From the blog

The negative ICP: who not to sell to

A negative ICP is the set of conditions under which a company can never be a good customer for you, whatever the timing. It is the mirror of your ideal customer profile, not a do-not-email list. The hard part is not listing your bad customers. It is proving that a pattern in them is actually a rule.

By Kshitij Maheshwari, co-founder · Updated August 2026 · 12 min read

Share

The test

The five gates a candidate exclusion has to pass

Before a pattern in your worst customers becomes a rule that deletes accounts, run it through five checks. Each fails differently, and each failure routes the observation somewhere.

Gate	The question	It passes when	If it fails, what you have
1. The mirror	Do any of your wins carry this condition?	No good customer has it, and you looked.	A coincidence. Keep the note, drop the rule.
2. The count	How many accounts is the pattern built on?	More than one, and flipping one does not erase it.	An anecdote. Date it, set a recheck.
3. The base rate	What share of your market carries the condition?	A share you can afford to delete.	A market decision, for the profile, not a filter.
4. The observability	Can you see it from outside, before you send?	A field, a public fact, or a visible artifact.	A qualification criterion, for the call.
5. The clock	A state that holds, or an event that expires?	It is a state.	A suppression entry, with a date on it.

Gates 3 and 5 are arithmetic, not findings. Run those two on your own list before you argue about the other three.

The boundary

A suppression list and a negative ICP are different objects

One spreadsheet does both jobs on most teams, so neither gets done well. One is about this send, the other about your market.

Definition

A negative ICP is the mirror of your ideal customer profile: the conditions that make a company impossible to win or unprofitable to keep. It describes the company, not this week's send.

Also called anti-ICP · full glossary

Suppression list

•Answers who must not receive this send
•Customers, open opportunities, live sequences, unsubscribes, bounces
•Every row has a date, and most of them expire
•Built and frozen step by step in our lead list building guide

Negative ICP

•Answers who can never be a good customer
•Conditions about the company, not this week
•No expiry, a re-test trigger instead
•It changes what you build, not what you send on Tuesday

Confusing them is how a timing problem becomes a permanent decision. A closed-lost account in its cooling window is a suppression row. A company whose economics never work is a profile row.

The mirror gate

Why "look at your churned customers" is half a method

Every published method runs three steps: list the customers who went badly, find what they share, exclude it. Every one of those steps looks at the same half of your book.

You only looked at the ones that went wrong. That gives you what your bad customers have in common, and nothing about whether your good ones have it too. Selection bias is the textbook name.

The error that follows is treating whatever they share as the cause. Draw the other two cells and it usually stops looking like one.

Draw it on paper	Stayed and paid	Churned or never bought
Condition present	Anyone here and you have no exclusion, just a pipeline description.	The only cell the standard method ever looks at.
Condition absent	Wins without the condition. The baseline the pattern has to beat.	Losses with other causes. Skipping them inflates the pattern.

An exclusion built only from your losses is half a study. You know what your bad customers have in common. You do not know whether your good ones have it too.

The count gate

How many bad customers is a pattern

At three customers, one bad experience is a third of your evidence, and the instinct is to explain it rather than count it. Counting takes a minute, and it settles the argument.

Count it, flip it, file it

1

Count it

Write down how many accounts, not how much it hurt

An exclusion feels big because one account hurt. The number that decides it is smaller: how many accounts carry the condition. Say three of your last nine customers churned and all three carried it. The pattern is three accounts wide.

2

Flip it

Ask how many would have to flip to erase it

Flip them one at a time. One left because its champion quit, so the pattern is two. A second left over price, so it is one. If flipping a single account erases it, you have a story, not a condition.

3

File it

Two homes: the tool, or a dated note

A pattern that survives the flip goes into the tool as a query. One that does not goes beside the profile as a note, dated, with a trigger to look again. It becomes a rule the day the count holds, not the day it annoys you.

Our ideal customer profile guide runs the same arithmetic on a campaign, where the unit is a reply. Here the unit is an account.

Operator note

Name the account

The check I run before an exclusion becomes a rule is boring. Name the account that would have to be different for me to be wrong. If it is the only one holding the pattern up, I have a note, not a rule.

KM

Kshitij Maheshwari

Co-founder, Real Good GTM

The observability gate

Conditions you can see before you send

A negative criterion you can only detect on a call is a qualification criterion, not a list criterion. There is no field for it, so no list build applies it.

