Dear GTM Strategist!
I have had a few calls recently that went almost exactly the same way.
Founder: “We’re ready to scale. We just need to figure out how to do it using as much AI as possible!” (People are really sold on AI-Native GTM by now - yaas - the idea that something else will do the work and we don’t have to pay humans to do it is very appealing.)
Me: “Great. Who are we selling to?”
Founder: “Well, honestly, our product can help pretty much anyone who...”
And there it is. The sentence that kills more launches than I’d care to admit.
As a founder myself - I get it. I really do. When you invest your time and money into building something that you find genuinely useful, narrowing feels like leaving money on the table.
But it is not. Most of us don’t have the resources or product state that would equally serve everybody we could potentially help just yet. We have to narrow down our market and make difficult pricing, positioning, and ICP choices that will speak directly to it.
People usually get it when I show them this illustration 🍪
By the time most of us get involved in go-to-market, the product is already largely a given input. We can influence it at the margins, but it exists. Which means the customer decision is usually the first serious decision that is actually ours to make.
There are two ways teams get it wrong. Either they take it way too lightly, defining an ICP in a 60-minute persona brainstorm on a Tuesday afternoon with a branding consultant who knows absolutely nothing about your industry and calling it strategy. Or they take it way too seriously, disappearing into six months of research and emerging with a beautiful deck and zero customers.
There are only two roads I have seen work reliably:
Launch broadly, then reverse-engineer. Go wide, watch which segment retains and monetizes best, then double down. This works if you are PLG, can fund a launch with real reach, and have the capacity to service those early customers decently well. Plenty of great companies got here this way.
Do the digging first. Narrow the market before you start pumping out marketing and sales, using beachhead logic to find the segment most likely to convert in the next three to six months, with reasonable willingness to pay, healthy growth potential, and enough delight that they will actually refer you.
This article is about road two, because it is the one most teams need.
Here is what we’ll cover:
Market segmentation - segment by job to be done, not by employee count
Segment scoring - narrowing for winnability, not for size
Research methods - why people doing beats people saying, every time
The ECP prototype - your first answer is an ECP, not an ICP
The GTM roadmap - when to expand
Let’s find the one 🗺️
Step 1: Market Segmentation (Who Could This Help?)
I have to poke fun at personas one more time, because most I see out there are hilarious.
The majority of ICP definitions I see in the wild are essentially industry plus headcount on a slide. “Mid-market SaaS companies, 200 to 1,000 employees, North America.” That could be roughly 3,000 companies - can you access them at the same time with the same messaging and serve them equally well? If 🫨- this definition is probably way too broad for you to operate with (and it is pretty narrow compared to most ICP slops I have the sadness to review).
The most useful basis for segmentation is almost always the job to be done, or the problem being solved. Something meaningfully more specific than smashing a random employee/revenue range onto your ICP deck.
Now, I am not anti-firmographics. Company size could correlate with willingness to pay, and in B2B you will absolutely end up with firmographic parameters eventually, because they help with targetability. But they are an output of this step, not the starting point.
The starting point is a much simpler question: who can this solution genuinely help, and what are they trying to achieve?
You should come up with between 10 and 20 segments fairly quickly (AI can massively help here - here are a few useful skills you can use right away). That is perfectly fine at this stage.
The only real quality bar is this: is each segment defined specifically enough that you could actually go and target it later?
Not “everyone who would like to reduce AI costs” but “Europe-based companies with a high volume of data to process with AI, operating with clients in regulated industries, with min $10M (Series A) in funding at the edge of scaling stage.” Specificity is your best friend here to unlock the next four steps.
That is then a workable answer at this stage. It is also completely useless on its own, which is exactly why step two exists.
Step 2: Segment Scoring (Where Can We Win?)
For most teams, targeting and properly exploring 15 segments at once is fantasy.
The real cost of trying isn’t just wasted effort and time. It is that you never reach critical mass in any single segment, which means you never gather enough evidence to say with confidence that you validated anything. You end up with 15 half-answers and still no decision.
So segment scoring answers one question:
Segmentation tells you who you could help. Scoring tells you who you can actually win in the next 3-6 months. Those are two very different lists.
