How to Build an ICP That Actually Predicts Who Buys (Not Just Who Fits)
Mid-market SaaS, 200-2,000 employees, VP of Sales or above” describes a category not a buyer. Here’s how to build an ICP from the customers who actually bought.
Ask most early-stage founders for their ICP and you’ll get a version of the same answer: “mid-market SaaS companies, 100-500 employees, VP-level buyer.” That’s not an ICP. It’s a category description broad enough to include tens of thousands of companies with wildly different buying behavior — and a category doesn’t close deals (Salesmotion).
The gap between a real ICP and a wish-list ICP shows up everywhere downstream: in every list your team sources, every lead a rep chases, and every hour spent on a company that was never going to buy. Here’s how to build the version that actually predicts.

Start backward from who already bought not who you wish would
The single biggest mistake founders make building an ICP is designing it from aspiration instead of evidence. A real ICP is built backward from your closed-won customers — the attributes and signals that actually correlate with a purchase, not the traits that sounded right in a planning meeting (Clay’s ICP guide).
Practically, that means:
- Pull your top 20-30 customers, ranked by revenue, retention, and expansion — not by logo size or how good they’d look on a case study.
- Ignore the ones who churned, haggled endlessly, or needed constant hand-holding, even if they were big accounts. Outliers teach you the wrong lessons if you build around them.
- Look for what the winners have in common — industry, size, tech stack, and crucially, the situation they were in right before they bought.
If you can’t yet do this because you’re pre-revenue or very early, use your best 10 closed deals — or the customers you’d clone if you could only sell to 100 companies total (Cleanlist).
The three-layer framework that separates a real ICP from a guess
1. Firmographics — who they are. Industry, headcount, revenue band, geography, business model. This is table stakes, and most founders stop here. It tells you who could fit, not who’s likely to buy.
2. Technographics — what they already run. The tools and platforms an account uses tell you a lot before you ever talk to them. A complementary tool in their stack is often a strong fit signal; a direct competitor’s tool can be a disqualifier worth screening out early.
3. Behavioral signals and buying triggers — why now. This is the layer most early ICPs skip entirely, and it’s the one that actually predicts timing. A funding round, a new VP hire, rapid headcount growth, or a compliance deadline is what separates “fits the profile” from “fits the profile and is in-market right now” (Martal — ICP Sales).
Firmographics tell you who could buy. Signals tell you who’s actually behaving like someone about to.
Just as important: define who you say no to
The strongest ICPs include a documented negative ICP — the accounts you deliberately don’t chase, even when they’re tempting. Deal characteristics that reliably predict churn, slow cycles, or low lifetime value are disqualifiers, not just low-priority notes (SalesHive).
This matters more at the early stage than it sounds like it should. A founder-led sales motion has limited hours. Every hour spent chasing a bad-fit account that will churn in six months is an hour not spent on the ten accounts that actually convert. A negative ICP gives you — and eventually your first sales hire — permission to say no without re-litigating it deal by deal.
Turn it into something your team can actually use
An ICP that lives in a slide deck or a founder’s head isn’t operational. The version that works gets scored:
- Weight 4-6 attributes based on how strongly each one correlates with your actual closed-won data (industry, size, tech stack fit, geography, and buying trigger are common starting points).
- Score prospects out of 100. A common tiering: 70+ is high-fit and worth full effort, 40-70 is worth monitoring, below 40 gets deprioritized or disqualified outright regardless of how attractive the logo looks (Factors.ai ICP framework).
- Feed the score into whatever you use to prioritize outreach — even a simple spreadsheet column beats an unscored gut call.
Revisit it — this isn’t a one-time exercise
The signals that predicted a purchase last quarter decay. A profile built around a 2024 funding wave is sending you after stale accounts by the time you’re reading this. The firmographic boundary — industry, size — moves slowly. The signal layer needs a refresh on a fixed cadence, ideally quarterly for an early-stage company still learning who its buyers are (Clay’s ICP guide).
Why this matters more at the early stage, not less
It’s tempting to think ICP work is a “later” problem — something to formalize once you have a sales team and a RevOps function. It’s the opposite. A founder doing outbound personally, with limited hours and no team to absorb wasted effort, has the least room for a fuzzy ICP. Every message sent to a bad-fit account is a message not sent to a good one, and at pre-seed or seed stage, that opportunity cost is the whole game.
This is core to the 0-to-1 GTM work we do with early-stage founders at GTM Playroom — building the ICP from your actual closed-won evidence, not a wish list, before you scale outbound, hire your first salesperson, or hand the motion off to anyone else.
If your outbound feels like it’s hitting a wide, low-converting list instead of a sharp one, talk to us — getting the ICP right is usually the fastest lever available before you spend on hiring or tooling.
Sources: Clay — The Complete Guide to ICP, Martal — ICP Sales: Meaning, Scoring Criteria, and How to Build One, SalesHive — Simple Guide to Identify Your ICP, Salesmotion — Ideal Customer Profile Template, Factors.ai — ICP Marketing Guide, Cleanlist — How to Build an ICP That Actually Converts
Want to Apply These Insights?
Book a 45-minute GTM diagnostic and get a clear roadmap for your next 30-60-90 days.
