When was the last time you updated the ICP doc?
As a founder you’re juggling a lot. This article will help revisit that ICP motion but with the GTM Engineering impact.
In the GTM Engineering founder's guide, we walked through the five engines every GTM team needs: ICP, signal, outbound, CRM, and reporting. This piece goes deep on the first one, because it's the foundation the other four sit on.
The problem we see commonly is that the ICP lives in a deck from a planning offsite, and nobody's looked at it since. Everyone from sales, marketing, CS, and the CRM, all have a different mental model of the ideal customer. So reps chase accounts that look right but don’t close.
The fix is turning the ICP from a document into a system: a fit score that runs in your CRM, routes every account, and tightens itself each quarter against what is actually closed. Let’s see how to build it.
Before the build, one distinction that the rest of this article hangs on: Fit and timing are different things, and they belong in different places.
Fit is who the account is: industry, size, stack, economics, etc. It's stable and changes only when the company changes (rare). This is your ICP.
Timing is when the account is in motion: a funding round, a new VP, a competitor rip-out, a product launch, etc. It's volatile and changes weekly. That's the signal engine from the parent article and it should be kept as a separate system.
For example, when we use a "raised a Series B last week" as a +20 points on the fit model. You end up with a score that swings wildly as the signals go up and down.
An account that's a poor fit will start looking "hot" because a signal fired up this week. So there’s no clean way to answer the only two questions that matter:
There are three failure modes we see repeatedly across B2B:
The slide deck ICP. The team had a workshop in the early days and agreed on firmographics, pain points, and a buyer persona. It all went into a deck. But that deck is now buried in a shared drive. Sales didn't build their territory plan around it. Marketing didn't filter their ad targeting by it. The CRM has no field that corresponds to it.
The vibes-based ICP. There's no deck at all. Instead, there's an implicit sense shared loosely between the founder and a few early reps.
It's based on the first five deals that closed, and it hasn't been pressure-tested against the deals that didn't. It hasn't been written down so every new hire gets a slightly different version of it during onboarding.
The over-engineered ICP. This is rarer but it kills the velocity just as effectively. A RevOps team built a 100-attribute scoring model with weighted fields, decay logic, and custom objects. It took three months to build and nobody trusts it because the scores don't match what reps see in the field. It is technically impressive but practically ignored.
Companies with clearly defined ICPs see a 36% conversion rate advantage. Precision in targeting drives returns but precision requires infrastructure.
A working ICP system stacks five layers, each adding precision that the one below can't provide alone.
This is the floor. Industry, employee count, revenue range, geography, business model (B2B vs B2C, SaaS vs services). If a company doesn't match these, nothing else matters. Firmographics filter out the obviously wrong accounts so you're not wasting scoring logic on companies that were never going to buy.
For a 0–1 stage founder, firmographics might be all you need. Five attributes in a spreadsheet, applied manually to every prospect before you reach out. That's a functional ICP engine at that stage. Don't build the rubric yet, build the filter.
This is where most ICP frameworks get lazy. They treat tech stack as a single checkbox, "do they use HubSpot, yes or no." That misses the signal entirely. Tech stack has three distinct categories, and each tells you something different about fit.
These are time-bound events that create urgency: a new funding round, a leadership change (new VP Sales, new CRO), a recent acquisition, a hiring spike in the department your product serves, or a public statement about a strategic initiative that maps to your value prop. Triggers don't define fit, they define timing. An account that scores well on firmographics and technographics but has no trigger is a good account for next quarter. An account with a trigger is a good account for this week.
This layer connects directly to the signal engine from the parent article. The ICP system identifies who to watch. The signal engine identifies when to reach out.
Buying committees in B2B average 6–10 decision-makers. A company might be a perfect firmographic and technographic match, but if there's no identifiable champion, no one in the org who owns the problem your product solves, you're reaching into a vacuum. Organisational signals include: presence of the right roles (does the company have a RevOps lead, a VP Demand Gen, a Head of Growth?), recent hires in relevant departments, and reporting structure patterns that suggest centralised vs. decentralised purchasing.
This is the precision layer. Everything above tells you who to pursue. This layer tells you who to exclude, and it's where most teams leave the biggest gains on the table. We'll cover this in detail in its own section below.
