An ideal customer profile is usually delivered as a filter: this industry, that headcount band, this region, a funding stage. It is easy to agree with, it runs without argument, and it is close to useless, because none of those fields describe the thing that made your best customers buy. Here is the version that survived contact with a real pipeline, including the two rules we got wrong and had to delete.
01Start from the customers you already have
Not the market you could address. The addressable market is a number for a board deck; your existing accounts are evidence. Take the ten or twelve highest by lifetime revenue and ask what they have in common that is behavioural rather than firmographic.
When we did this on our own book the answer was unusually sharp: the top accounts by lifetime revenue cluster on three target families. That is not a coincidence about those particular customers. It is the reason they are customers, and it is the first thing a profile should encode.
The useful question at this stage is not who they are, it is what they were doing that made the purchase necessary:
- What were they already doing at volume before they bought? A company that has never done the thing manually rarely buys the automated version of it.
- What would break, and how quickly, if what you sell disappeared on Monday? If the honest answer is "not much", they are a user rather than a buyer.
- Who inside the company felt the problem? A named function with a budget is a prospect. A vague benefit to the whole organisation is not.
- What did the ones who churned have in common? That list is usually shorter and more informative than the list of wins.
02Write the test as a sentence somebody could argue with
A filter cannot be wrong in any interesting way. It just runs, and when it returns the wrong companies you adjust a threshold and it runs again. A sentence can be wrong out loud, which is the only property that makes a profile improvable.
So the centre of ours is a paragraph rather than a set of fields. It says what a prospect is doing, at what volume, and what signal shows that they would feel it if that stopped. A salesperson reading it should be able to say "no, that is not right, our best account does not look like that" and be immediately correct or immediately wrong. Fields do not offer that.
It also has to be written in the language of the buyer rather than the seller. A profile that describes your product with the nouns rearranged will match companies that describe themselves the way you describe yourself, which is a competitor, not a customer.
03The disqualifiers are worth more than the qualifiers
Qualifiers get you a long list. Disqualifiers are what stop you writing to the companies that look perfect on paper and can never buy, and they are almost never derivable from the target segments themselves.
The one we measured is the plainest example. A company that sells what we sell matches every qualifier we have: right segment, right vocabulary, right technical sophistication, obviously interested in the problem. Three of them cleared every gate and reached an approved message before anyone noticed they were competitors. Nothing in the target families says otherwise. It had to be written down as its own rule.
- Competitors, which match your qualifiers better than your customers do.
- The wrong end of the transaction. In some markets the famous name in a segment is the party being measured rather than the party buying the measurement.
- The wrong buyer inside a right-looking company. Agencies, outsourcers and consultancies genuinely do the work, and the budget sits with their client.
- Anyone who has asked not to be contacted, which outranks every qualifier you own.
04Two rules we deleted, and why they were wrong
Both were written the same way: somebody looked at a test run, saw companies that felt wrong, and reasoned backwards to a rule. Both read plausibly. Both cost real prospects before they were removed.
The first said a company large enough to do this in house will not buy it. The correction came from the person who runs the business: the largest companies in our market are actively moving toward becoming customers on exactly that path. Doing it internally at scale is not evidence that somebody will never buy. It is evidence they have the volume, the budget, and a team who already knows what it costs to keep working, which is the team that eventually stops building.
The second said that using the thing internally does not count, only reselling it does. Same mistake, same shape of reasoning, and it refused a run of large, well funded companies on the grounds that the work served their own business. That is precisely the situation that gets outsourced, and it is often the bigger account.
05How to know when yours is wrong: read the refusals
The approvals tell you almost nothing. Every profile approves companies that look like the profile; that is what a profile is for, and it is true of a wrong one as well as a right one. The refusals are the only place the shape of the thing is visible.
Over six months against our own business the pipeline read 4,182 companies, turned down 4,171 of them, and delivered 11 messages a human approved. That is 379 rejections for every one contacted, and the ratio is only worth quoting because every rejection carries a written reason. Both of the deleted rules above were found by reading those reasons in order rather than by inspecting the companies that qualified.
The exercise is cheap and most teams never do it. Take the first hundred companies your process turned down, in the order it turned them down, and read the reason attached to each. You are not looking for one bad call. You are looking for a run of the same wrong call, which is what a mistaken rule looks like from the outside.
If your process cannot produce that list, that is the finding. A profile with no refusal log attached cannot be corrected, only replaced with a differently confident guess, and a guess that has been rewritten three times is not more likely to be right than the first one.
The short version: derive it from the accounts you already have, write it as a sentence a colleague can contradict, spend most of the effort on the disqualifiers, and check it against your own rejections rather than your own wins. Ours was wrong in two places after six months of running, and the only reason we can say which two is that we wrote down every no.