Google Ads documentation, read August 2026, says negative keywords "won't match to close variants or other expansions": exclude "flowers" and your ad still serves on "red flower". Inclusion criteria expand on their own. Exclusions remove only what you named.

What you actually learned	Visible from outside?	So where does it go?
A formal security review ate two months	Yes. Trust centers, compliance badges, security job posts.	A list criterion. Write it as the query.
They wanted a custom integration before signing	Partly. API docs, marketplace listings, a readable stack.	A list criterion that flags, not deletes.
Nobody could move money without a committee	No.	A qualification criterion, for the first call.
The buying process ran three quarters	No, and often the row above in disguise.	A qualification criterion, plus a forecast note.
They needed hand-holding the product could not give	No, and it is not really about them.	A roadmap note. Re-test when the gap closes.

The tool decides what is expressible. Apollo's published API reference exposes two exclusion parameters for finding companies: headquarters location, and a technology in use. No industry, no keyword, no headcount.

Everything else is build the list, then subtract, which runs only when somebody remembers.

Want the profile and the exclusions written properly, by the two of us?

Book a Fit Check

The person layer

The negative persona, and why it is the dangerous half

A company-level exclusion removes a company you cannot serve. A person-level one removes a route into a company that is in fit.

The category-adjacent professional

Consultants and agencies who serve your buyers. Same title, opposite intent: they reply to resell or partner. Visible from the company type, so handle it at the account layer.

The researcher

Analysts, students and competitors doing landscape work. Best questions of the month, no purchase. The tell is a company without the problem.

The title collision

Head of Growth at a nine-person agency and at a ninety-person product company are two jobs sharing a string. Expensive, because it passes every seniority filter.

The mandate-free senior

Innovation, Transformation and Digital titles carry seniority, curiosity and no line item. Check whether the role has a budget, or only a remit.

The stale record

The title is right and the employment ended two months ago. Nothing is wrong with the persona. The row is stale, which is a refresh problem, not a fit one.

Which seats to source, and what changes per role, sits in our buyer personas guide.

Operator note

Demote, do not delete

The default for a negative persona is almost never delete. Move them down the send order and check who else at the account is sourced. Roles do not churn, people do, and deleting one can delete your only way in.

RB

Rahul Bageria

Co-founder, Real Good GTM

The routing rule

If it fails a gate, what was it actually?

A failed gate does not make the observation worthless. It tells you which document it belongs in, and all five have different homes.

The gate it failed	What you actually have	Where it goes instead
The mirror	A coincidence. Wins carry it too.	A note beside the profile, with the win that broke it.
The count	An anecdote, resting on one account.	The same note, dated, with a trigger to look again.
The base rate	A market decision, not a filter row.	The profile, plus a count of what is left.
The observability	A qualification criterion, invisible until you talk.	The first-call script and the disqualification rule.
The clock	A suppression entry, true this month only.	The suppression file, with a date and an expiry.

The base-rate row is the one founders skip, and it is arithmetic rather than judgment. What share of your list carries the condition? The line we hold is a third, past which you are choosing a smaller market, and counting what is left is what TAM enrichment does.

Five gates, five homes. The only observation wasted is the one nobody wrote down.

The honest read

What excluding actually changes, and what it does not

Four things move when you exclude a segment properly. Three are worth the work. The one everybody expects is not one of them.

Four outcomes

List size

It falls by the base rate, not by the pain

A condition deletes accounts in proportion to how common it is, not to how much one customer cost you. Run the fraction first: it decides whether this is a filter or a retreat.

Cost avoided

The cheap version of firing a customer

Removing a customer you should not have sold is slow and risky. Harvard Business Review made the case in April 2008: employees and remaining customers "may wonder whether they're next". The framework there puts divestment last, after reassess, educate, renegotiate and migrate.

Reputation

Sends nobody was ever going to pay you for

Every account outside the profile still costs a send, a mailbox and a slice of your domain's standing.

Reply rate

It does not move, and nobody publishes that

The accounts you excluded were mostly not replying, so removing them moves the denominator and little else. No credible dataset shows a reply-rate lift from excluding a segment, and the firms that publish numbers sell the service.

Excluding rarely lifts your reply rate, because the segment was not replying. What it lifts is the value of the replies you already get, and that shows up two quarters later.

Failure modes

When a negative ICP is over-fitted

Four ways an exclusion goes wrong, and one that costs enough to get its own box.

Built only from losses

The wins were never checked, so the shared attribute describes your pipeline, not a cause. The default failure, and why the mirror goes first.

