it’s time to think bigger. Even the most sophisticated marketing experts need to start viewing campaigns as assets that have the potential to adapt to every buyer interaction.

Case Studies from Calls: Upload a Zoom recording or transcript, and AI generates a case study following your brand’s template. It fills in missing details (headquarters, industry, metrics) from open sources, then writes narrative copy and quotes.

Decks from Webinars: Instead of manually building slides, upload the recording and let AI generate a presentation in your brand’s slide template, complete with key takeaways, graphics, and speaker notes.

Ads and Social Posts: Feed an eBook or whitepaper into the system, and it generates LinkedIn posts, Google Ads, and blog drafts that align with your brand voice and channel best practices.

Each of these use cases removes a bottleneck that previously required design, ops, or agency help.

Founder takeaway: The AI-powered campaign isn’t a collection of disconnected tasks. It’s a flywheel where research, creation, and execution all feed each other.

5. Personalization at Scale Is Finally Real

For years, account-based marketing was more of a pipedream than a reality: one-to-one experiences at scale. In practice, most teams settled for “one-to-few.” That’s changing.

One-to-One ABM Campaigns: With TOFU’s templates, a solo marketer can select target accounts from the CRM, personalize landing pages for each one, and auto-generate matching SDR email sequences and LinkedIn ads. The system uses existing brand templates and research data to populate industry-specific language and imagery.

The result is a true end-to-end ABM workflow that connects every touchpoint, from personalized page to email to ad, without needing a team of specialists.

Founder takeaway: ABM is no longer reserved for enterprises. A marketing team of one can now build one-to-one campaigns in hours, not weeks.

6. AI Is Only as Good as the Data Behind It

Elaine often reminds teams that AI’s superpower is context, but only if you feed it enough of the right context.

Think of AI as a junior marketer who learns fast but starts blank. If you train it on the wrong assets or forget to update your inputs, its output will reflect those gaps.

That’s why Elaine advocates for centralizing your content repository. Whether you use a CRM, a CMS, or a shared drive, give AI access to call recordings, eBooks, case studies, and customer data. The more you connect those dots, the smarter your system becomes.

In Practice → Data Enrichment: Tofu automatically pulls firmographic and behavioral details from 12 datasets and pairs them with your CRM records. Ask it, “Which accounts host frequent webinars?” and it will crawl the web, cite sources, and generate a list for a targeted campaign.

Founder takeaway: Data is the fuel that makes AI marketing go. Keep it clean, connected, and current.

7. Carve Out Time to Play

The best AI marketers aren’t just using tools, they’re experimenting with them. Elaine recommends leaders “force the time to play.”

“Carve out dedicated, repeated time to kick the tires on tools. At least once a month,” she says. “Stuff is changing so quickly.”

Make AI practice a habit:

• Host a monthly “AI lab day” where your team tests new prompts or models.

• Assign someone to re-audit your best performing assets with AI assistance each quarter.

• Capture and share your prompt templates internally,

AI won’t replace marketers, but marketers who don’t adopt AI thinking will get replaced by those who do.

Founder takeaway: Treat AI fluency like a muscle that will grow with practice.

8. How to Avoid the Next Wave of Shiny Objects

Elaine acknowledges the temptation to try every new AI tool. Her advice: “Focus on bigger tools that can do more things for more people on your team, the ones that get you 90% of the way there.”

The goal is to reduce tool fragmentation. Pick a few core platforms that integrate well and enable cross-team visibility. A lightweight stack might look like:

• CRM + MAP: HubSpot or Salesforce for data and automation

• AI Execution Layer: Tofu or equivalent for research → creation → distribution

• Analytics Layer: Mixpanel or Looker for performance tracking

This stack keeps your team nimble while still giving you the benefit of AI scale.

Founder takeaway: Less software, more integration.

9. Reframing AI as Creative Leverage

Perhaps the biggest shift AI introduces isn’t efficiency but creative leverage. When you can generate assets in minutes, you can spend more time testing ideas.

The marketer who launches 10 mini experiments a week will outlearn the marketer who launches one big campaign a quarter.

AI lets you:

• Turn a webinar into a deck in minutes

• Build personalized landing pages for every target account

• Launch and iterate ads with matching visuals and copy in real time

And it does so without adding headcount, just better thinking.

Founder takeaway: AI doesn’t erase your creative instincts; it amplifies them. Use that leverage to experiment fearlessly.

Final Thought: It’s Still About Strategy

The most important sentence from my conversation with Elaine Zelby was this: “AI doesn’t deliver strategy.”

That line stuck with me. Because underneath every tool, every automation, every algorithm, someone still has to decide what story we’re telling and why it matters.

If you start with a clear strategy and a point of view, AI becomes your multiplier. If you don’t, AI just helps you do the wrong thing faster.

So as you experiment with AI in your go-to-market motion, remember:

• Campaigns are now systems. They learn and adapt.

• The stack is flattening. The barrier to execution is dropping.

• Strategy is still the moat. Your insight, creativity, and taste can’t be replicated.

The marketers who win the AI era won’t be the ones who buy every tool. They’ll be the ones who understand where AI ends and where their strategic judgment begins.

Subscribe now

At Stage 2 Capital, we spend every day thinking about how go-to-market can be done better. With 50+ portfolio companies in our network, we’re constantly scanning for ways AI can accelerate execution. Over the past year, we’ve noticed a major shift: what started as hesitation (“Let’s wait and see”) has turned into urgency (“I know I should be using AI, I just don’t know where to start”).

One of the biggest questions we hear from early-stage founders is:

“Do I really need to start with an enterprise CRM, like Salesforce or HubSpot, or is there a new path forward?”

For years, these systems have been the default answer. But they come with baggage: expensive licenses, heavy setup, and the eternal struggle of getting salespeople to input data. The result? Leaders spend more time chasing updates than actually learning from the CRM.

This is where Day AI comes in.

The Case for a Native AI CRM

Unlike the well-known enterprise CRMs, Day AI doesn’t require teams to do a lot of data entry. Instead, it listens, captures, and surfaces insights from the interactions sales teams are already having — email, meetings, and calls. Instead of requiring structure before delivering value, Day AI builds structure around a team’s day-to-day work.

When we first invested in Day AI, we saw the potential for something fundamentally different: a CRM that works for humans, not the other way around. And now that early customers are using it in production, the results are compelling.

A Founder-Led Sales Team’s Breakthrough

Finch is a company that handles pre-litigation for law firms, using AI-powered software. One of its promises is to handle administrative work so law firms can serve more clients. So it makes sense that Finch would also avoid administrative work wherever possible.

Bennett Northcutt is on the strategy and operations team at Finch. “I came in tasked with building out our go-to-market operations,” he said.

The team already had a CRM but it wasn’t doing what they needed and it was hard to manage.

“We were going to a lot of conferences so we wanted to be able to easily manage lists of attendees, including who was already in the CRM, who we’ve reached out to, the opportunities that exist,” Bennett said. “And it was just kind of convoluted. We would look at it and scratch our heads wondering, ‘am I looking at a list of companies? Is this a list of objects?’ The CRM had a certain way of organizing everything and we were supposed to figure out how to make it work for us. But it just never felt quite right.”

“I assumed we would go with either HubSpot or Salesforce. Right? Everybody uses one of those.”

Bennett’s CEO encouraged him to give Day AI a try before investing the dollars and the time that an enterprise CRM implementation requires.

Within hours, their pipeline number appeared automatically — no spreadsheets, no weeks of setup, no painful migration.

“It was amazing that it just sort of appeared,” Bennett said. “We hooked up Day AI, and all of a sudden we had pipeline visibility without anyone having to manually update a thing.”

Ease of Use That Meets the Team Where They Are

Another big difference from enterprise CRMs was that Bennett and his team found there was virtually no learning curve. “It was obvious how to use it because there wasn't any action we needed to take,” he said. “We just connected it to our systems, and then everything appeared, and it was readily available. And we can always query using the chat functionality, which is very nice.”

Like many young companies, Finch’s sales data was spread across various email inboxes, including the CEO’s.

“It would have been a huge administrative burden to get data from all of the conversations that had been happening out of people’s inboxes and into a CRM,” said Bennett. “It was amazing we didn’t have to do that.”

Today, Finch is using Day AI primarily to manage pipeline. For example, Bennett can submit a query: “Is there anything outstanding for any opportunity in discovery that has fallen through the cracks?” Day AI then recommends actions like, “Make sure to follow up with Charles.”

Day AI has saved Finch’s sales team countless hours of admin work, but this wasn’t just about convenience; it was a strategic choice. Bennett explained that Finch as a company believes in taking asymmetric risks if those risks could yield a unique competitive advantage. Bennett and his team felt that the time and cost they would save by not using a legacy CRM would give them a leg up over their competitors.

By adopting Day AI, Bennett’s team gained:

• Instant pipeline visibility without manual input.

• Proactive follow-up reminders to prevent deals from slipping.

• Automatic meeting context via call recordings and notes.

• Daily sales briefings that kept the team focused on action, not admin.

And Day AI did all of this straight “out of the box.”

Why This Matters for Early-Stage Companies

At the earliest stages, sales is often founder-led. Time spent updating fields in Salesforce or HubSpot is time not spent selling. Day AI flips that equation:

• For founders: It keeps focus on customer conversations while building the pipeline automatically.

