Your ICP Filter Is Hiding Your Hottest Accounts

Somewhere in your market, an account is on your pricing page for the fourth time this week, and your system has decided nobody should be told. It fails one line of your ICP definition, so it is out of your addressable market, out of your ad audiences, and out of every alert. The intent is real. The rule that hides it was written a year ago and nobody has read it since.

This is not an argument against an ICP. It is an argument against treating one as a wall instead of a hypothesis.

How a good rule goes bad #

Cole D’Ambra, who runs growth at Plain, put a live example on screen in front of a room of GTM engineers. A large security company had an intent score of 217, 17 sessions on the site and over an hour of time on it. Its card read cold.

Plain’s intent platform, sensibly, does not raise an account to act-now unless it is inside the ICP. One of the ICP rules, written in 2025, required at least 30% of a company’s staff to be engineers. This company was at 24.5%, because at enterprise size even a very technical company has a lot of people who are not engineers. The rule was right for the companies it was written about, and wrong for the ones it had never seen.

Cole’s summary of the old state: if a company did not fit a static ICP that changed every few months (at some companies he has worked at, every few quarters), its high-intent engagement was suppressed. Not in target audiences, not in ad audiences, and invisible to the team.

Why nobody notices #

Three things make this failure silent:

• Suppression produces no event. An account that is scored and alerted leaves a trail. An account filtered out before scoring leaves nothing to look at.
• Cliffs hide near misses. A threshold at 30% treats 29.9% and 5% the same. The accounts closest to the line, often the most interesting ones, get no special attention.
• The rule outlives its reason. Whoever wrote it knew why 30% made sense for the market they were looking at. That reasoning was never written next to the number.

The query to run this week #

You do not need an agent to start. You need one question asked on a schedule: which accounts are showing strong intent but are excluded by fit?

SELECT a.domain, a.fit_tier, a.excluded_by, i.intent_score, i.sessions_30d
FROM accounts a
JOIN intent i ON i.account_id = a.id
WHERE a.fit_tier = 'out'
  AND i.intent_score >= 150  -- your "act now" bar
ORDER BY i.intent_score DESC
LIMIT 50

The column that makes this useful is excluded_by: the name of the rule that put the account out. If your scoring does not record it, make that the first change. Once it does, group the results by rule. One rule responsible for most of the high-intent exclusions is the rule to revisit.

Look at the top of the list by hand the first few times. Some accounts will be genuinely out of market: a student, a competitor, a consultancy doing research. Write those down as known exceptions. The rest are the pipeline your filter has been hiding.

Change the rule, not the philosophy #

When a rule is suppressing real buyers, there are better fixes than deleting it:

• Loosen the threshold, and price it first. Plain lowered the engineering density bar from 30% to 20% after previewing the change against all 55,000 accounts: about 200 moved into tier 3, 86 up to tier 2, and 23 up to tier 1.
• Make the cliff a slope. Replace a hard cutoff with a score that falls off near the line, so 24.5% counts for less than 35% but more than 5%.
• Let strong intent open a review, not a tier. Intent above a high bar sends an out-of-ICP account to a human to look at, without promoting it automatically.
• Write the reason next to the number. A comment saying why the threshold is 20%, and what data it came from, is what lets the next person change it safely.

Then automate the looking #

Once the query is useful, the next step is to have it run itself. At Plain, a weekly job asks whether high-value accounts are being suppressed, checks whether the responsible rule still makes sense, and opens a pull request with the fix and its impact across every account. A human still decides. We walk through how to build it in letting an agent propose changes to your scoring model.

The ICP stays. It just stops being a wall nobody inspects.

FAQ #