Most ICPs get written once and never checked. This page walks through the one test from the episode, a part at a time: play the part, then work the numbers yourself. Every example here is made up.
Fit score, fit tier and lift, decoded
Once you’ve written down your ICP, every new lead can be scored against it. That’s a fit score. Then you group leads by how well they fit: strong, good, weaker. Those groups are fit tiers.
Lift is the simple bit. It’s how much better your strong-fit leads do than the rest, the good and weaker ones combined.
Lift = strong-fit rate ÷ the rest’s rateEach rate is the share of that group that converted.
Why an unchecked ICP is decoration
The test is simple: do the leads that fit your ICP best actually do better? It needs no new tools, just leads you already have. If they don’t do better, your ICP isn’t steering anything yet. It’s decoration.
Strong fit vs the rest: the 2× line
Two groups, not three:
Strong fit
The strong tier on its own
vs
The rest
Good and weaker, combined
The line we use is twice as often: strong-fit leads should convert at least twice as often as the rest. It’s a rule of thumb, so pick yours. But pick it before you look, or any result can be called a pass.
Here’s the made-up example from the episode. Try each step before you open the answer.
Your turn · made-up example
| Group | Leads | Replied |
|---|---|---|
| Strong fit | 40 | 12 |
| The rest | 38 | 6 |
1. Strong fit: 12 of 40 leads replied. What share is that?
Show the answer
Replies divided by leads in that group.
12 ÷ 40 = 30%
2. The rest: 6 of 38 replied. What share?
Show the answer
Same sum, the other group.
6 ÷ 38 ≈ 16%
3. What’s the lift?
Show the answer
Strong fit’s share divided by the rest’s share.
30% ÷ 16% ≈ 1.9×
4. Your line is 2×, set before you looked. Does it pass?
Show the answer
1.9 is under 2. The line was set first, so it doesn’t get rounded up after the fact.
Close, but not yet
Why shares, not counts: 12 replies against 6 looks like exactly double. But the groups aren’t the same size, 40 leads against 38, so the shares are what you compare. Counted properly it’s about 1.9×, just under the line.
When the rest never convert
This is the case that tripped us up. When the rest had never converted at all, our numbers said “not enough data”. That’s backwards.
Strong fit 12 of 40. The rest 0 of 38.Made up. No ratio, and none needed.
You can’t divide by zero, so there’s no lift to write down. But zero from the rest is about the best news an ICP can get. Say it in counts, as above, and call it what it is.
Enough to judge
Before the ratio means anything, I want about 20 leads in each group, and at least 5 strong-fit conversions. Below that, the answer is “not enough yet”. Both are rules of thumb.
The second floor matters as much as the first. With only a couple of strong-fit conversions, one lead either way can move the lift past your line or back under it.
What the result means
Three ways to read it:
1
Lift well above your line
It’s working.
Tighten what you import to it.
2
Lift around 1×
It isn’t separating good leads from bad.
Revisit the disqualifiers and triggers from episode 2.
3
The rest beat strong fit
The ICP may be too narrow, or aimed at the wrong segment.
Decoration vs a working ICP
Bad looks like an ICP written once, with every lead treated the same. Good looks like a pass line set before you look, lift checked on your own leads, and every revision dated, old leads rescored, and checked again in about six weeks.
Bad looks likeThe ICP is written once and filed away.
Good works likeThe ICP is checked against real leads on a schedule.
Why it mattersAn ICP nobody checks can’t steer anything. It’s decoration.
Bad looks likeEvery lead is treated the same, whatever its fit.
Good works likeLeads are scored and grouped into fit tiers: strong, good, weaker.
Why it mattersWithout groups there’s nothing to compare, so you can’t see lift.
Bad looks like“Is it working?” is answered after looking at the numbers.
Good works likeThe pass line is written down before you look.
Why it mattersSetting the line afterwards lets you call any result a pass.
Bad looks likeJudging on a handful of leads.
Good works likeAbout 20 leads per group and 5 strong-fit conversions, or “not enough yet”.
Why it mattersTiny samples swing wildly. One deal can flip the answer.
Bad looks likeChanging the ICP quietly, mid-stream.
Good works likeEach revision dated, with what you expect to move, old leads rescored, checked again in about six weeks.
Why it mattersOtherwise you can’t tell whether the change helped, and old and new scores aren’t comparable.
Rescoring is the step that gets skipped. Leads scored on the old ICP aren’t comparable until they’re rescored on the new one. Otherwise you’re comparing two different tests.
For scale, one outside number. Ebsta and Pavilion’s 2024 B2B Sales Benchmarks: H1 Update found win rates 3.1 times higher when opportunities matched the ICP, yet only 18% of 2024 pipeline did. The sample was 4.7 million opportunities. It’s a vendor’s dataset of its own customers, and a correlation, not cause. It also measures win rate on opportunities, which is a different thing from the 2× pass line on lead conversion here. So I read it as a sign that fit tends to matter, not as the line to hold your own ICP to.
Revising is a bet too
Changing who you sell to is a bet, and it can be a big one. HubSpot’s Brian Halligan, talking to AdExchanger, described their customer mix moving from about 80% small-business owners to about 80% marketers.
So treat each revision like one: date it, write down what you expect to move, rescore the old leads, and check lift again in about six weeks.