• For sales leaders: It removes the administrative tax that often drags down adoption of traditional CRMs.

• For companies: It represents an “asymmetric risk” — the chance to gain competitive edge by using a tool that competitors stuck on legacy systems don’t have.

Or as Bennett put it: “Day AI would be my number one recommendation. Let it handle your pipeline — you focus on selling.”

The Bigger Picture

AI isn’t just automating tasks; it’s reshaping workflows. CRM may be one of the best examples of this shift; but that doesn't mean it's right for every company.

For example, with a traditional enterprise CRM, the user begins the day reviewing opportunities and thinking about how to prioritize the workday. With an AI-native CRM, all of the context gathering and prioritization is handled by the agent. It creates a brief and delivers recommendations with pre-written email drafts to the prospect.

This is a fundamental change in workflow. It will be interesting to see how quickly user expectations adapt to this shift — and to what extent enterprise CRMs come to support them.

Is an AI-First CRM Right for You?

Questions to Ask

Team Size and Complexity

• Do you have a large team supporting multiple verticals that requires the depth of a traditional CRM, including sophisticated pipelines and custom reporting?

• Do you have a smaller or more junior team that would benefit most from the prioritization, speed, and automation of an AI-native CRM?

Business Requirements

• Do you and your company’s finance team need enterprise-level infrastructure and integrations today?

• Would you rather have an agile tool with fewer but quickly iterating features? (Many AI-native CRMs may not have all of the integrations out of the box, but they can grow with your company.)

• What kind of reporting will you need for board meetings, fundraising, or forecasting?

Existing Workflows and Processes

• Will your team actually use an enterprise CRM daily and see the value in it, or will it become shelfware?

• Is your customer data already housed within a CRM or is it distributed across email, product, and various spreadsheets?

• Do you have the time and resources to train your team to fully utilize new tools?

Marketers know that inbound lead conversion is throttled by delayed responses, SDR bottlenecks, and qualification friction. But with the rise of AI-native sales agents, the conversion funnel is being rebuilt for speed, precision, and scale. As we see more and more buyers using LLMs to search for solutions to their challenges, and then using vendor websites to qualify themselves, using AI to facilitate this process seems like a no-brainer.

As part of my series on GTM AI that actually works, I’m showcasing three compelling use cases in which companies replaced legacy inbound workflows with AI—and saw measurable results. While each of these companies is using a different platform based on their size and needs, they are all seeing similar results. Let’s dive in!

Subscribe now

1. Rho and Spara: Automating Speed-to-Lead in a Timing-Sensitive Business

The Challenge
Rho is an all-in-one banking platform that sells to early-stage startups where timing is critical to their survival. Think funding events: a young startup’s ability to open accounts quickly can mean the difference between dying and thriving. Before they adopted AI agents, Rho was relying on basic lead routing; they didn’t even have an SDR team. With around 100 inbound signups a day and only seven salespeople, deals were falling through the cracks due to slow response times and lack of real-time qualification. “We had no one to sit behind Intercom or Drift and qualify traffic,” said CRO Tommy McNulty. “It was very bare-bones.”

The Solution

That’s when Rho turned to Spara, an agentic AI platform that autonomously qualifies, engages, and converts leads across chat, email, and voice. The goal: filter out unqualified leads, accelerate time-to-meeting, and free up sales reps to focus on high-value prospects.

The onboarding process was fast and lightweight, which was important to Rho given their lean operation. Spara’s team scraped Rho’s site, absorbed demo content and sales collateral, and was up and running within days. Embedded directly on the site, Spara handles two main workflows:

• Demo Qualification: Triggered by “Book a Demo” CTAs, the agent asks pre-qualifying questions and passes meetings to sales.

• Intent-Based Routing: It branches conversations based on user interest (e.g., partnerships, support, or product questions) and routes inquiries accordingly.

Results

The results were impressive, especially for such a small revenue team:

• 548 meetings booked and ~1,500 qualified leads engaged in 12 months

• ~$45 cost-per-meeting vs. $570 using the previous method

• 120+ support inquiries deflected from sales

• Speed-to-lead reduced to under 4 seconds

• Booked a $40M revenue customer directly via AI chat

Insight
It’s notable that buyers appreciated interacting with an AI agent that openly identified itself—especially when it quickly answered pre-purchase questions. Spara mimicked the intent and drive of a top-performing SDR but ran 24/7, meeting customers on their timeline. In addition, the setup doesn’t require rev ops or integrations, making it ideal for companies that don’t have the resources to build and maintain a custom or more complex solution.

2. Demandbase and Qualified: Scaling AI SDRs for Inbound Pipeline Generation

The Challenge
Demandbase’s marketing team was generating inbound traffic, but converting those leads to pipeline depended on human SDRs, which caused delays due to working hours, ramping, and attrition. Demandbase implemented Qualified in 2021. While they did see an improvement to customer conversion rates with better response times, they were still dependent on a human in the loop to answer those initial questions. The biggest continuing challenge answering customer questions that came in outside of working hours.

The Solution
“Piper” is an AI SDR agent built by Qualified specifically for inbound conversion. Piper interacts with site visitors, qualifies them, answers questions, and books meetings on its own, without an SDR. Demandbase converted their standard Qualified chatbots to Piper for all inbound traffic, except Tier 1 accounts.

Demandbase trained Piper using all their existing content (website, support site, and previous chat history) for two weeks and then turned it on for a couple of pages for two weeks to test its knowledge and ability to respond. After that, Piper went live across the entire site. They still train Piper monthly on any updates.

Demandbase is using Piper for two main use cases:

• On-site chat to answer questions and convert to meetings

• Email follow-up to prospects that engage with content on the site or attend events

Results

• 4x the amount of qualified meetings booked

• Faster, 24/7 response to inbound leads

• CMOs regained confidence in pipeline conversion without relying on junior SDR teams

• Reduced hiring and training costs; faster scale and more consistent pipeline flow

Insight
Hannah Jordan, Head of Digital Marketing, described their legacy process as a “hope-based pipeline” reliant on an SDR team. Piper eliminated that dependency and created repeatable, scalable conversion—even during off-hours.

3. Hubspot and 1Mind: AI-Powered Lead Engagement for PLG

The Challenge
Hubspot’s Product-Led Growth (PLG) motion allows prospects to do their own research on the site and sign up for a trial without ever talking to sales. Yet key conversion moments—like integrations, pricing, or compliance questions—still need human-like guidance to move forward. Potential customers who wanted answers before talking to a sales rep could get frustrated trying to find the information and drop off the site. Hubspot tried chatbots to keep visitors engaged, but still saw a drop-off before taking a meaningful action like booking a demo or upgrading their solution.

The Solution
Hubspot worked with 1Mind to create “Fiona,” an AI agent, who engages buyers in conversation, asks the right questions, empathizes with pain points, and uses visuals to explain products or integrations. She can scope projects and confidently discuss pricing and packaging, giving prospects a complete and informed experience. Once the prospect is ready, she can connect them directly to sales.

Results

The results speak for themselves with a much higher engagement than legacy chat that is converting to pipeline. Fiona’s conversational, empathetic approach is proving that customers are receptive to an AI agent that can understand their challenges and answer their questions.

Insight
Fiona mimicked a helpful product expert rather than a scripted bot, meeting PLG users in their discovery flow and guiding them toward the right solution for them, providing a better customer experience. Hubspot is already looking at other ways to expand Fiona’s role across the website in other products.

The Takeaway

Across these use cases, three themes emerged:

1. Speed wins. Inbound prospects want answers now. AI can respond in seconds 24/7—humans can’t.

2. Scale doesn’t require headcount. AI agents now handle qualification, booking, and segmentation with greater consistency than people, without the turnover and ramp time.

3. Buyers are comfortable with AI—when it’s done right. Transparent, capable AI creates trust and accelerates decision-making. Potential customers are comfortable asking questions without the pressure of a live salesperson.

If you’re not sure about investing in a solution, you can try building your own.I recently shared a deep-dive on How to Build an AI Agent for Account Selection, which can help you get started. Whether you’re running a lean Series A startup or a scaled enterprise marketing funnel, AI for inbound conversion isn’t futuristic—it's working today. The playbook is evolving fast. And those who embrace it are pulling ahead.

Subscribe now

Subscribe now

Why Account Selection Still Breaks ABM

If you’re leading ABM at an early-stage company, you’ve likely felt this tension: everyone agrees account selection matters, but it’s harder to agree on which accounts to prioritize, let alone how to find the correct contact information.

You’ve probably defined an ICP. Maybe even mapped tiers. But when it’s time to build a list, everything gets muddy: Marketing leans toward broad reach and signals. Sales wants warm intros and correct contact information. The data sources don’t agree with each other. And everyone’s still manually fixing errors in the CRM. And worse, all of this takes precious time. It’s been a challenge in every organization where I’ve led revenue, and it’s still a challenge in our portfolio companies today.

That’s why I wanted to share a new approach I’ve seen work. At Rally, a Stage 2 Portfolio Company, Shreesh Naik, an independent GTM Ops consultant, built a two part ABM system:

1. A modular, AI-powered account selection workflow that solved this challenge in a thoughtful, repeatable way. Instead of relying on static filters or pre-built segments, they inserted an OpenAI agent into their Clay workflow to act as the logic layer, qualifying accounts based on real-world context.

