Your ICP is probably three ICPs wearing a trench coat

Ask most sales teams to describe their Ideal Customer Profile and you’ll get one confident paragraph: mid-market, a couple of named industries, a headcount range, a short list of target countries. Ask them to describe the three most recent deals that actually closed, and the paragraph starts falling apart — because those three deals rarely came from the same profile at all. One was an enterprise account that took nine months. One was a scrappy scale-up that closed in three weeks off a single good conversation. One barely resembles either. There isn’t one ICP in that story. There are three, sharing a document because nobody separated them.

How the blend happens

Nobody sets out to build a blended ICP. It accumulates. Someone writes down the profile of the first big win. Someone else adds “or a smaller company if the fit is otherwise strong.” A third person widens the geography because a good deal came from an unplanned market. None of these additions are wrong on their own — each one describes a real customer. The problem is treating the union of all of them as a single target, then feeding that single target into one scoring model and expecting the output to mean the same thing for every lead it touches.

What a blended ICP does to your scores

A scoring model can only optimise for the profile it’s given. Feed it a blend and it does the mathematically honest thing: it finds the average, which usually isn’t a great fit for anyone. An enterprise-shaped lead scores lower than it should because the model is also weighing scale-up signals against it. A scale-up-shaped lead scores lower than it should for the opposite reason. Everyone lands in the “potential match” band, nobody lands in “excellent,” and the queue stops being useful for deciding who gets contacted first.

Symptoms worth watching for:

• The 90–100% band is nearly empty, even though you know strong-fit leads exist in the list
• Deals that close well came from leads that scored only moderately
• Reps informally override the queue order, because their gut says the model isn’t finding the good ones
• Company-size fit and industry match keep pulling the score in opposite directions on leads you’d call obviously good

If your best customers don’t cluster, your ICP isn’t one profile with some noise around the edges. It’s several profiles you haven’t separated yet.

Splitting it apart

The fix is rarely to narrow the ICP until it only fits one segment — that throws away real, working parts of the business. It’s to stop asking one profile to describe all of them and instead define each segment on its own terms: its own industries, its own size range, its own geography, sometimes its own weighting of what matters. An enterprise profile might weight company size and outsourcing probability heavily, because a nine-month sales cycle needs a company with the budget and structure to sustain it. A scale-up profile might weight role and technology relevance instead, because speed matters more than scale there.

Running multiple ICP configurations in parallel — rather than one compromise profile — means each segment gets scored against criteria that actually describe it, and the queue produces genuinely separated priority lists instead of one muddled one.

Three questions that surface the blend

If you’re not sure whether your ICP is secretly three, three questions usually surface it fast:

1. Do your best-fit customers share a company size, or do they cluster into two or three distinct sizes?
2. Would the same outreach message genuinely work for all of them, or does it only really land for one group?
3. If you removed any single deal from your “great customer” list, would the profile description actually change?

If the honest answer to any of those is “it depends which one,” the ICP isn’t one profile yet. It’s a coat rack with three coats on it, and the scoring model is trying to describe the coat rack.

Fewer, sharper profiles beat one wide one

A narrower, correctly-scoped ICP will always look smaller on paper than one wide compromise profile. It will also produce a scoring model that actually separates excellent leads from mediocre ones — which is the entire point of scoring in the first place. Splitting a blended ICP into two or three honest ones is not more work for the sake of it. It’s the work that makes the rest of the pipeline mean something.