Customer discovery interviews, and how many are enough
B2B customer discovery interviews: the sample size the research converges on, why your segments change that number, and the questions that get honest answers.
Read
The version on the slide describes a market and excludes nothing. The version that is worth writing tells one person which accounts to put in this week's sequence, and which accounts to leave alone permanently.
Most ideal customer profiles are written once, presented on a slide, and then never opened again by the person who actually has to send the emails. It describes a market in language nobody can act on: mid market companies in growth sectors with a need for efficiency. Every account passes. Nothing is excluded. The list gets built from whatever the data vendor had in stock.
An ICP that earns its keep does something narrower. It tells one person, on a Tuesday morning, which accounts to put in this week's sequence and which to leave alone, and it does it without a meeting. That is a different document from the one in the board deck, and it is built differently.
The strategy version answers who we sell to. It exists to align a budget and it can afford to be broad, because nothing downstream breaks if it is.
The prospecting version answers two questions instead, and the second one is the one people skip. First, is this account eligible at all. Second, is there a reason to contact it this week rather than in eight months. Eligibility is firmographics. The reason is a signal, and without it you have a list of companies that look right and no idea which of them is in motion.
That distinction decides the shape of the whole document. An ideal customer profile for outbound is a filter plus a trigger, not a portrait.
Run a draft through these before anyone starts building. Each one fails a common version of the document.
Take your last twenty to forty closed won deals. Not your current customer list, which flatters you with accounts that have since grown. What you want is what each account looked like on the day it decided to buy.
Then look for the boring shared attributes rather than the flattering ones: size band, structure, who held the budget, what they were already using, what had just changed. The pattern is usually duller and narrower than the slide version, and that is the point.
Do the counter check as well, because it is where the real exclusions come from. Pull the closed lost deals that looked identical to your winners on paper and ask what separated them. Often it is one thing: no internal owner, a system you cannot integrate with, a procurement route that adds nine months. Those become disqualifiers, and disqualifiers are the most valuable lines in the document.
The method generalises past B2B sales, and it is the same instinct behind starting from the people who already paid rather than from assumptions on a slide. The uncomfortable part is that the winners often share something you did not choose deliberately, such as a referral route or a single integration, and the honest version of the document names that rather than dressing it up as strategy.
Keep the two halves separate in the document itself, because they are maintained on different clocks. The filter changes once or twice a year. The signals are checked weekly.
| Line in the ICP | What it decides | Where you actually get it |
|---|---|---|
| Size band and headcount | Whether the problem is big enough to pay for | Company registry, their own site, a list vendor |
| Sector and sub sector | Whether your proof applies | Your own closed won set |
| Buying structure | How long the deal will take | Closed won and closed lost notes |
| Systems in use | Whether you can integrate or must replace | Job ads, their careers page, public case studies |
| Hard disqualifiers | Which accounts never enter a sequence | Closed lost analysis |
| Trigger events | Which eligible account to contact this week | Funding, hiring, leadership change, new location |
The last row is the one that turns a profile into a prospecting instruction. Which triggers are worth acting on varies by what you sell, and we went through the ones that tend to pay in trigger events in B2B prospecting. The fields themselves, and how few of them you need, are in how to build a B2B prospect list.
This is the part that justifies the hour spent on the document, and it is worth citing rather than asserting.
Gartner's sales survey of 632 B2B buyers, published in June 2025, found that 73% of B2B buyers actively avoid suppliers who send irrelevant outreach, and that 61% said they prefer an overall rep free buying experience. Robert Blaisdell, VP Analyst in the Gartner Sales Practice, put it plainly in the release: "Bad prospecting actively damages relationships with potential customers." The full figures are in Gartner's press release on the survey.
Read what that means for a loose ICP. A badly targeted sequence is not a neutral event that produces no reply. It removes the account from the set you can reach later, because the next person there who hears your company name has already filed it under irrelevant. A list of three hundred poorly qualified accounts can burn more future pipeline than it creates now.
It also explains why volume cannot compensate for a weak profile. If more touches on the wrong accounts actively subtracts, the fix is upstream of the sequence. Whether outbound is even your constraint is the prior question, and we argued it in outbound versus inbound for B2B.
Format matters more than it sounds, because the document has to be usable at the moment someone is deciding whether to add an account.
One page. Disqualifiers first, since they do the most work per line. Then the filter fields with the exact values, not adjectives: "50 to 400 staff", not "mid sized". Then the trigger list with where each one is checked. Then one line naming who may approve an exception and on what grounds, because exceptions will happen and an undocumented exception becomes the new de facto profile.
Keep scoring out of it at this stage. A score is useful once you have enough history to calibrate it, and premature scoring hides the reasoning inside an arithmetic that nobody questions. That failure mode is worth reading about before you automate anything: rivl covers it for scoring models in AI lead scoring for sales, and when the number is just a guess.
If you have fewer than about ten closed won deals, you cannot build this from data, and pretending otherwise just dresses a guess in a table. Write it as an explicit hypothesis instead, note the date, run it for thirty to sixty days, and review it against what actually replied and what actually closed. The early version of this document is a bet you are keeping score on, not a finding.
Two more honest caveats. An ICP is not a persona: it describes the account, while the persona describes the human you write to, and conflating them produces a document that is bad at both. And it decays. Your win pattern from two years ago reflects a product and a price you may no longer have, so re check it against the last two quarters of closed won at least twice a year.
Last, the uncomfortable one. A tight ICP will sometimes exclude an account you would have won. That is the cost of the trade, and it is usually the right trade, because the alternative is spending the same hours across a list where most accounts were never going to buy. If the exclusions start to feel expensive, that is a reason to examine them deliberately at a review, not a reason to widen the filter quietly in the middle of a quarter.
Send three things: your last twenty closed won accounts with their size and sector, the two or three closed lost deals that looked identical on paper, and the filter you are using now. We will come back with the disqualifiers we think are missing and which trigger is worth checking weekly.
Book a Free Consultation