ICP analysis software turns your closed-won deals into a weighted, scoreable definition of who you actually sell to, then scores open opportunities against it. The good ones derive that definition from your outcomes. The rest ask you to write it down, then score against your own guess.
That distinction is the whole category. It's also the one thing most buyers never check before signing.
What ICP analysis software actually does
Every B2B company has an ideal customer profile somewhere. Usually it's a slide. Someone wrote it in a workshop, listed four or five traits, and nobody has revisited it since.
The problem isn't that the slide is wrong. It's that it isn't a scoring system. "Mid-market SaaS, 200 to 500 employees" doesn't help a rep decide which of their forty open deals to work on a Tuesday morning. It has no weights, so it can't rank anything.
ICP analysis software does three jobs:
1. Derives or captures the profile. What do your best customers actually have in common? Industry, size, region, tech stack, deal shape, how long the cycle ran.
2. Weights the traits. Not all traits matter equally. If industry alignment explains far more of your wins than company size does, the model needs to say so, with a number.
3. Scores against it. Every open opportunity or target account gets a score showing how closely it resembles the pattern. In the tools worth paying for, it also shows which signals fired and which were missing.
Job three is the one people buy. Jobs one and two are where the tools differ, and where the value actually comes from.
The four kinds of ICP tool
Almost every product in this category is one of four things. They get marketed with similar language, so it's worth knowing which you're evaluating.
Enrichment tools add data to records you already have: firmographics, technographics, headcount, funding. They make your existing definition better informed. They don't tell you whether the definition was right.
Intent tools tell you which companies are researching your category right now. That's a timing signal, not a fit signal. A company can be actively in-market and still be a bad customer.
Product-usage tools score based on how much someone is using your product or engaging with your content. That measures attention rather than fit. A poor-fit prospect who opens every email will outscore a perfect-fit prospect who doesn't.
Outcome-derived tools compute the profile from your closed-won deals and weight the traits by what actually correlated with winning. The model is derived, not declared.
Get pipeline intelligence insights weekly
Join sales leaders learning to turn CRM data into pipeline confidence. No spam. Unsubscribe anytime.
Most stacks need more than one of these. But only the fourth answers "who should we be selling to?" from evidence rather than from opinion.
The question that separates them: who set the weights?
This is the one to ask every vendor, and the answers vary more than the marketing suggests.
A lot of tools, including good and expensive and well-reviewed ones, let you assign weights by hand. You pick the signals that matter and give each one a multiplier. The tool then scores consistently against the model you built, and you tune it over several weeks until the results look right.
That's a genuinely useful thing to have. It's faster and more consistent than a spreadsheet. But be clear about what it is: your hypothesis, applied at scale. If you set the weights, the model inherits your assumptions, including the ones that are wrong. And "tune it until the results look right" means tuning until the model agrees with what you already believed.
The alternative is to compute the weights from outcomes. Feed in the deals you've closed, let the correlations decide what matters, and accept the answer even when it's uncomfortable. The most valuable output is usually the trait you didn't expect to matter.
Neither approach is fraudulent. But they answer different questions, and only one of them can tell you that you've been wrong.
A buyer's checklist
Ask these of any vendor, including us. The answers will tell you more than a demo will.
1. Where do the weights come from? Did someone assign them, or were they derived from outcomes? If assigned, by whom, and against what evidence?
2. Can I see why a specific account scored what it scored? Not a confidence blurb. The actual breakdown: which signals fired, which were missing, what each one contributed. A score a rep can't trace is a score a rep quietly stops using.
3. What happens when there isn't enough data? This is the sharpest question on the list. A tool that always produces a confident answer is a tool that will eventually produce a confident wrong answer. Ask what it does when the correlation isn't there.
4. Whose data is being scored, mine or yours? Scoring your CRM records and scoring a vendor's database are different products. You can verify the first by recognising the names. The second requires trusting someone else's data before you can act on it.
5. Does it change, and can I see what changed? An ICP built eighteen months ago describes a market that has moved since. Ask whether the model updates, how often, and whether you can see a record of what shifted and when.
6. What's the minimum? Every model has a floor below which the pattern isn't statistically real. If a vendor won't name theirs, that's an answer.
7. Where does the data live, and who else sees it? Relevant for anyone in the UK or EU, and increasingly a procurement question rather than an IT one.
What ICP analysis can't do
Worth stating plainly, because most vendors won't.
It needs enough deals. Telepath works best from around 50 closed-won deals. Below that the patterns aren't statistically meaningful. You'll get directional signal at best, and anyone promising a robust model on twelve deals is selling you noise.
It can't see what isn't in the data. If the thing that predicts your wins is a relationship, a referral, or a champion's personal history, no firmographic model will find it. The model describes what it can measure, not everything that matters.
Closed-won is a biased sample. A customer who signed and churned nine months later still counts as a win in the training data. A model trained on signings alone will happily recommend more of them. That's a real limitation of the whole category, ours included, and it's why we're building retention as a second and permanently separate score rather than folding it into this one.
It describes; it doesn't decide. A score tells you how closely something resembles your wins. It doesn't tell you the deal will close. Any tool presenting a resemblance score as a probability is overstating what it knows.
How Telepath approaches it
Telepath reads your closed-won deals, from a CSV or directly from HubSpot, and computes the weighting from what correlated with winning. Nobody assigns the weights.
Every load-bearing number is computed in code. The AI is only ever allowed to narrate a figure it has been handed. It never calculates one, and it never sees the raw deal data. When the data won't support a claim, it says so rather than producing a confident-sounding paragraph.
That last part costs us demos. "We can't find a correlation here" is a much worse slide than a chart. We've kept it, because a product that always finds something is a product under pressure to find things.
Built for the UK and Europe. Company data is enriched from Companies House, the statutory register, rather than a scraped third-party guess. Data is resident in the EU, encrypted at rest and in transit, on SOC 2-certified infrastructure, with full deletion on disconnect. We never train on your data.
See what your own data says
The ICP report is free. Upload a CSV of your closed-won deals, with company name, deal size, industry and close date at minimum, and you'll get your computed profile back in about three minutes. No signup, no card, no demo required.
If it tells you something you didn't know, that's the point. If it tells you nothing, that's worth knowing too, and we'll say so.
Related reading
- The signals that predict closed-won deals: how the analysis actually works
- ICP drift: why your ideal customer profile is outdated: what happens when nobody revisits the model
- ICP fit vs lead scoring: why they're not the same: activity versus fit
- How the T-Score works: the scoring architecture in detail
- Why your ideal customer profile isn't working: the workshop ICP problem