2. A workflow to identify and qualify the right buyer personas within the shortlisted accounts, and find and validate the correct contact information (emails and phone numbers) to make it easier to improve your connection rate. We’ll briefly touch on this here, but stay tuned, the next article in this series will go much deeper into this part of the process.

In this post, we break down exactly how he built part one so you can too. In our follow-up post, we’ll do the same for part two.

Whether you’re a RevOps lead, demand gen marketer, or founder wearing both hats, this is a roadmap for building a better ABM foundation from day one.

The Problem: Why Account Lists Are So Hard to Get Right

Even with powerful tools like ZoomInfo and Apollo, teams still struggle to build clean, confident account lists.

Here’s why:

• ICP filters are too shallow. You can filter by industry or size, but not “does this company sell a digital product to mid-market teams.”

• Title searches often return junk. For example, let’s say you’re looking for UX Researchers, the title “researchers” may mean a very different role depending on the industry.

• Enrichment happens too early. Teams enrich thousands of records before knowing which ones are actually worth it.

• There’s no feedback loop. Lists don’t get better over time because no logic gets updated. Lists get moved into a spreadsheet and often get whittled down, but the decision criteria isn’t fed back into the logic up front.

Shreesh recalls a time early in his client and Stage 2 Portfolio company’s Rally’s outbound motion: “We pulled a big list using Apollo and tried to filter it based on title and industry. But we were getting all kinds of false positives. Clinical researchers instead of user researchers. Healthcare SaaS instead of product-led B2B. It was messy.”

The Solution: Add a Logic Layer Between Search and Outreach

Instead of relying solely on filters, Rally decided that it needed an AI agent between raw list generation and enrichment. The agent applies real ICP logic, evaluates each account in context, and improves over time through feedback.

“We didn’t get rid of Apollo or ZoomInfo,” says Shreesh. “We just stopped expecting them to do something they weren’t built to do, apply judgment.”

Before getting into the weeds, let’s look at the workflow that this is based on. Shreesh calls out the importance of stepping back before you get started to map out a workflow that breaks down each part of the process into discrete and manageable steps. This helps you identify a focused set of tasks to build the AI agent around and allows you to adjust and tune each part of the process.

Here’s the high-level structure Rally followed:

1. Search Broadly – Start with a wide-net list using tools like Apollo or Sales Nav.

2. Build Initial Account List and Upload into Clay – Centralize your messy data before taking action, so you can enrich and organize it.

3. Qualify with AI – Use OpenAI to assess if accounts match your ICP. The agent reviews each row and flags what's in or out based on your criteria. (This also works for other use cases like qualifying tradeshow attendee lists.)

4. Enrichment with AI – Improve results through feedback and prompt tuning.

5. Filter the List – Keep only the accounts the AI flags as high-quality.

6. Enrich Contacts – Use Clay to pull clean contact info only for qualified accounts (this will be covered in Part 2)

7. Push to CRM – Send the enriched, qualified list to sales tools with a fit rationale attached.

Ok, let’s get building.

How to Build the Workflow

1. Start with a Broad List of Potential Accounts

First, start with a broad list of accounts. This isn't the time to go super narrow; that's what the agent will help do. Shreesh recommends using Sales Navigator, ZoomInfo, Apollo, or other tools to capture the broadest set of accounts that might fall within your ICP. Think of it as casting a wide net.

Keep filters loose: you want to capture all possible matches, even imperfect ones. Keep filters loose: you want to capture all possible matches, even imperfect ones.

Example targeting criteria: - Industry: SaaS, IT, Internet - Company size: 11–500 employees - Geo: North America

“This first list was big,” Shreesh says. “About 17,000 rows. We knew a lot of it would be junk. That was the point. We just needed a surface area to run logic against.

2. Upload That List into Clay

Bring that raw CSV into Clay. I should note that you can also build your query in Clay directly. Each row should include at least:

• Company name

• Website or domain

• Basic firmographic tags (industry, employee count, etc.)

Clay becomes your staging ground for logic, enrichment, and decision-making.

💡 Pro Tip: This may sound obvious, but Shreesh recommends keeping ChatGPT open while building your prompts. The model knows how it wants to receive input—let it guide you. You provide the strategy, and the LLM helps translate that into clear, structured instructions.

Step 3: Set Up Your AI Agent in OpenAI

Next we move into the logic layer of the process. This is where you’ll tell the agent what it should look for, who your target ICP is, and how it should approach making these decisions. You’ll use OpenAI’s Responses API, not ChatGPT,CustomGPTs, or even Clay’s Claygent

3.1. Create Your Assistant

• Go to platform.openai.com

• Navigate to the Chat tab (on the left menu)

• Create a new prompt

• Write clear step-by-step logic, including input expectations and rules (we’ll share more on the prompt below)

• Output format: Specify a JSON schema if you want structured responses

• Save the prompt and grab the Prompt ID

Design Your Prompt

As you’re doing this, don’t forget to use ChatGPT to co-create your prompt! Feed it your ICP and ask it to format a structured, step-by-step prompt. You want precision and decision-tree-like logic.

Shreesh shares some tips when doing this: “One of the best practices I’ve picked up is keeping a ChatGPT window open while building prompts. The model knows how it likes to receive input, so let it guide you. You provide the strategy and rules, and let the agent turn that into structured, step-by-step logic.

Prompt Structure Example:

• Define what the agent does (e.g., assess ICP fit)

• Specify the inputs (company LinkedIn summary, company size, HQ location, industry, etc.)

• Set rules:

  • “If none of these account keywords are present, return ‘Poor Fit’ and stop.”

  • “Only accept if the company sells a consumer-facing digital product.”

  • “Evaluate industry and GTM motion.”

Prompt example: [Given the company name and website, determine whether the company is a good fit for a mid-market B2B SaaS product selling to design and research teams. Return account_fit: good | poor | manual_review, and provide one-sentence reasoning.]

Shreesh explains: “We basically encoded our ICP criteria into plain English and handed that to the agent. It wasn’t fancy. But it worked.”

💡Why not just build the logic inside a CustomGPT instead of using OpenAI's developer tools?" Technically, even a CustomGPT is a simple agent in that it can take actions (like web search, look up files), make decisions, and pursue goals, but it can't be executed across thousands of records. It works better for a chat-like experience.

Once you’ve completed your prompt, you’ll need to generate your API Key in OpenAI. To do this:

1. Go to your OpenAI API key

2. Click Create new secret key

3. Copy this key — you’ll use it in Clay

4. Connect Clay to OpenAI and Run the Evaluation

Now we connect your Prompt to your data table in Clay.

Use Clay’s HTTP API Enrichment

In Clay:

• Click Add Enrichment

• Choose HTTP API

Set up the enrichment as follows:

• Method: POST

• URL: OpenAI Responses API endpoint (from OpenAI Docs)

• Headers: Include your OpenAI API Key

Body: Add your prompt_id and insert the input variable from your Clay column (e.g., Key Company Details)

💡 Again, ChatGPT can write this JSON body for you if you give it your Prompt structure and endpoint docs.

Retrieve and Use the Agent’s Response

Once your first HTTP API call sends the input to OpenAI, the Prompt starts processing the data. However, you won’t receive the final output immediately; instead, OpenAI will return a response ID associated with that specific run.

To complete the loop and pull in the actual evaluation results, you’ll need to:

1. Set up a second HTTP API enrichment in Clay

This call uses the response ID from the first step to fetch the final output from OpenAI.

2. Extract key fields from the response

The OpenAI agent will return structured results based on the JSON schema you defined in the prompt, such as:

• Fit rating (e.g., Good Fit, Poor Fit, or Manual Review)

• Step-by-step reasoning that explains how it reached that conclusion — such as:

  • Company HQ is within target geos confirmation

  • Company size or GTM alignment

  • Industry or product relevance

  • Size of the relevant department

3. Write the results back into Clay

These results can be saved directly into new columns in your Clay table. From there, your team can:

• Filter for top-tier accounts

• Add context to prioritize outreach

• Push the cleaned, qualified list to your CRM or sequencing tools

Filter to Keep Only the Qualified Accounts

Now, filter your Clay table to include only rows where account_fit = good.

You’ve turned a noisy 17,000-record list into a clean, ranked target list, with logic and reasoning attached to each record. And the best part is, as you update your prompt, this list will only continue to become more refined and nuanced to your business.

Push the Final List to CRM or Outreach Tools

Clay integrates with tools like Salesforce, HubSpot, and Outreach. When you’re done:

• Push the final, enriched, logic-qualified list directly to your CRM.

• Include fields like account_fit_reason so reps know why they’re being asked to work the account.

Measuring Success

One question I asked early on was whether this kind of system actually leads to better performance, better contact rates, cleaner pipeline, real results. Shreesh shared that it was really about clarity. “We knew who we were going after, why they were a fit, and what to say. That alone made the outbound sharper—and the results followed."

But there’s more under the surface. Shreesh pointed out that one of the biggest early benefits was operational: saving reps time. "This evaluation work is usually done by the sales rep," he said. "They spend 20–30% of their time just checking contact data or reviewing leads manually. That goes away."

He’s also seen that narrower, cleaner lists lead to stronger results—not just for cold outbound but for warm outreach too. "Sending the same message to 100 well-targeted people beats sending it to 100 with 30 bad fits. The resonance just improves."