Nobody can query it

The exclusion lives in a document, as a sentence. Until it is codes, keywords and domains inside the tool, the next list gets built without it.

It was really a timing entry

Closed-lost eight weeks ago says nothing about fit. Put it on the suppression file with a date, or you will still be excluding it in two years.

It deletes a third of the market

One bad quarter produces a condition true of most of your addressable list. The pain is real and the arithmetic still says you chose a smaller market.

!

Caution

The permanent exclusion with no date and no reason

A segment gets cut in month four because the product was not ready for it. The line goes into a filter, the reason never gets written down, and nobody re-tests it. Two years on, the product handles that segment easily and nobody looks at it.

Do this instead

Give every exclusion a date and a re-test trigger. No trigger, no exclusion.

How we would run it

A worked example: one exclusion through five gates

An illustrative walkthrough of the method, not a specific client result. We report real numbers only when they are real.

A seed company sells a compliance workflow tool. Three of its five churned accounts were agencies, so the founder wants agencies out.

1

The mirror: check the wins

Two of the eight customers who renewed are agencies too, and one is the second-largest account on the book. The condition sits on both sides.
2

The count: ask what would erase it

Two of the three losses ran client work out of shared inboxes. The third had its own compliance team and left for an unrelated reason. Flip it and the pattern is two.
3

The base rate: do the fraction

Agencies are a third of the current list. Cutting them is not a filter, it is a decision to work a smaller market.
4

The observability: find the real condition

The two real losses share something narrower: client work run from shared mailboxes. Visible from outside, in job posts, service pages and listed tools.
5

The clock: write it as a state

"Runs client work from shared mailboxes" holds for months, so it is a fit condition. Into the tool as a keyword and domain query, dated, with one re-test trigger: shared-inbox support ships.

The artifact

Writing it down, and where it has to live

An exclusion that lives in a document gets re-litigated every time a list is built, and it loses. It has to be a row in the tool.

Five fields

One row per exclusion

Anything shorter and the exclusion turns into folklore inside a quarter. Nobody remembers who decided it, or what would change their mind.

Google's negative keyword list is the shape to copy: "a single, global, account-level list" that applies to everything eligible, without anyone attaching it.

1 The condition. Something true of the company, not a feeling about it.
2 The query. The codes, keywords, domains or stack tells that execute it.
3 The evidence. The accounts it came from, wins included.
4 The date. So a reader later can tell a rule from a reflex.
5 The re-test trigger. What would make you look at the segment again.

Key takeaways

4 points

1 A negative ICP is about fit. A suppression list is about this send.
2 Check every exclusion against your wins before it becomes a rule.
3 If you cannot query it, it is a qualification criterion, not a list criterion.
4 Every exclusion carries a date and a re-test trigger.

FAQ

Questions founders ask

What is a negative ICP?

A negative ICP is the set of conditions under which a company can never be a good customer, whatever the timing. It mirrors your ideal customer profile at the company level. It is not a do-not-email list: it decides who goes on the list at all.

What is the difference between a negative ICP and a suppression list?

A suppression list answers who must not receive this send: customers, open opportunities, live sequences, unsubscribes, bounces. A negative ICP answers who can never be a good customer. One expires, the other holds.

How do I build a negative ICP when I only have three customers?

Slowly, and against your wins rather than only your losses. Find the pattern in the accounts that went badly, then check whether any good customer shares it. If one does, it is a coincidence. Write it down, dated, with a re-test trigger.

Is "no budget" a negative ICP criterion?

No, and it is the most common mistake in every published exclusion list. Budget is invisible from outside and it changes, so no data provider has a field for it. It belongs on the first call.

Does excluding bad-fit accounts improve reply rates?

There is no credible dataset saying so, and the numbers in circulation come from firms selling the service. What excluding buys is capacity, a result you can read, and the wrong-fit customer you did not sign. None of the three lands inside a send window.

How much of my market can I safely exclude?

Run the fraction before you commit. What share of your list carries the condition? The line we hold is a third: past that you are choosing a smaller market, and that decision belongs to the profile.

About the author

Kshitij Maheshwari

Co-founder of Real Good GTM. He has been the first business hire and Chief of Staff at seed-stage B2B startups, building outbound pipeline before any playbook existed. He wrote this one because exclusion is the half of targeting nobody audits, and the method everyone publishes for it never looks at the customers who stayed.

Connect on LinkedIn

Keep going