This is where you bring in the specifics of your product and your sales process. The criteria I score against:
Realistic conversion window. Is this customer plausibly going to convert in the next three to six months? Not “would they benefit,” but “would they buy, soon.”
Sufficient willingness to pay. Can they afford you at a price that builds a business, not a hobby?
Strength of the pain. How badly does this segment feel the problem? Are they motivated to solve it now? A problem they have comfortably worked around for years is not a problem you can sell into.
Healthy market. Is it growing? Is there room?
Proximity and access. Do I know these people? Can I reach them? Do I speak their language? This one gets skipped constantly and it is often the deciding factor. A slightly less perfect segment you have genuine access to beats a perfect segment you have no way in.
The Market Problem Map, which I built with Simon Belak, is better when you are under 10 and want to investigate each space with real depth.
I still believe the best way to do this is human-to-human. But you can use AI to pressure-test your scoring afterward. Running your results past AI for an objectivity check is a genuinely useful second pass, and it tends to make teams more comfortable committing to the answer.
The output of this step is your beachhead segment candidate to validate.
Step 3: Research Methods (Now Prove It)
One core governing principle here:
People do > People say.
Every single time.
The closer you can get to someone doing something real - signing up, paying, integrating, replying to an outbound message - the more confident you can be. And how far you can push along that spectrum depends partly on how much product you have to show.
Which brings me to synthetic personas and AI-only research. And maybe I am being a little old-school here, but talking to actual humans still hits differently than reading an AI-generated document or talking to a bot. When you sit with someone and feel their frustration, you build empathy. You get ideas you would never have thought to prompt for. They refer you to someone else. The AI version is competent and somehow empty.
I’m not saying AI/syntetic personas are useless. I’m saying: please don’t skip at least 10 real conversations (interviews, sales discovery calls) before making any final calls.
Here is a live example of what good looks like. I was in the pre-beta early access program for Grok Bot that launched this week, and watching that product team work was genuinely instructive. We signed an NDA, had a shared Slack channel, and I sent feedback as I used it. What made it work is that I felt like I was in it with them.
Compare that to receiving a cold LinkedIn message with a link and “tell me what you think?” If I like the person, I’ll send something back. Mostly I’ll forget. I can’t prioritize it.
For Grok Bot, it resulted in a stellar launch, with high-level opinion leaders like Lenny Rachitsky and Ruben Hassid posting about it for free, just because they felt involved being in a pre-beta program. One caveat: the product needs to be sincerely impressive (and this one is).
My favorite method once interviews are done is a small marketing or sales test that brings some real data in. Teams that I work with usually use these channels/tactics to run it:
Founder branding - building in public on LinkedIn or X
Advertising budgets - Google Ads usually works best, but go where the audience is (sometimes that means working with a niche publication, newsletter, or an influencer)
Direct outreach - preferably to an email list of relevant event attendees - that works great
Community launches - again - go where the audience is … and don’t be cringe
More info on that with many, many examples in this article if you need more context.
Ideally, you will get some traction to reverse-engineer really fast; otherwise, go back to the drawing board and try more validation or segments - it is not one shot, one opportunity. In practice, it takes 5-7 iterations to get it right.
Step 4: Build the ECP Prototype (Meant to Be Broken)
I hesitate to call it a persona, but I know you will 🙂
I call it ECP - Early Customer Profile. And it looks something like this.
One simple diagram, not a 40-page research report nobody opens.
It captures the main insights from previous stages and serves as a team alignment tool.
For now, in early stages, it’s the single source of truth, but it’s built to be broken. In my book, I said this holds for 3 to 18 months. I’m revising that down. In current AI cycles, I’d say three to six months, realistically. Markets are reshaping faster than our documents are.
Ideally, the ECP/ICP doc lives and evolves somewhere in your GTM brain and gets revisited automatically. At minimum, set a calendar reminder and ask three questions: what changed in the market, what changed in the product, and what changed in our traction? Is my ECP/ICP doc still relevant?
If you are further along the way and already have a real volume of customers, opportunities, and closed-lost data, Eddie Reynolds wrote the most in-depth piece on turning that reminder into an actual system: The AI-Driven Continuously Self-Improving ICP. His argument is that most mature ICPs are broad, shallow, and sitting in a doc nobody opens, with no feedback loop as deals get won and lost. Sound familiar? Same problem, later stage.