As for Rally, he shared that the team is rebuilding its outbound engine from scratch—and expects to see hard performance lift data in Q3."

Why This Works

This approach doesn’t replace traditional data providers. It augments them.

• Modular: You can update the logic without starting from scratch.

• Transparent: Every account has a reason attached.

• Compound: Feedback improves the agent, not just the list.

I’ll put it simply: the power isn’t in the agent, it’s in how this system turns your best judgment into repeatable logic your whole team can use.

Subscribe now

Subscribe now

If you've spent any time on LinkedIn, then you know how noisy the AI space has become. It can feel like you're getting left behind, that everyone else has it all figured out. And, it seems like there’s an AI tool for everything: agents promising to disrupt your demand gen strategy, flashy bots that'll automate your SDRs, GPT-powered sequences that'll handle your entire outbound effort. We're all hungry for new leads and more efficient ways of working, but what's actually moving the needle? What's just noise?

I sat down with Andrew Ettinger, Technical CRO and GTM Strategist, to unpack his take on the real opportunity most teams are missing: using AI not just to create demand, but to convert it into revenue. Andrew even shared some handy step-by-step plays and prompts you can implement immediately (we love it when it gets tactical!).

When I asked Andrew about the AI hype, he pointed out that we're missing the bigger picture. "AI should be amplifying value, not just cranking up volume," he told me. "I see a lot of teams blasting inboxes and calling it innovation, but we're actually burning through our ICPs and our reps are spending time on tools that aren't really moving the needle."

So where does Andrew see real opportunity? Conversion. “We spend all this energy building 4x pipeline coverage, but then we lose three out of four deals,” he said. “That's where the real impact is.”

Instead of focusing on generating more leads, we should be using AI where it can have the biggest impact—converting the prospects we already have into actual revenue. This means using AI to better understand buyer pain points, craft content your executive buyers will actually read, and get it in front of them.

Here are the five mid-funnel plays that Andrew walked me through, with the step-by-step instructions and prompts you can use right away.

Play #1: Use AI to Extract Buyer Signals from Real Calls

Why this works: Every recorded sales call is a goldmine of real customer language, pains, blockers, metrics. That’s the foundation for revenue-generating assets.

Your Move:

1. Grab transcripts from Gong, Chorus, or Zoom.

2. Feed them into a Large Language Model (ChatGPT, Claude, Gemini).

3. Surface:

   • Top 3 business pains (with verbatim quotes)

   • Related metrics

   • Roles (Champion, EB, Blocker)

   • What’s at stake if they do nothing

4. Do this across multiple calls to find patterns.
   
   Prompt: Pain & Metric Extractor

You are a revenue analyst. From the transcript below, output:

• 3 biggest business pains mentioned (with verbatim quotes)

• Any quantitative metrics tied to those pains

• Role of each speaker (Champion, EB, User, Blocker)

• One sentence describing the stakes of inaction

→ Paste transcript here

Play #2: Build a 2-Page Whitepaper Executive Buyers Will Actually Read

Why this works: Your buyers rarely open lengthy decks and PDFs. Instead, use the insights from your calls to build a short, punchy whitepaper that speaks the EB’s language. It’s easier to consume, and it is specifically tailored to your most recent conversations.

Page 1: The Stakes

• 150-word narrative summarizing the 3 biggest pains

• One peer benchmark

• One “cost of inaction” stat

Page 2: The Value Framework

Prompt: Whitepaper Builder

Using the bullet list of pains & metrics below, draft:

• Page 1 narrative (≤150 words)

• Page 2 four-column value table

• Closing 40-word EB invite

Tone: data-driven, executive

→ Paste extracted list here

Play #3: Use the Whitepaper to Book the Executive Buyer Meeting

Here’s the part most people miss. Send the PDF with a direct ask, and text your champion before you email it. You want to get in the room, not just in the inbox.

If you don’t have their cell number? That’s a bigger problem. And if you can’t get an in-person EB meeting, your close rate just got cut in half.

Prompt: EB Email Hook

Distill the whitepaper into a 120-word email:

• Lead with projected ROI

• Cite one peer proof point

• CTA: 15-min meeting

Tone: concise, confident

Play #4: Reuse the Whitepaper for Top-of-Funnel

You just built high-value Middle of Funnel content, so why not use it to pull in new prospects?

Reuse Channels:

Also:

• Post insightful comments under ICP content.

• DM: “Do you mind if I share a whitepaper we put together on this?”

• Host a weekly 20-minute micro-webinar with Product and a customer.

Play #5: Instrument & Iterate

Run a 90-day test comparing deals that use your AI-generated whitepaper versus your current process. Track win rate, sales cycle length, EB meetings per opportunity, and revenue per rep hour. Do monthly check-ins to see what's working, then double down on the winners and kill the losers.

Track two cohorts:

1. Deals with the AI-generated whitepaper

2. Deals without

Metrics to Track:

1. Win rate

2. Sales cycle length

3. EB meetings per opp

4. Revenue per rep hour

Do a quarterly rinse and repeat.

Why MoFU (Middle of Funnel) Wins Over More Leads

Most teams think they need more leads to hit their numbers. But here's what actually happens when you focus on converting what you already have:

Use AI Where Revenue Dies

Top-of-funnel automation is useful, but not if your middle is leaky. Instead:

1. Extract buyer insights from real conversations.

2. Package into content that gets read.

3. Use it to close more—and attract more.

If you improve the win rate from 25% to 33%, that’s an extra $400K ARR per rep. Push to 50%? Add $1.2M. This isn’t just better sales,it’s exponential growth.

Let’s stop leaking revenue. Let’s close what’s already in front of us.

Subscribe now

What if you could build pipeline without ever hiring a sales development rep? That's exactly what Ryan Milligan and his team at QuotaPath accomplished. At QuotaPath, Milligan developed a method to run outbound campaigns without a single SDR, generating real pipeline without the overhead of a traditional outbound sales team.

In this article, I’m breaking down how they did it, why it works, and how early-stage founders can borrow the same playbook to drive outbound pipeline without the cost and complexity of an SDR team.

The Problem: When Hiring SDRs Doesn’t Make Sense

I talk to early-stage founders all the time, and they often ask me: “When should we actually invest in SDRs?” And honestly? The answer is rarely straightforward. And it’s possible that the annual contract value means that hiring an SDR team might never make sense. I've seen plenty of companies burn through cash trying to build outbound teams when the math simply doesn't add up.

“Our customers are upper SMB through the mid-market,” Ryan said. “Our ACV is in that range. The cost of an SDR relative to our ACV and win rate is probably not justifiable.”

They had a choice: either force-fit an expensive SDR model into their business or find a different way to build pipeline.

They went with option two.

The Solution: The SDR Team of Zero

Instead of hiring a team of SDRs, QuotaPath built a smart, lean outbound engine that’s completely automated but still human where it counts.

The foundation of their approach is something Ryan calls warm outbound. They only reach out to people who already know about QuotaPath or have interacted with the product in the past.

“We have a big body of people who know or have interacted with QuotaPath in some way,” Ryan explained. “And we just constantly tap back into that group.”

The Playbook: How They Did It

Here’s exactly how they built their human-optional outbound machine.

Step 1: Identify Your Warm Prospect Pools

The first step is zeroing in on warm leads — those who already know and have some understanding of the company. For QuotaPath, that includes:

• Past admins and reps who’ve changed jobs

• Free trial users who didn’t convert

• Closed-lost deals

• Website visitors and content downloaders

Step 2: Enrich & Update the Lists

Once they have their list, they run it through Clay to update and enrich the data. Clay is a one-to-many enrichment service that queries more than 100 different data providers simultaneously.

Ryan explained how it works: “You load a CSV with first name, last name, company name, and email. Clay finds their LinkedIn profile, checks if they still work there, and if they don’t, it’ll try to get their new email at their new company.”

Clay lets you enrich data across multiple platforms using a flexible credits model—no need to negotiate separate contracts with each vendor.

That gives them current, accurate data, so their outreach is relevant. It can also help to disqualify people who may have been relevant leads in the past. For example, if Clay identifies that a former QuotaPath admin has moved into a different, non-commission role at the same company, they won’t send an email to that person. But, if Clay shows that a former QuotaPath admin is now a sales manager at a new company, they’ll receive an outbound email.

Step 3: Personalization That Actually Matters

The team focuses on signal-based personalization: job changes, past product usage, and specific reasons for lost deals — not shallow, surface-level attributes like the university they attended.

“I care more about personalizing the problem and the solution than I do about you knowing I minored in French in college,” Ryan said.

For example, if someone was a past admin who’s now in a new role, they’ll get an email like this:

“Saw you used to use QuotaPath at Jones. Looks like you’ve taken on a new RevOps role at Marker. Is paying commissions still a challenge for you?”

How are these emails written? Some will be surprised that this is intentionally the most manual part of the process.

“I don't use AI to write any of the drafts,” said Ryan. “Honestly, I want them to feel like they're coming from a human. These are 40-word emails, max, and AI can be verbose. So I just write them myself.”

Step 4: Protect Your Domain Reputation

Deliverability is everything. The team uses Smartlead, which helps protect QuotaPath’s primary domain by spinning up alternate domains specifically for outbound.

