The most useful research is the research that makes you uncomfortable. The research that confirms what you already believe is just expensive reassurance.
I want to write honestly about a finding from my customer research that I spent several weeks trying to explain away before I finally accepted what it was actually telling me. I am writing it down partly to hold myself accountable to having learned the lesson, and partly because I think it describes something that affects far more founders than talk about it publicly.
The pattern I did not want to see
About eight months into building, I started noticing an anomaly in my interview data. I had been conducting structured discovery interviews with founders at various stages — early idea, post-customer-discovery, post-launch, post-raise. I was coding the interviews for themes and tracking patterns across the cohort.
The anomaly was this: founders who had stopped building — who had wound down or paused their ventures — were not primarily citing running out of money as the reason. A minority were. The majority cited something else. Something that was harder to categorise and that I initially resisted accepting as a real signal because it did not fit my thesis.
They had run out of conviction.
What running out of conviction actually means
This phrase sounds like a resilience failure. It sounds like the founders in question simply did not have the grit to push through. That interpretation was comfortable because it implied the problem was with the founders, not with the environment or the tools available to them. If it was a grit problem, it was not a problem I was responsible for solving.
But when I went back and asked follow-up questions — really dug into what running out of conviction felt like and when it happened — a different and more uncomfortable picture emerged.
In almost every case, conviction collapsed not because the evidence was definitively against the founder, but because the founder had no way to interpret the evidence they were seeing. They did not know whether what was happening to them was normal or a warning sign. They had no reference class. Every setback felt like a potential verdict on the entire venture.
Is a 15% email open rate good or bad for a cold outreach campaign? Is three months without a second paying customer a product problem or a distribution problem or a normal part of the journey? Is burning through £6,000 in the first quarter fast or slow for a pre-revenue startup at their stage?
None of these questions have universal answers. The answers depend on the business model, the market, the stage, the channel, and dozens of other contextual factors. But first-time founders, operating alone without experienced operators around them, were making these interpretations in isolation — and they were often getting them wrong in the pessimistic direction.
The implication I had been avoiding
The uncomfortable truth this finding pointed to was that my product needed to do something I had not designed it to do. It needed to give founders context, not just data. Not just a measurement of where they stood, but an interpretation of what that measurement meant relative to comparable businesses at a similar stage.
The difference between data and context is the difference between telling someone their heart rate is 95 beats per minute and telling them that is elevated for resting but normal for their recent activity level. The number means nothing without the frame.
I had been building a measurement tool. What the market needed was an interpretation tool. These are not the same thing. Measurement tells you where you are. Interpretation tells you what to think about where you are. The first is useful. The second is necessary.
What changed in the product
This finding drove the single largest architectural change I have made to the product. Instead of surface-level scores and metrics, the core output is now contextualised analysis — what your numbers mean relative to the stage you are at, the market you have chosen, and the business model you are operating. The benchmarks are not generic. They are specific.
This is significantly harder to build than a scoring system. It requires much richer data, much more nuanced logic, and much more careful presentation to avoid creating false precision in the other direction. But it is what the founders I am building for actually need.
The finding I almost explained away turned out to be the most important signal in my entire research corpus. I tell that story to myself every time I am tempted to dismiss something that does not fit the story I am already telling.