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
Structured interviews are the strongest predictor in the selection literature and also the most inconsistent one. The gap between those two facts is made of decisions you are already making without noticing: how many elements of structure, how you score, whether you probe, and how the question is worded.
Roughly half of B2B sales hires are not hitting quota. The Bridge Group's 2026 account executive report, a survey of 158 B2B companies run in the first half of 2026, puts 48 per cent of reps at quota, down from 51 per cent in 2024. The same report says the experience companies demand at hire has risen to 3.7 years, up from 2.7 in 2022.
Read those together and you get the shape of the problem. Companies are raising the bar at the door and still getting a coin flip. Buying more years of experience has not fixed it, which suggests the filter is not experience. It is the interview.
The industrial psychology literature has a clear answer on interview format, and it got clearer recently. In a 2023 commentary in Industrial and Organizational Psychology, Huffcutt and Murphy report that in the Sackett and colleagues meta-analysis, structured interviews emerged as the selection predictor with the highest mean validity. Not cognitive ability tests, not work samples, not assessment centres. Structured interviews.
They add a second finding worth more than the first to anyone hiring in a small company: structured interviews showed significantly lower impact on racial groups than the other top predictors, including biodata, knowledge tests, work samples, assessment centres and general mental ability. The best predictor is also among the fairest.
Then comes the uncomfortable part, and it is the reason this article exists. The same paper reports that structured interviews have the highest variability in validity among the top predictors, at .42 plus or minus .24. That spread is enormous. A method whose average is the best available, and whose variance is also the worst available, is not a method you can adopt by name. "We do structured interviews" tells you almost nothing about whether this particular process predicts anything.
Huffcutt and Murphy group the sources of that inconsistency, and three of them are decisions you are making right now, probably without noticing.
How much structure, and which parts. They write that there are as many as 18 different structuring elements, citing Levashina and colleagues, and that an average of six are actually incorporated. Six of eighteen. Two companies can both say "structured" while sharing only one or two elements in common.
How you score it. Rating each question individually using behaviourally anchored scales is common, but the paper notes that some interviewers use generic scaling such as a Likert scale, and some wait until the interview is over before rating anything. Those are different instruments producing different numbers.
Whether you probe. They name this explicitly as another common methodological variation: whether to allow, discourage, or even prohibit follow-up questions. Most interviewing teams have never made this decision deliberately, which means it varies by interviewer and by candidate, which is exactly how structure leaks out of a structured process.
Decide these three before you book the first call, write them down, and apply them identically. That is most of the gap between the top of that .42 range and the bottom.
The paper makes a point I had not seen put so plainly: surprisingly little attention has been paid to how structured interview questions are worded, and wording changes what the question measures.
Their example is, conveniently, a sales question. "Can you tell me about an occasion where you had to deal with an angry client?" The authors point out this could easily be answered with an experience reflecting either typical performance, meaning day-to-day tendencies, or maximal performance, meaning the best the person has ever managed.
Those are not the same thing and you almost certainly want one of them. If you are hiring someone to handle twenty renewal conversations a week, their finest hour of client recovery is close to irrelevant. You want the day-to-day. But the question as written invites the highlight reel, and any candidate who has read interview advice will give you exactly that.
The fix is in the wording, not the follow-up. Ask for the typical case by naming it: the last time, not a time. "Tell me about the most recent client who pushed back on price" cannot be answered with a career highlight, because it specifies which memory to retrieve. Our write-up of discovery call questions that decide the deal makes the same argument from the other side of the table.
The common convention, the paper notes, is to ask about all aspects of an experience at once in a single question, frequently structured around the STAR format of situation, tasks, actions and results, and to expect a unified and complete response shortly after the question is read.
They count what that demands. A past-experience question involves no fewer than five cognitive steps: deciphering the question, searching long term memory, comparing experiences to pick the best one, recalling its details, then formatting all of it for oral delivery. Under time pressure, in front of strangers, with a job at stake.
What you are measuring at that point is partly sales ability and partly the capacity to perform a memory task at speed. Those correlate, but not enough to confuse them, and the second is heavily practised by candidates who interview often.
Their suggested alternative is borrowed from how legal professionals gather eyewitness testimony, which they note is conceptually very similar to describing a past experience: guide the person through recall in a series of small steps rather than demanding the finished narrative in one go.
In an interview that means asking the situation first and stopping. Then what they were responsible for. Then what they did. Then what happened. Four short exchanges instead of one large request. It removes a variable you were never trying to measure.
There is an encouraging implication buried in the paper. Discussing why cognitive load varies between jobs, the authors note that in some occupations really difficult challenges do not occur often, which makes the memory search cumbersome, while in areas such as customer service difficulties occur frequently and are easily recalled.
B2B sales sits firmly in the second group. Anyone who has worked a real pipeline has lost deals, been ghosted, been beaten on price and had a champion leave mid-cycle, repeatedly and recently. The retrieval problem that makes behavioural interviewing unreliable for some roles is much smaller here, because the material is abundant and fresh.
That has a direct diagnostic use. If a candidate with claimed experience cannot produce a specific recent loss with detail, the most likely explanation is not a bad memory. It is that the pipeline was thinner than the CV suggests.
Two things this process does not decide. Whether you should be hiring at all, which is a different and prior question covered in when to hire your first salesperson, and which of the sales jobs you are hiring for, since prospecting and closing are different roles with different interviews, as we argued in how to structure a small B2B sales team.
The validity figures above are means across many jobs and studies, and the point of the .42 plus or minus .24 finding is that the spread is wide. None of them is specific to B2B sales, and I could not find a published estimate that is. Treat the direction as well evidenced and the magnitude as not transferable.
The paper drawn on here is a commentary, not a meta-analysis in its own right. The Sackett and colleagues figures reported above come through that commentary, which is where I read them, not from the original paper. The adverse impact finding concerns racial groups in a largely United States literature, and should not be assumed to transfer to hiring in the Gulf or North Africa.
The Bridge Group figures are survey data, which the report itself describes as observational rather than experimental, with a sample that skews toward organisations with engaged revenue leadership. Forty eight per cent at quota is a real signal about the market, not a benchmark your next hire should be measured against.
And the ceiling on all of it: a better interview raises your odds, it does not make the decision safe. The strongest predictor available still leaves most of the variance in performance unexplained, which argues for a short, well instrumented ramp with early checkpoints rather than more interview rounds. Their timing is in sales ramp time for new hires.
Send three things: the role as you would advertise it, the questions you are planning to ask, and your median sales cycle over the last ten deals. We will come back with a scored question set, the anchors for each answer, and the two questions worth cutting because they measure interview practice rather than selling.
Book a Free Consultation