“You don’t want the core QuotaPath domain to be marked as spam,” Ryan shared. “So you buy these separate domains, and Smartlead warms up those inboxes.”

Here is the process to follow:

1. Purchase similar, alternate email domains

Ryan explained that there are a few options for purchasing alternate domains:

• You can buy and spin up domains directly through Smartlead.

• You can buy them yourself through domain registry companies such as GoDaddy.

Ryan advises finding domains for sale that are adjacent to your primary domain. So, if your business is called Softwarexyz.com, consider buying hi.softwarexyz.com, hey.softwarexyz.com, and get.softwarexyz.com.

Domains are typically very cheap to purchase—usually around $10 per year—so this is a small investment that can yield significant returns.

2. Set up Google Workspace inboxes for the new accounts

In order for this process to work, you’ll need to connect your new adjacent domains to a Google inbox—just like you have your primary email account.

“Then, you’ll connect all that to Smartlead, which will do the warming for you,” said Ryan. “Over two or three weeks, it will automatically send emails, reply back to emails, and go back and forth to mimic a live conversation. This shows that the email domain is real.”

Once the inbox is “warmed,” you can start using it to send emails. For example, if QuotaPath wants to email 300 former QuotaPath admins, Smartlead will divide the list among several domains.

Ryan emphasized a few best practices:

1. Only sending 25-40 emails per domain per day. “You can't just send a thousand emails a day,” Ryan said. “It has to mimic the flow of an actual person.”

2. Only have two or three inboxes per domain. “You can start to raise some flags if you're trying to make like 40 different people on a Google Workspace account on one domain, all of a sudden, overnight,” he said.

Ryan shares his perspective on why domain health is critical to a successful outbound program.

Step 5: Automate and Sequence Outreach

Once lists are ready and domains are set, the outreach is fully automated through Smartlead.

They use short, simple email sequences, focused on quick yes/no decision points. And they apply spintax—slight word variations—to improve deliverability.

“If you send the same email over and over and over again, spam filters will see that you're just sending the same email to a hundred or a thousand people,” said Ryan. “But if you have thousands of permutations of the email, it increases your likelihood of landing in the inbox.”

Ryan also spoke to the importance of designing campaigns and messages based on the type of interaction the lead has had with QuotaPath, from the top of the funnel down.

For example, leads that have had a free trial of QuotaPath, but never opened an opportunity, would receive a different message than closed-lost opportunities.

“For closed-lost opportunities, we’ll reference the reasons we lost right in the email,” Ryan said. “Maybe it was a product gap, or because of budget or pricing, or whatever the reason is, if you’ve uploaded that information to Smartlead, it can be included in the email.”

While the outbound emails are automated, QuotaPath keeps them on a schedule, targeting one of each campaign per quarter. “This month, we’re outbounding to closed-lost opportunities; next month, it will be former admins; and then the following month, it will be former users,” Ryan said.

Best practice: It’s important to deconflict outreach, so one person isn’t receiving identical outbound emails over and over. This can be hard to manage manually with spreadsheets. “Fortunately, Smartlead sees the email addresses of the people you've loaded and they won't re-register somebody in the same campaign they’ve already been in,” said Ryan.

How about results?

“We don’t track open rates anymore, because you now get dinged for including the pixel that tracks open rates,” explained Ryan. “That’s OK because our reply rates on these campaigns are what I actually care about.”

Those reply rates are between 5%-15%—which exceeds the B2B outbound email reply average of 1%-5%.

“It’s all about sending a very relevant contextual message to somebody who has some awareness of what you're doing,” Ryan said.

Step 6: Add Human Touchpoints — but Only When It Matters

The only time a human gets involved is when someone replies or books a demo. That’s when the sales rep steps in.

Currently, QuotaPath centralizes all email replies into one active inbox. This makes it easy for a single person—Ryan—to monitor replies and direct inquiries to sales reps.

“I usually like to respond to the prospect first, and cc a rep, including a calendar link,” he said. “The reps don’t do a ton outbound. They mainly receive outbound demos that I’ve booked on their behalf. And then from there I'm kind of hands off.”

The Results: What Changed

Before they rolled out this system, outbound efforts at QuotaPath were inconsistent. Reps were responsible for their own outbound, but it rarely moved the needle.

“They generated very little pipeline,” Ryan said. “QuotaPath sales reps did not have a lot of experience doing outbound—but they’re very good at convincing someone who shows up for a demo to buy QuotaPath.”

Once they set up these automated workflows, everything changed:

• Outbound generated meetings jumped from 5-10% of pipeline to 20%

• Pipeline grew 15% year over year

• ARR per rep almost tripled

“Candidly, it allows me to have fewer reps that I load with more demos that are retained really well,” Ryan shared.

So, Is This the End of SDRs?

Not exactly.

Ryan’s clear that there’s still a place for human-led outbound—especially in enterprise sales, or when ACV and deal cycles justify it.

“If QuotaPath cost $100K+, I would certainly have SDRs doing more, sending gifts, flying to prospects, taking them to VIP events,” Ryan said. “But for us, this model is cost and margin supportive.”

While a team of SDRs may not be in the cards for QuotaPath, Ryan does see potential value in a single SDR. “We're definitely sold on automation and will continue to automate as much as possible, so I think we would consider adding an SDR to be an orchestrator of the tools and workflows,” he said. “But that will only come when we feel like we absolutely need another human in the loop.”

The AI Factor

So how much of this is AI magic, and how much is just good process?

A lot of it is simply good workflow design—any company can implement Clay and Smartlead. But Ryan and his team use AI in two smart ways:

1. Building the workflow itself: “AI has helped us actually build a number of these flows,” Ryan said. “Show me how I get this data here, here, and here.”
   

2. Campaign ideation: They used ChatGPT to brainstorm new campaign plays and even surfaced the idea of targeting former reps.

So, AI doesn’t write or send QuotaPath emails—but it did help to design the machine.

If you’re an early-stage founder wrestling with how to scale outbound without blowing up your budget, this model is worth a serious look.

The goal isn’t to replace humans across the board. It’s knowing when you don’t need them—and where you can use data, automation, and workflow discipline to do the heavy lifting.

You might find that the best SDR team is the one you never hired.

Final advice for founders? Ryan shares his perspective on how to get started.

Subscribe now

Subscribe now

I meet with startups every day and have looked at hundreds of GTM AI tools - it's a crowded space. At Stage 2 Capital, we've invested in a few, like Momentum, which helps enterprise teams turn conversations into insights. But for early-stage startups running lean, the journey often starts with getting creative using basic, readily available tools.

That’s what I loved about MedScout’s approach. As part of our 2023 Catalyst program, they built a solid sales playbook using the Science of Scaling framework. But they didn't want it gathering dust - they used AI to turn it into something their team could actually use day-to-day to work more effectively and efficiently.

I had Brian Aoyama, their Head of Marketing, share how they're implementing AI to do this. Their goal was simple, let AI handle the heavy lifting so their team can focus on what humans do best: building genuine connections and closing deals.

The best part? MedScout didn’t overcomplicate it.

They started with a simple use case, enhancing the call discovery process, and then kept tweaking and improving based on real results. What started as a few practical experiments has turned into a powerful AI system driving their entire GTM motion.

Ok - Let's get into how they did it.

Why MedScout Focused on Discovery First

Discovery is hands down my favorite part of the sales process. It's where great sales teams set themselves apart – playing detective to uncover what's driving their prospects' needs. At Stage 2, we've consistently seen that companies who excel at discovery close larger deals faster and build more predictable revenue engines. Many sales teams rush to showcase their product, thinking that's all it takes to win the deal. But in today's market, that approach falls flat.

Great discovery requires balancing three critical elements:

1. Depth of understanding. This means getting the full picture of your customers' situation – their business objectives, challenges, and what's driving their needs. Your reps need to play detective, asking the right questions and reading between the lines to uncover what matters most.
   

2. Consistency of execution (across calls and reps). It's not enough to have one or two reps who excel at discovery. Every conversation needs to maintain the same high standards, following proven frameworks that surface the insights your team needs to be successful.
   

3. Speed of follow-through. The insights from discovery calls are most valuable when they're fresh. Waiting days to document key learnings or send follow-up materials dramatically reduces their impact.

The challenge is that in a startup with limited resources, these three elements are often in tension. The more thorough your discovery and documentation, the slower your follow-up. This is what led MedScout to embrace AI. Brian explained, "It takes a lot of time and effort to execute our discovery process well. We needed a way to make cycles faster and more reliable without sacrificing quality."

Three Stages to AI-Enhanced GTM

Stage 1 (Crawl): Start with GTM Playbooks

Outcome: First, nail your foundation.

In my conversations with startups, I often see teams excited about the promise of AI and eager to jump straight into implementing AI tools. While the enthusiasm is understandable, this approach typically leads to frustration – overlapping tool capabilities, mounting costs, and low team adoption. Without the right foundation, AI tools can generate a lot of extra "noise" that ends up gumming up your GTM motion rather than improving it.

MedScout took a different path, one that I believe every early-stage company should follow: they invested first in building a strong GTM playbook.

