CB Insights read through 101 startup failure post-mortems and found one reason ahead of all the others: building something nobody actually needed. Not running out of cash. Not a bad team. Just a solution to a problem that, once tested against a real customer, turned out not to be a problem worth paying to solve. That single cause was cited in 42 percent of the failures (CB Insights, 2014), more than twice the next most common reason. It's an old study by internet-years standards, and CB Insights has run variations of it since with similar results, which is the part that should worry you: this isn't a phase teams grew out of.
Customer discovery is the stage that's supposed to catch this before the money is spent. Done properly, it isn't a courtesy step between the diagnostic and the build. It's the point where an assumption either survives contact with an actual customer, or it doesn't, and you find out for a few thousand dollars and three weeks instead of a full build cycle.
What a discovery interview is, and what it isn't
A lot of what gets called customer discovery is a survey with a microphone. Someone asks, "would you use a tool that does X," the person on the other end says something polite, and the team writes down "validated" and moves on. That's not discovery. It's asking someone to predict their own future behaviour, which people are reliably bad at, especially when they're being nice to you.
A real discovery interview asks about the last time, not the next time. Not "would you use a scheduling tool," but "walk me through the last time a shift got double-booked, what did you actually do about it." The first question gets you a hypothetical. The second gets you a real Tuesday afternoon, with names, tools, and workarounds attached to it. Workarounds are the tell. If someone has already jerry-rigged a fix out of a shared spreadsheet and a group text, that's a stronger signal of real pain than any amount of enthusiastic nodding at a pitch.
Who to talk to, and how many
Six to twelve interviews is the range we use for a single diagnostic hypothesis, the same range that shows up across our own Diagnostic Mirror engagements. Discovery usually needs more breadth than that, because you're rarely testing one assumption. A retail client mapping why online orders weren't converting needed three separate interview groups: people who bought, people who abandoned a cart, and people who never got that far. Each group told a different story, and only one of the three matched what the merchandising team had assumed going in.
The bias that ruins more discovery processes than any interview technique mistake: talking only to easy, friendly customers. Your best client in Moncton who takes your calls and likes you personally is the least representative person in your customer base. The interviews that actually change a decision usually come from someone who churned, someone who chose a competitor, or someone who's never heard of you at all but fits the profile.
Turning conversations into something a leadership team can act on
A stack of interview transcripts isn't a deliverable. Three things need to come out the other side of synthesis, and if any is missing, the discovery process didn't finish its job.
- Themes, not anecdotes. If three of nine interviewees independently describe the same workaround without being prompted, that's a theme. One person's colourful complaint is not.
- Verbatim quotes attached to each theme. A paraphrase gets forgotten by the next board meeting. A direct quote ("I just keep a second spreadsheet because I don't trust the system's numbers") carries the point on its own.
- Personas tied to revenue impact, not marketing colour. "Busy parent, age 35-44" tells you nothing about the decision in front of you. "Segment that generates 60 percent of repeat revenue but has the highest support-ticket volume" tells you where to look first.
Why teams skip this, and what it costs
Nobody sets out to skip discovery. It gets rationalized away, usually with one of three lines.
- "We already talk to customers all the time." Ad hoc feedback from whoever happens to call in is real information, but it's self-selected and unstructured. It's not the same as a deliberate sample designed to include the people who don't call in.
- "We don't have six weeks for this." A scoped discovery pass, eight to ten interviews around one specific decision, runs closer to two weeks. The six-week version is for a full strategy engagement, not a single go/no-go call.
- "The founder is the customer." Sometimes true, and worth real scrutiny when it is. A founder who used to do the job the software replaces has real domain knowledge, and also a sample size of one.
Ignoring customer input outright showed up in 14 percent of the same CB Insights post-mortems (CB Insights, 2014), roughly tied with poor marketing for eighth place on the list. Combined with the 42 percent who built the wrong thing in the first place, well over half the failures in that study trace back to a version of the same root cause: the team never had a reliable read on what the customer actually needed, and built ahead of finding out.
Where this sits in the pipeline
Customer Discovery is Stage 2 of our nine-stage methodology, and it's not always the right next step after a diagnostic. If your diagnostic recommendation is entirely internal (a process fix, a data cleanup) discovery usually isn't needed yet. It becomes essential the moment a recommendation touches something a customer will see, use, or pay differently for. We wrote separately about how to run a diagnostic that produces decisions, not a binder, which is usually Stage 1 and the step before this one. If you're still deciding whether your organization is ready to start any of this, our piece on 5 signs you're ready for digital transformation (and 3 signs you're not) is a reasonable place to check first.
The honest test of a discovery process is whether it changed a decision. If nine interviews confirmed exactly what the leadership team already believed, something in the sample or the questions was probably too safe. Real discovery produces at least one uncomfortable finding. That's the one worth paying attention to.