The difference is what you have to work with. At the earlier stage (that we focus on in this article), you have 10 sales call transcripts and a hypothesis. At the stage Eddie talks about, you have five full years of CRM data and can ask AI which signals actually separate your best customers from the rest.
Step 5: The GTM Roadmap (Earn the Right to Expand)
Your ECP is not your forever customer. These are the early customers who will give you valid insight and real money without demanding massive social proof first or dragging you through nine-month procurement cycles.
If someone still shouts at you that “we are leaving money on the table and we can help everybody,” show them this image and tell them Cursor, Clay, Harvey, Lovable, Uber, Slack, and Figma bake in the same logic to their early go-to-market roadmaps.
So come to peace with the fact that most companies don’t have resources to scatter across non-related segments. You choose one - the beachhead segment and then work your way up the market from here. Framing it this way makes narrowing feel like a smart strategic choice rather than a sacrifice, which is exactly what it is for most folks.
Now the danger zone 💀 Once you start getting traction, new opportunities will find you - you’ll get new ideas on sales calls, investors will pull you sideways, competitors will trigger your FOMO, inbound leads will try to lure you into 999 other directions, and the highest paying customer will demand one more feature - if not, they are gone.
When to move from the beachhead segment to other attractive segments or lean in to insights to change the direction?
In theory, you would commit to beachhead for min. three to six months or till you win a certain % of market share there - the answer is usually from 60-80% - this is how narrowly the beachhead should be defined.
If you have more resources to validate multiple segments independently at the same time and go all in on multiple fronts, you can absolutely do it - but do it quietly, in relative isolation.
The cost people underestimate is positioning drift. Public communication gets confusing fast when you’re serving several masters. Your marketing and sales materials need to reflect a vertical experience the buyer actually identifies with, and you cannot do that convincingly for three segments simultaneously.
One more thing while we’re here: be extremely careful not to take product requests à la carte from sales calls. At this stage, customers are still largely buying your vision of how this should work, so every “but it would be even better if it had/did this” should be taken with a grain of salt if it is really aligned with your product and go-to-market vision. If you lean in too much, you may lose direction and sink your resourcing into serving “a market of one”
So Where Does That Leave You?
The reason I wrote this is simple: to help you answer the question “how do I define my ICP?” And while I am triggered to write “You don’t - you find it in interactions,” I know that you deserve better answers.
It’s more work than a workshop. And it’s work that mostly can’t be delegated. Please don’t outsource this decision to someone with no knowledge of your industry, and don’t rely on AI slop that doesn’t survive contact with the market.
And even when you follow this process, your work in the ICP arena is never done. Things change constantly - a new launch from a competitor, new technology, or market saturation will shake your product-market-fit. These frameworks come in super handy whenever you’ll be forced to defend it.
There are 135+ more helpful frameworks, workshops, templates, examples and AI prompts in my GTM checklist that has helped 9,500 business owners plan & execute their GTM strategy. Get it here
If you’re interested in my other products:
📘 Get my best-selling GTM Strategist book that helped 9,500+ companies to go to market with confidence - frameworks and online course included.
📈 My latest course: AI-Powered LinkedIn Growth System teaches the exact system I use to generate 7M+ impressions a year and 70% of my B2B pipeline.
🏅 Are you in charge of GTM and responsible for leading others? Grab the GTM Masterclass (6 hours of training, end-to-end GTM explained on examples, guided workshops) to get your team up and running in no time.
🤝 Want to work together? ⏩ Check out the options and let me know how we can join forces.
Thanks for reading, and let’s go to market!
Maja
The criterion people skip is access to the buyer, and it is the one that decides whether your reps close or just collect very pleasant discovery calls. "Version one, meant to be broken" is the right posture too, because ICPs go stale faster than most GTM leaders like to admit.
I think the single best piece of advice I've heard is actually speaking to real humans. The nuances that a short conversation can surface are totally impossible to be replicated by AI.
When someone's eyes light up using the product, their body language shifts, or when you hear them getting excited or nervous about something you just showed them always helps us with our decision making.