A good playbook doesn't just align your team - it's the foundation for growth. At Stage 2, we've seen that companies who document their GTM process can iterate faster and spot what works. MedScout built their playbook during our Catalyst program and kept refining it, using Winning by Design's SPICED framework - though whether you use SPICED, MEDDIC, Challenger, or another approach, what matters is having a systematic way to understand your prospects and capture key information.

Example of SPICED Framework - Credit: Winning by Design

"You don't want to upend your process to fit off-the-shelf AI tools," Brian explains. "We wanted the process to serve us, not the other way around. So we used our GTM playbooks as a starting point, then layered in AI to execute those playbooks more consistently and effectively than we could before."

When you take the time to document your GTM fundamentals properly, it's much easier to start layering AI into your workflows. Let's walk through how MedScout used AI to eliminate the biggest bottlenecks in their discovery process.

Watch Brian Aoyama talk through MedScout's strategy for creating an AI playbook.

Stage 2 (Walk): Basic GTM AI Implementation

Once you’ve documented your GTM fundamentals, it’s fast and easy to start enhancing your process with AI.

We'll look at two approaches:

1. Quick Start: Simple steps you can implement right now.
   

2. Optimize & Expand: The step-by-step process that MedScout followed to level up their entire discovery process.

Quick Start: Put Your Playbooks to Work

Outcome: Create an AI-powered GTM coach for your team in under 5 minutes.

The quickest way to boost your GTM with AI is straightforward: feed your existing playbook into tools like Claude, ChatGPT, or Perplexity. Just this simple step helps these AI tools understand your company and approach, leading to much better outputs.

"When we first started, it helped to think about our AI tools as new team members – extremely productive but lacking context or direction," Brian explains. "The more context you can give them about your company, objectives, and expectations, the better your results will be."

Claude AI, trained on MedScout’s GTM playbooks, can analyze discovery calls and provide guidance and coaching based on your business and sales framework.

When you take advantage of this capability, you'll see immediate improvements in everything from prospect research to follow-up materials.

Quick tip: If you haven't completed your GTM playbooks yet, use ChatGPT to create a simple one-page document capturing critical information about your company, target customers, and value props. Some tools will even analyze your website to draft this for you. While not comprehensive, this gives you a foundation to build upon.

Optimize & Expand: Build a Custom Discovery Prompt

Outcome: Deliver thorough, personalized follow-up within minutes of every call.

After getting familiar with general-purpose AI tools, the next step is creating specific prompts that strengthen key parts of your GTM process.

MedScout focused on the discovery phase as their biggest opportunity. They built a custom prompt that combined their playbook and discovery frameworks, starting with a basic goal: create better sales call summaries than their automated software.

Want to build better AI prompts? Let AI help. "We used AI to help us build better AI," Brian explains. "We just told it what we wanted to achieve and what information we needed, and asked it to create a template. The key was being specific about our goals, what data to capture, and how we wanted it structured."

Their first implementation was intentionally simple and manual. By using actual discovery calls as their testing ground, the team could evaluate and refine their approach multiple times each day. After each discovery call, reps would:

1. Copy and paste the sales call transcript into Claude or ChatGPT
   

2. Drag and drop the latest version of the shared discovery prompt into the chat
   

3. Review the AI output and provide feedback. Where did it succeed? Where did it fall short?
   

4. Paste the final notes into the CRM

The team revised the prompt every day for 2 weeks. Eventually, the prompt grew to include sophisticated components like:

• Prospect Voice: How to handle prospect quotes: cleaning up filler words while preserving the client’s distinct voice, terminology and key phrases

• Reading Between the Lines: How to make informed inferences based on unstated or implied topics (while clearly marking them with confidence levels)

• Customer First: Improving focus on the prospect's perspective rather than the sales rep's talk track

• Rep Coaching: Scoring the rep's discovery performance and providing coaching feedback

This collaborative process showed two significant benefits beyond just improving the prompt.

1. It built the team's trust in AI-generated outputs. While early results weren't impressive, the rapid improvement cycle helped overcome initial skepticism.
   

2. It strengthened the team's discovery skills. The team became more attuned to what good discovery looks like and how to capture key insights systematically.

Here’s how MedScout structures its prompt to transform its sales playbook and team insights into a powerful AI tool. Use this template as inspiration for your own sales process.

INTRO & ROLE

Open your prompt by stating exactly what outcome you want—such as a SPICED discovery call analysis—and why it matters. This immediate clarity helps the AI target your core needs from the start. Then define the AI’s role (e.g., “You are a senior sales strategist”) so responses match the level of expertise you expect.

COMPANY & CONTEXT

Provide brief details on your product, target customers, and market positioning so the AI can interpret transcript references in the right context. Tying the analysis to your unique value proposition and ICP avoids generic output and ensures recommendations align with what truly sets you apart.

ANALYSIS FRAMEWORK

Introduce your chosen framework—whether SPICED, MEDDPICC, or another—and outline how each element should be examined. Then detail a simple step-by-step process (like extracting key facts, grouping them by SPICED element, and suggesting next steps) so the AI covers every important point without skipping or merging crucial details.

GUIDELINES

Clarify how you want the AI to handle uncertain data, flag missing information, and optionally coach the sales rep. This maintains consistency and trustworthiness: the AI can highlight where the conversation falls short, rate completeness, and provide targeted feedback for improvement.

OUTPUT FORMAT

Finally, specify how the results should look, whether it’s a bullet-point summary or a detailed SPICED breakdown with quotes. If you can share an example of the final format, the AI will know exactly how to structure its response for seamless handoff to your CRM or other team members.

Successful Prompts Aren’t Without Challenges

The custom prompt approach proved highly successful. When tested against specialized products in the market, MedScout's homegrown tool consistently outperformed more expensive alternatives. More importantly, the sales team had fully embraced it – every rep was using the prompt daily and trusted its output.

This success, however, highlighted new challenges. The prompt had grown to 35 pages after months of refinement, making updates increasingly difficult. The manual copy-and-paste workflow, while valuable for early learning, now felt cumbersome. MedScout was ready for their next evolution: automating these proven processes to create a truly AI-native GTM motion.

Stage 3 (Run): Evolving into an AI-native GTM team

Outcome: Build a GTM system that gets smarter with every customer conversation.

Quick sidebar: I love the crawl-walk-run approach. It gives you time to work out the bugs and build trust with your team (and we all know how skeptical sales teams can be). My advice? Don't jump straight to automating workflows until you've earned your team's trust and you're consistently getting high-quality output.

Ready to level up? Here’s how MedScout did it.

Automate AI Workflows

The first thing MedScout did was break its extensive prompt into discrete, specialized components. Then, using accessible tools like Zapier and Clay, they connected these components into automated workflows. And they didn't need expensive specialized software – they used the same tools as before, just connected intelligently.

"Our manual process had served its purpose, helping us refine and validate our approach," Brian explains. "But the repetitive steps were taking time away from actual customer interactions. This friction helped us identify exactly which elements to automate first."

This automation dramatically improved their discovery process. Sales reps could now focus entirely on customer conversations, knowing that detailed call summaries, SPICED framework analysis, and personalized follow-up materials would be ready in moments. This consistent, rapid follow-up accelerated their pipeline velocity.

The Next Iteration: Augmenting the Team with AI Agents

Traditional automation platforms like Zapier require pre-programmed sequential steps, following rigid "if this, then that" logic. But GTM work requires flexibility, which is why MedScout is now exploring AI agents as their next evolution.

AI agents function as digital team members who understand your GTM playbooks, think contextually, and work directly within your tech stack. Unlike fixed automation workflows, agents can determine the best approach based on context and collaborate with team members to achieve objectives.

MedScout is now using Toolflow.ai to build AI agents trained specifically on their specific GTM playbooks. These agents access the team's tech stack to handle sophisticated tasks: researching prospects, analyzing discovery calls through the SPICED framework, enriching CRM data, and creating tailored sales enablement materials.

"We're not looking for agents to replace any functions on our team," Brian explains. "Human judgment remains critical. Instead, we want to augment our team with AI assistants that help execute our playbooks more consistently and at a greater scale than previously possible."

The impact of this AI-native approach is already visible in three areas:

1. Customer Intelligence: A smarter customer data model. Each interaction adds context, helping MedScout engage more effectively throughout the customer journey.
   

2. Cross-Team Learning: Insights flow automatically between teams. Product learns about pain points, Customer Success gets rich context for handoffs, and Marketing sees what resonates in discovery calls.
   

3. Scaling: AI handles routine tasks, freeing the team to focus on work that needs human judgment.

Key Principles for GTM AI Success

What's exciting about MedScout's story is how they built sophisticated AI capabilities starting with basic tools. Through quick iterations, they got there faster than you might think - and this is huge unlock startups.

The market's creating a perfect opportunity: while big companies are stuck wrestling with legacy systems and resistance to change, smaller teams can build AI into their DNA from the start. We're seeing this play out across our Stage 2 portfolio - companies taking a thoughtful approach to AI are actually outmaneuvering much larger competitors.

Five principles that enabled MedScout's success:

1. Start where you can learn fastest. MedScout chose discovery calls because they could test improvements multiple times per day and see immediate impact on deal progression. This rapid feedback loop accelerated their learning and helped them avoid common implementation pitfalls.
   
   

2. Build team trust through visible iteration. By starting with manual processes, MedScout's team could see and shape how their AI tools evolved. This transparency was crucial – by the time they moved to automation, the team trusted the system because they had helped build it.
   

