Everyone in B2B now agrees you should build your ICP from data rather than opinion. Almost nobody shows the working. The advice stops at "look at your closed-won deals", as if the weights fall out of the spreadsheet on their own.
They do not. Deriving weights from wins is a method, with steps, and with a couple of traps that quietly produce a wrong answer if you skip them. This post is that method, start to finish: what data you need, how a field earns its weight, what to do when your sample is small, and how to turn the result into something a rep can act on. The case for why weights matter deserves its own piece. This is the how.
What is a weighted ICP?
A weighted ICP is an ideal customer profile where every attribute carries a number that reflects how strongly it predicts a win. Instead of "industry: SaaS, size: mid-market", it says something closer to "industry alignment carries three times the weight of company size in our win pattern, and region barely registers".
The difference sounds academic. It is not. An unweighted ICP treats every attribute as equally important, which means a deal that matches on three trivial attributes looks identical to a deal that matches on the one attribute that actually decides outcomes. A weighted ICP separates them. That is what turns a profile from a description into a decision tool.
Why build it from closed-won data?
Because your closed-won deals are the only unarguable record of who actually buys from you. Everything else is opinion.
The traditional way to build an ICP is a workshop. The VP of Sales says mid-market. The VP of Marketing says SaaS. The CEO says enterprise. Everyone compromises on something that helps nobody, writes it in a slide, and never opens the slide again. The result is consensus, not evidence.
Your CRM has the evidence. Every won deal carries a constellation of signals: industry, company size, deal value, sales cycle length, lead source, region. Individually they are just fields. Together, across a body of wins, they reveal a pattern. The pattern is your real ICP. It has been sitting in your database the whole time, and it frequently disagrees with the slide.
One honest caveat before you start: closed-won data tells you who buys. It does not tell you who stays. A weighted ICP built from wins is a targeting instrument, not a retention model, and it is worth keeping those two questions separate in your head from the beginning.
What data do you need?
You need your closed-won deals from the last 12 to 24 months, with the company-level fields populated, and ideally your closed-lost deals from the same period for contrast.
Twelve months is usually the floor. Shorter than that and you are reading noise. Longer than 24 months and you risk building a profile of a market that no longer exists, because your product, pricing and positioning have all moved since.
The fields that tend to carry signal:
Get pipeline intelligence insights weekly
Join sales leaders learning to turn CRM data into pipeline confidence. No spam. Unsubscribe anytime.
- Industry or vertical
- Company size (employees or revenue band)
- Deal value
- Sales cycle length
- Lead source
- Region
Do not agonise over having all of them perfectly populated. Gaps are information too, and the method below handles them honestly.
How do you derive the weights?
The weight of a field comes from how differently your wins are distributed compared to everything else you pursue. Concentration is signal. Uniformity is noise.
Work through it field by field:
Step one: profile your wins per field. Take industry. Bucket your closed-won deals and look at the spread. If 60 percent of your wins sit in two verticals, that field is telling you something. If your wins are spread evenly across ten verticals, industry is not part of your win pattern, however strongly the room feels about it.
Step two: contrast against your losses. This is the step almost everyone skips, and it is the one that separates a real ICP from a description of your outbound list. If 60 percent of your wins are in two verticals but 60 percent of your losses are too, that concentration is not a win signal. It just means that is where you spend your time. A field earns weight when wins and losses diverge on it.
Step three: assign weight in proportion to divergence. Fields where wins concentrate and losses do not get heavy weights. Fields where the two distributions look alike get light weights. Fields with no discernible pattern get zero, and writing a zero down is not a failure. Knowing that region does not matter is as commercially useful as knowing that industry does, because it tells you where to stop filtering.
Step four: resist rounding to the story you expected. The weights will surprise you. Deal value might matter more than industry. Lead source might dwarf company size. When the data contradicts the slide, the slide is wrong. That discomfort is the entire value of the exercise.
How do you handle small samples?
Refuse to overclaim. If you have 15 wins, you can read direction from the data but you cannot read precision, and any method that hands you confident weights from 15 deals is making them up.
Two rules keep you honest:
Set a minimum per field. If only six of your wins have the industry field populated, that field does not get a learned weight this quarter. Mark it as unknown and move on. An unknown labelled as unknown is useful. An unknown dressed up as a weight is a landmine.
Blend towards neutral when thin. With a small corpus, pull your weights back towards even rather than trusting extreme values. As the win count grows, let the data take over. The alternative, treating 15 deals with the same confidence as 150, produces an ICP that is precisely wrong instead of approximately right.
This is also the answer to a question you should ask of any tool or consultant offering you an ICP: what happens when the data is insufficient? If the answer is anything other than "it says so", walk away.
How do you turn it into a scoring system?
Score every open deal against the weighted profile, then band the scores so a rep can act on them at a glance.
The mechanics: for each open deal, compare its attributes to the win pattern field by field, multiply each match by that field's weight, and sum. The output is a fit score. On its own a number like 71 means little, so band the range into tiers: the top band is where your wins have historically lived, the bottom band is where your time goes to die.
Then use it for the decision it was built for. When a rep has 40 open deals and time for ten, the bands answer the question the adjective ICP never could: which ten.
Two disciplines matter here. First, the score should describe, not instruct. "This deal resembles your wins" is a fact. "Drop this deal" is a judgement call that belongs to a human who can see what the data cannot. Second, when a deal is missing the fields that carry the heaviest weights, the score should say so rather than quietly pretending to full confidence.
How often should you rebuild it?
Every time a meaningful batch of deals closes, and at minimum quarterly. An ICP is a moving target because your market, product and pricing are all moving.
This is the quiet failure of the workshop ICP: even when it starts accurate, it decays. The version most teams are using describes the company they were two years ago. A weighted ICP built from closed-won data has a rebuild button by construction. New wins arrive, the weights shift, the profile stays current. Watching the weights move over time is its own intelligence: if industry is fading and deal size is rising, your market is telling you something before your revenue does.
Where do you start?
Start with a rough manual pass, because even an imprecise version will show you things the workshop never did. We have written a 30-minute spreadsheet walkthrough for exactly that: how to analyse your closed-won deals. Do that first if you have never looked at your win data properly.
What a manual pass will not do is stay current, handle the small-sample maths honestly, or score your live pipeline every day as deals move. That is the part we built Telepath to do: it learns the weighted ICP from your HubSpot closed-won data and scores your open pipeline against what actually closes. Learned from your wins, not configured in a workshop.
See what your own data says. Three minutes, free: telepath.pro