Knowing the five layers is necessary. It's not sufficient. The next step is turning those layers into a scoring model that produces a single number per account, a number the CRM can act on.
The model we use is a 100-point rubric. It's not the only approach, but it solves the most common problem with ICP scoring: treating every attribute as equally important. They're not. Firmographic fit matters more than whether they posted a job listing last week. The rubric forces you to make that explicit.
Here's how the 100 points distribute across the five layers:
This is the largest allocation because firmographic misfit is the hardest to overcome. You can work around a missing trigger or a complex buying committee. You can't work around a company that's in the wrong industry and a tenth of your minimum viable size. Within these 30 points, weight by the attributes that most strongly predict closed-won in your data. If industry is a stronger predictor than geography, give industry 15 and geography 5. The weights should come from your pipeline data, not from a workshop.
Split across the three categories. Integrating tech, direct compatibility with your product, carries the most weight here, typically 10–15 points, because it predicts implementation speed and time-to-value. Adjacent tech, the maturity and ecosystem signal, gets 5–8 points. It's a good indicator of buying sophistication and adoption velocity, but it's a proxy, not a direct predictor. Competing tech is where scoring gets interesting. A weak competitor presence might add 5 points (they've identified the problem and are spending on it). A deeply entrenched enterprise competitor might subtract points, this is a negative signal, which we'll formalise in the Anti-ICP section.
Funding events, leadership changes, hiring spikes, public strategic announcements. These are high-signal but time-decaying, a funding round from six months ago is less relevant than one from last week. If you're automating this in Clay, set decay logic: full points within 30 days, half points at 60, zero at 90.
Presence of the right buying roles, recent hires in the relevant department, identifiable champion potential. This layer is harder to automate than the others, it often requires LinkedIn research or enrichment through tools like Apollo or Cognism, but it's a strong predictor of deal velocity once engaged.
This is a deduction layer, not an additive one. Accounts that trigger negative criteria lose points from their total. This is where competing tech reappears, a long-term enterprise contract with a direct competitor isn't just "not a positive." It's an active reason to deprioritise.
The rubric above is the full model. You don't need to build it all at once.
At 0–1, score on 5 firmographic attributes in a spreadsheet. No automation. You're pattern-matching manually across a small number of prospects, and the goal is to learn which attributes predict conversion, not to build infrastructure.
At 1–10, move the scoring into HubSpot or your CRM as custom properties. Add the technographics layer. Start using Clay to enrich records automatically. You're building the system now because you're generating enough pipeline to need it, and enough closed-won data to validate the weights.
At 10–100, add all five layers. Automate scoring through Clay workflows that enrich and score on record creation. Build routing rules based on score tiers. Connect the score to SLAs, a Tier 1 account gets a response within 4 hours; a Tier 3 gets a nurture sequence. This is the full ICP engine, running continuously.
Every B2B team has a set of accounts they shouldn't be selling to. Not because those companies are bad, but because selling to them is reliably unprofitable. They churn faster. They require more support. They push for discounts. They consume sales cycles and don't convert. Or they convert and then drag down NRR for the next twelve months.
The Anti-ICP codifies those patterns into explicit disqualification criteria.
These are accounts that should never enter the pipeline regardless of how good they look on paper. Common ones: company size below your minimum viable threshold (if your product requires a RevOps team to implement and the prospect has 12 employees, it's not going to work); industry verticals where you've historically seen 0% win rate or extreme churn; companies in regulatory environments that make your product non-viable (healthcare data restrictions, government procurement requirements you can't meet).
These don't exclude an account outright, but they lower the ICP score enough to change routing and prioritisation. This is where competing tech shows up as a negative signal. A prospect locked into a three-year contract with your main competitor isn't a hard no, they might be in a renewal window in nine months. But right now, they should not be in your Tier 1 pipeline consuming your best reps' time. Other soft disqualifiers: no budget holder identified after enrichment, company on a public hiring freeze, previous deal that went dark or was lost on price.
One benchmark from a Series C FinTech that implemented Anti-ICP scoring: pipeline volume dropped 40%, but win rates increased 22%. Total revenue went up. They were selling to fewer accounts and closing more of them because the accounts they were selling to actually fit.