3. Keep humans at the center. AI should handle mechanical tasks like note-taking, framework analysis, and quote extraction, freeing your team to focus on what matters most: building relationships and making strategic decisions. The goal isn't to replace human judgment; it's to amplify it.
   

4. Focus on cross-functional value creation. The real power of AI in GTM comes from building systems that help your entire organization learn and improve. When MedScout automated their discovery process, they didn't just make individual reps more productive – they created a systematic way to capture insights that benefit their product, marketing, and customer success teams.
   

5. Establish a strong foundation before laying in AI. Start with clear GTM playbooks and processes, then layer in AI thoughtfully. This foundation ensures your AI tools enhance rather than complicate your GTM motion, enabling you to iterate quickly toward more sophisticated capabilities.

   
   Good Luck!
   

   Subscribe now

A huge thank you to Brian Aoyama, head of marketing at Medscout, a Stage 2 Portfolio Company, for sharing his insights and learnings with us. We’re all figuring this out together, and I can’t wait to see what comes next!

Subscribe now

There was a time when scaling your content strategy meant climbing the Google ladder: write great content, optimize for keywords, earn backlinks, and wait. That playbook still has value, but it’s no longer the whole story.

Today, buyers are skipping Google altogether. They’re asking LLMs like ChatGPT or Claude for recommendations. They’re relying on Perplexity to summarize competitive landscapes. They’re getting product suggestions directly from AI Overviews in Google search.

And here’s what might surprise you: that traffic converts better.

"We’re seeing LLM traffic convert at 30% versus 5% for traditional SEO," said Eric Hann, Head of Growth at Order.co. "It started as paranoia; we were just trying to make sure we weren’t losing traffic. But now it’s a core part of our strategy."

At Stage 2, we’ve heard similar stories across the portfolio. Founders are watching LLMs emerge as a new top-of-funnel driver, and scrambling to figure out what to do next. If that’s you, here’s a practical guide to breaking through the noise and getting started.

Before You Begin: Remember that Focus Matters

If you're a small team, you can’t afford to overhaul your SEO strategy every quarter.

The fundamentals still matter:

• Write for humans first.

• Maintain clean site structure and page speed.

• Focus on helpful, intent-driven content.

But treat LLMs like a new distribution channel. Start small: pick a few LLM prompts that matter to your buyer, see how you show up, and test from there. While some features like AI Mode are still in Search Labs, Google’s AI Overviews have already started rolling out more broadly, meaning this shift is happening fast. Founders should hold onto their SEO fundamentals, but reserve time to test, learn, and iterate as this new layer of search evolves. This shift has sparked new terms like Generative Engine Optimization (GEO), content strategies designed for how LLMs surface answers, not just how search engines rank pages.

Step 1: Understand Where You Stand Today

Start by understanding how and where you're already showing up in LLM results.

There are a myriad of new tools that have popped up to help you get there. Peec AI or Profound are a couple of good options and they’ll help you track how your site is referenced across AI systems like ChatGPT, Perplexity, and Google’s AI Overviews. Think of a tool like Peec as the “SEMrush for LLMs", it lets you monitor prompt visibility, benchmark against competitors, and zero in on what content is breaking through.

Understanding where you’re showing up in LLM rankings is important, but how can you connect that to the traffic you’re actually seeing in your analytics?

It can still be a bit murky, but there is data to show you a signal of what’s working.

In Google Analytics, LLM-referred traffic can show up inconsistently. ChatGPT sometimes tags its referrals, allowing you to filter traffic from it directly. But most AI platforms don’t yet consistently include referral headers or identifiers, meaning that traffic often appears as "direct" or "organic."

Chris Long, VP of Marketing at Go Fish Digital, explained this challenge: “ChatGPT does tag itself and is making efforts to add more tracking. Occasionally, you’ll see it in GA. But most of the time, LLM referrals show up as homepage traffic. In many instances, someone types your brand directly into their browser after seeing a recommendation in an LLM."

That’s why Eric at Order.co also recommends a low-tech approach: just ask.

"When a prospect conversation happens, ask how they found you. We’ve had multiple buyers say, ‘I asked ChatGPT,’ or ‘I saw you in Perplexity.’ Those mentions matter, and we manually log them when our attribution stack misses it - in Q3, AI will be capturing this and writing it to the CRM right from call transcripts automatically."

Pro tips:

• Use Peec or Profound to understand which prompts your brand is surfacing in, and who else is appearing beside you.

• Monitor homepage traffic spikes as a possible leading indicator of LLM-driven discovery.

• If you're recording calls, search transcripts for mentions of ChatGPT, Perplexity, or “I asked AI.”

• Remember: this is a moving target. Keep revisiting your analytics setup and test new LLM reporting features as they emerge.

This qualitative signal is often your earliest clue that LLMs are driving demand.

Step 2: Start Small. Pick 10-25 Prompts to Target.

Founders often ask: where do I even begin?

The answer is: treat this like a sprint. Pick 10–25 prompts that matter to your ICP. Use Peec to see how you rank. Then optimize, or create, content designed to win those queries.

Order.co compiles prompts they use to monitor and compare their performance against competitors within Peec.

Peec tracks and visualizes how specific prompts perform compared to those used by competitors.

Eric shared how Order.co doubled content velocity last quarter and focused that effort on commercial-intent LLM prompts like:

• “Best procurement tools for schools”

• “Finance automation platforms for mid-size teams”

Partnering with Omniscient Digital - they built structured, specific pages that solved for those intents and started tracking rank and conversion monthly.

But what made the difference wasn't just volume, it was use-case specificity.

"I don’t think broad educational content will hold up in LLM results long-term," said Eric. "We used to rank for ‘What is procurement?’ But now we focus on things like ‘What’s the best procurement platform for charter schools?’ because that’s the kind of specific question someone might actually ask an LLM."

I also reached out to another portfolio company to see how they're tackling this challenge. Jonathan Kvarford, Head of GTM Growth at Momentum, shared his systematic research process. He starts by identifying five competitor tools, then feeds all of that data into Claude—competitor analysis, customer pain points from sales calls, and questions people are asking about their space.

"The prompt asks it to identify patterns across all the research and highlight what's unique. Sometimes those unique findings are exactly what we need to double down on." Jonathan added.

This approach allows Jonathan to create a content flywheel that can be repurposed across LinkedIn posts, YouTube videos, TikToks, and multiple blog articles.

And Chris sees similar success patterns across clients: "Mid-funnel, hyper-specific content is what gets picked up. ICP-targetedages like ‘SMS for restaurants’ or ‘CRM for law firms’ convert better from LLM results than generic educational content."

So instead of rewriting your whole blog strategy, create one or two use-case pages per ICP vertical. Then monitor visibility.

Pro tips:

• Consider adding competitor comparison pages (e.g., “YourTool vs. CompetitorTool”)—these are common prompts users type into LLMs.

• Focus on creating intent pages. For example, a Fintech solution might create specific pages or blog articles dedicated to the intent behind what the searcher is looking for vs. the solution itself.

• "How to accept recurring payments for SaaS startups" > instead of "subscription billing software"

• "Setting up payment processing for nonprofit donations" > instead of "payment gateway"

• "Handling international payments for e-commerce businesses" > instead of "global payment solution"

Step 3: Know Your LLM Source Bias and Platform Strategy

Not all LLMs think alike. Where you show up depends a lot on where those platforms are pulling their information from.

A recent study by Profound, which analyzed over 30 million citations across LLMs, revealed platform-specific differences:

• ChatGPT: ~50% of citations came from Wikipedia, with Reddit also contributing ~11%.

• Google AI Overviews: Draws heavily from community-driven content like Reddit (21%), YouTube (19%), and Quora (14%).

• Perplexity: Strong Reddit dominance (47%), with YouTube a secondary source.
  
  

This means your strategy should flex by platform. If you’re optimizing for Perplexity, you’ll want a presence on Reddit. If Google Overviews is your main concern, think YouTube clips, Reddit threads, and Quora answers. For ChatGPT, ensuring you’re cited in Wikipedia or other “canonical” sources can help.

But proceed with caution.

As Chris Long notes, “Reddit is increasingly influential, but it’s not a place you can just drop in with a link and expect to rank. Communities sniff that out quickly.”

We'll get into Reddit in a future piece, but for now: listen before you speak. Look at where your brand is mentioned today, understand the sentiment, and look for authentic participation opportunities.

If you're just starting out, nominate someone internally (many times the founder or early GTM hire) to lead Reddit engagement. This is a long-tail strategy that requires patience.

Their job:

• Spend time understanding Reddit’s norms and subreddits that matter to your buyer.

• Track discussions where your category or competitors show up.

• Participate authentically. Build reputation first. Don’t sell, listen.

Once you’ve earned that presence, Reddit can be an incredible source of feedback. Ask questions. Test ideas. Bring product discussions to life. But only after you’ve built trust.

Also keep in mind: There are regional and bias-based nuances. Reddit’s user base skews U.S.-centric, which may not reflect global buyer patterns. And while LLMs rely on community content today, what data is licensed vs. free will influence what gets prioritized.

Pro tips:

• Use Profound or Peec to identify which domains LLMs are citing in your category.

• Build relationships with influential reviewers or publishers that LLMs commonly pull from (e.g., TechRadar for SaaS).

