We are starting a research company. There are already a great many of those, so the honest place to begin is with what we think is missing.

There has never been more data about customers than there is now, and rarely less understanding of them. Companies measure more than they ever have and still cannot say with much confidence why a customer did the thing that made them money, or what would make them do it again.

Begin with what a business actually runs on. It runs on things customers do: they buy, they come back, they recommend, they renew, they upgrade. Everything else matters only to the extent that it eventually shows up in one of those. Now look at what most companies know about their customers. Almost all of it is what people say. Almost none of it is what people do.

That would be a smaller problem if saying and doing were the same behavior. They are not. Answering a question about oat milk happens at a kitchen table, to a stranger, with time to think and a mild wish to sound reasonable. Buying oat milk happens in a store, in a few seconds, on the way to something else, with a half-full cart. Different place, different audience, different job. Nobody is lying on a survey. They are simply doing a different thing than the one you wanted to understand.

So why has an entire industry built itself on the weaker of the two? Not because anyone thinks it is better. Researchers have known for fifty years that one good conversation about what a person actually did teaches you more than a hundred rating scales. The problem was never the method. It was the price. Depth meant interviewers, recruitment, scheduling, transcription and a coding team, which meant depth had to be rationed: twelve people, six weeks, one question. Anything that needed scale got a survey instead. The industry did not settle for measurement over explanation because it preferred it. It settled because explanation cost twenty times more and took ten times longer.

That constraint is the thing that has just gone.

The moment a survey response lands, software can now read that one person's answers - their pattern, not the aggregate - work out what is worth asking about, and put a real interview in front of them the same evening, in their own language. At hundreds of respondents rather than twelve. Depth and scale stopped being a trade.

It is worth being precise about what that does and does not mean, because much of what is currently sold as AI in research points the other way: faster surveys, cheaper panels, synthetic respondents answering questions no human was ever asked. That is the bottom rung of the ladder, made more efficient. It produces more of the same data for less money. We are interested in the opposite trade, which is to use the cheap layer to work out exactly where the expensive one is worth spending.

There is a second reason we started, and it is older than any of the technology. Marketing works almost entirely on what happens before a purchase: the ad, the pack, the claim, the price, the shelf. All of it is well instrumented, because all of it is under your control. But whether someone buys again is decided by what happened after, by what the purchase actually did for them. What it solved, replaced, avoided, or made possible. Almost nobody measures that side, and it is the more stable of the two. Channels shift, algorithms change, attribution models get rebuilt every year. What a product does for a person moves slowly, sometimes not for decades. It is the most durable thing you can know about your market, and it is very nearly unmeasured.

So that is the company. We work in three steps, and they are a sequence rather than a menu. We start with the behavioral data a client already owns - transactions, loyalty, bookings, check-ins, logs - because it sits at the top of the evidence ladder and has already been paid for. Then studies: survey and interview together, asking what people did, when, who was with them, and what happened next. Then experiments: change one thing, hold a control, measure. The records tell us where to look, the interviews tell us what to change, and the experiment tells us whether the change worked.

None of this is a new idea, and that is the part worth being clear about. It is an old idea that was correct the whole time and could not be afforded. Our contribution is not the insight. It is the arithmetic.

We did not add AI to research. AI made the right method affordable.

Actions over attitudes. That's Superfact.