Benchmark: Series C FinTech after implementing Anti-ICP scoring
↓ 40%
Pipeline volume
↑ 22%
Win rates
↑ Net
Total revenue
Fewer accounts. Higher close rates. The pipeline that disappeared was never going to close.
That's the core insight: the Anti-ICP doesn't shrink your market. It shrinks your waste. The pipe that disappears was never going to close. The reps you free up go spend time on accounts that will.
In the CRM, implement this as a flag or a score deduction, not a deletion. You want to track Anti-ICP matches for two reasons: first, to validate the criteria over time (are Anti-ICP flagged accounts actually converting at lower rates?), and second, to recycle them into future pipelines when conditions change (contract expires, leadership turns over, company scales past your minimum threshold).
Here's how to wire it up. We'll use HubSpot as the reference platform because it's the most common CRM in the 1–10 and early 10–100 segments, but the logic applies to Salesforce, Pipedrive, or any CRM with custom properties and workflow automation.
Build a custom number property in HubSpot at the company level. Call it "ICP Score." Set it to calculate automatically based on the values of the individual attribute properties below it. Don't use HubSpot's native lead scoring here, that's designed for contact-level behavioural scoring (form fills, page views). Your ICP score is account-level and attribute-based. Keep them separate.
Each scoring attribute from your rubric gets its own property. Industry (dropdown), employee count (number range), tech stack, integrating (multi-checkbox), tech stack, adjacent (multi-checkbox), tech stack, competing (dropdown with options: none, weak competitor, strong competitor entrenched), recent funding (date field), champion role identified (yes/no). Each property maps to a point allocation in your rubric.
This is where Clay connects. Set up a Clay workflow that triggers on new company creation in HubSpot. The workflow pulls firmographic data (from Clearbit, Apollo, or Clay's native enrichment), technographic data (from BuiltWith or HG Data via Clay's waterfall logic), and recent events (from Crunchbase, LinkedIn, or news monitoring). Clay writes the enriched values back to the corresponding HubSpot properties. The ICP score calculates automatically.
A critical nuance on enrichment order: filter by firmographics first, then enrich deeper layers only for accounts that pass the floor. If a company has 8 employees and you sell to 50+ seat teams, don't burn a Clay credit enriching their tech stack. The waterfall should be: firmographic check → pass/fail → if pass, enrich technographics → score → if above threshold, enrich organisational signals. This sequencing matters at scale, if you're processing thousands of accounts per month, enriching everything first and scoring second will cost three to five times more in Clay credits with no improvement in output quality.
Tier 1
75–100
High-fit accounts. Immediate sales routing.
Assigned to a named AE with a 4-hour SLA. Accounts with active triggers jump to the front of the sequence queue.
Tier 2
50–74
Moderate fit. Nurture workflow.
Automated sequences, retargeting, content offers. Reps review weekly, not daily. Goal: improve score or qualify out.
Tier 3
< 50
Low fit. Data hygiene only.
No active sales motion. Inbound leads go through qualification before reaching a rep. Stays in CRM for recycling.
Define three tiers based on the ICP score:
Workshops generate hypotheses. CRM data provides evidence. Start with closed-won analysis, then validate with team input, not the other way around.
This treats all criteria as equally important. Industry matters more than geography. Tech compatibility matters more than webinar attendance. The 100-point weighted rubric forces these priority decisions.
Without the Anti-ICP, you're scoring accounts in but never scoring them out. Your Tier 1 list always grows and never tightens, which defeats the purpose.
If you're running Clay enrichment on every new contact before checking firmographic filters, you're burning credits on accounts that will never convert. Filter first. Enrich the survivors.
PLG self-serve and enterprise sales-led are fundamentally different buying experiences. A company perfect for self-serve is probably a terrible enterprise prospect. Build separate scorecards.
A scored account with no routing rule is a scored account nobody acts on. The score only matters if it triggers a workflow: assignment, SLA, sequence priority, alert.
RevOps updates the model quarterly but doesn't distribute changes to sales and marketing. The one-page summary is the mechanism that keeps the scoring model credible with the people who need to trust it.
Helps B2B Founders close the gap between present day MarTech and the GTM operations that haven't caught up yet
.png)
GTM Partner for Lean B2B SaaS Teams
.svg)