• Don’t fake your way into Reddit. Monitor threads, understand norms, and earn your presence.

As Chris puts it, “The sites that are influencing LLMs are highly dependent on the industry and audience. For example, let’s say you’re selling child car seats, it might be BabyGearLab.com. You have to understand which sites matter in your space, then find ways to show up there.”

By tailoring your content strategy to each LLM’s sourcing behavior, you’ll dramatically improve your odds of being included in the next buyer query.

Why Wikipedia Still Reigns Supreme

Across both the Profound study and a recent analysis from Ahrefs, one thing was clear: Wikipedia is the most dominant source across LLMs. While Profound analyzed citation patterns across 30M+ prompts, Ahrefs looked at domain-level citation frequency across ChatGPT, Claude, Gemini, and Perplexity. Profound’s analysis of 30M+ citations showed ~50% of ChatGPT’s references came from Wikipedia. Ahrefs similarly found Wikipedia was the top domain cited across ChatGPT, Claude, Gemini, and Perplexity.

So what should founders do about it?

Create a Wikipedia strategy. And think beyond just creating a company listing.

• Create category pages: If you're pioneering a new space, contribute to defining that category on Wikipedia.

• Founders as thought leaders: If the founder is well-published or featured, consider a personal Wikipedia page to increase surface area.

• Cross-link for authority: Wikipedia is a networked knowledge graph. Use relevant citations, credible references, and ensure you're connecting to broader ecosystem pages.

This isn't just about visibility, it's about perceived authority. Wikipedia remains one of the most trusted and broadly cited sources across LLMs, and getting your brand positioned there can drive disproportionate returns.

Pro tips:

• Contribute value-first. Wikipedia has strict editorial standards. Avoid marketing language and ensure all edits are backed by third-party citations.

• If your category doesn’t exist yet, define it. LLMs use these connections to understand context.

• Monitor changes. Because Wikipedia is public, competitors or editors might revise content—set alerts or check regularly.

Getting mentioned here isn't fast or easy, but it's one of the highest-leverage plays you can make for long-term LLM visibility.

Step 4: Shift Your Content Strategy to Align With LLM Behavior

This isn’t just a technical game. How you write, and what you write about, matters more than ever.

The old SEO model prioritized keywords, backlinks, and breadth. But LLMs seem to favor precision (how the data is structured), authority (which site it's coming from), and specificity (how it matches the intent of the searcher).

Let’s look at what’s working today:

1. Solution-first headlines beat generic how-tos

Eric emphasized this shift: “We used to rank for ‘What is procurement?’ But now we focus on prompts like ‘What’s the best procurement tool for charter schools?’ That’s what LLM users are actually asking.”

This pivot mirrors the shift toward use-case specificity: not just what your product is, but who it’s for and what it solves.

2. Editorial roadmaps need prompt research, not just keyword research
Platforms like Peec and Profound now offer prompt discovery tools. Treat those as your new editorial lens: start with 10–25 high-intent prompts that match your ICPs. Then build content clusters around those prompts. Think of this as the version of keyword research.

3. Align content to the platform’s preferred sources
Perplexity loves Reddit. Google favors community responses and YouTube. ChatGPT trusts Wikipedia. Map your editorial strategy accordingly and expect them to change.

• If you’re trying to rank in Google AI Overviews, consider integrating YouTube content or highlighting community answers.

• For Perplexity, think about building Reddit visibility or being cited on domains it favors (you’ll find these using Profound).

• ChatGPT? You may need to secure citations on Wikipedia or other structured sources.

4. Double down on mid-funnel content
Chris sees it across clients: “Mid-funnel, hyper-specific pages outperform top-of-funnel blog posts. Things like ‘CRM for dental practices’ work because they solve a buyer’s exact question.”

Your best content ideas are already in your sales calls. Jonathan at Momentum uses data from closed-won, closed-lost, and active pipeline deals to find the most relevant pain points and trends. “This gives me super rich detail that I’ve never had before,” he shared. Rather than relying on generic AI research or manually reviewing calls, he uses Momentum to surface insights automatically. It captures feedback from hundreds of conversations each month, flags recurring themes, and updates CRM fields with customer pain points, giving him a live pulse on what his ICP cares about. This process allows him to turn emerging questions or objections into new content within days of hearing it from a prospect or customer call.

Pro tips:

• Refresh old blog content with more definitive answers and use-case targeting.

• Build a monthly content sprint aligned to 3–5 prompts. Track changes in visibility via Peec.

• Collaborate with product marketing to convert feature releases into targeted, prompt-worthy landing pages.

This isn’t about abandoning good writing or SEO principles. It’s about evolving how we package and present that value. LLMs are quickly becoming the front door to your brand.

Step 5: Build for LLM Parsers, Not Just Humans

While traditional SEO rules still matter, LLMs parse and prioritize content differently than search engines. That means your content structure, metadata, and even site architecture need a fresh look.

Chris explains: “Structured content is doubly important for LLMs. Their crawlers aren’t as advanced as Google’s, especially in early-stage platforms like ChatGPT or Perplexity. They need clean, well-organized HTML to interpret and cite your content.”

Here are some tactical adjustments to consider:

1. Use Clear HTML Hierarchies

• Stick to clean heading structures (H1, H2, H3) to help LLMs understand content organization.

• Bullet points, lists, and Q&A formats are easier for parsers to extract.

Jonathan broke this down further, "If you know how to prompt well, those same principles apply to how AI consumes content. LLMs are looking for clear intent matching. They want to connect the intent of the user's prompt to the intent of your content. This means your content structure should follow prompting best practices: H1 with a clear answer, H2 with supporting answers, and a logical hierarchy that helps AI quickly extract relevant information.”

Jonathan notes that it’s more about whether the intention of this content actually answers this particular question, rather than just splattering SEO keywords throughout an article.

“Think of your content as a well-structured prompt that helps LLMs deliver fast, accurate answers to users.” Jonathan added.

2. Implement Schema Markup

• Use schema to tag articles, authors, FAQs, and products. This provides context that LLMs can ingest more reliably.

3. Eliminate JavaScript Dependencies

• LLMs like ChatGPT and Perplexity currently can’t render JavaScript-heavy content. Make sure key info is available in raw HTML.

4. Optimize Meta Titles and Descriptions

• While LLMs don’t rely exclusively on meta tags, a clear and consistent title structure improves how your content is presented and interpreted.

5. Audit Your Site Architecture

• Ensure your content is easily discoverable through internal linking.

• Create landing pages that group resources by topic, use case, or persona, this helps LLMs contextualize relevance.

Pro tips:

• Ask your web developer to install structured data templates site-wide (especially on your blog and solutions pages.)

• Test your content’s structure with LLMs. Drop a page link into ChatGPT and ask it to summarize, what it extracts will tell you what’s working.

• Consider simplifying overly designed pages. Text-over-image or interactive JS content often fails to get parsed.

• Don’t skip the basics: fast-loading pages, mobile optimization, and internal linking still matter.

Quick Note: Schema Markup for Non-Technical GTM Leaders

If you're writing blog content in HubSpot, using schema markup can help LLMs and search engines better interpret your content. While you can generate schema code manually, adding it post-by-post isn’t scalable.

Here's a more sustainable approach:

• Use a tool like TechnicalSEO’s Schema Markup Generator or ChatGPT to create a sample of the structured data you want (e.g., FAQ, article, etc.).

• Share that sample with a developer.

• Have the developer implement the schema at the template level in HubSpot. This involves editing the core blog template to dynamically insert schema based on fields like title, URL, and author.

Step 6: Measure, Iterate, and Learn as the Landscape Evolves

AI search is not static. Google’s AI Overviews remains in Search Labs, and each platform is rolling out updates rapidly. That means what works today may not work in six months, but that doesn’t mean you should sit still.

Instead, implement a rhythm of experimentation and feedback. This isn’t about rebuilding your strategy constantly, it’s about layering in LLM tactics while staying grounded in SEO best practices.

What to measure:

• LLM visibility: Use Peec or Profound to track prompt inclusion and competitor visibility.

• Attribution: Monitor homepage traffic and direct traffic lifts alongside any content or LLM prompt changes.

• Conversion rates by channel: Create dashboards that compare LLM-sourced leads (where possible) to organic and paid traffic.

• Qualitative indicators: Track mentions of ChatGPT, Perplexity, or “I asked AI” in sales calls and demo requests.

Suggested rhythms:

• Monthly: Review Peec/Profound data. Update content based on changes in prompt inclusion.

• Quarterly: Launch a few new content experiments targeting new prompts or platforms.

• Bi-annually: Revisit your technical setup to ensure it aligns with the latest LLM crawling behaviors.

Pro tips:

• When creating your board updates or GTM reviews, highlight AI search as a standalone channel to reinforce its importance. Show that you’re getting ahead of this new channel.

• We haven’t seen paid search results directly inside LLMs yet, but we expect them to arrive. In the meantime, use changes in your LLM results to inform your paid strategy. If you’re winning a specific prompt or keyword organically through LLMs, you might consider allocating spend on that same keyword or phrase in paid search.

Eric leaves us with a final reminder, “The rule of search still reigns true - no amount of optimizing will beat creating the things that help searchers get what they need. Solving for the user challenge is still the north star, it just has some new caveats.”

Above all, stay curious.