The Best Demo We Ever Built for a Market That Wouldn't Pay

In 2015 I co-founded Trensant and ran product as CPO. We did text analytics over enormous corpora: Twitter, Facebook, and tens of thousands of other sources, converted into numbers, assembled into a knowledge graph of the entities and concepts people were talking about, with analytics on top. We raised >$6M. Interos.ai acquired us in 2017.

The case study covers what we built. This is about the decision in the middle, which is the part founders actually ask me about, and the part that never makes it into a case study because it doesn’t flatter anyone.

Our primary market was consumer. That is where the team’s conviction was, where the product roadmap pointed, and where I would have told you in early 2016 the company was going.

Lesson 01

The demo that proves your technology is rarely the market that funds it

Months before the 2016 US presidential election, our data scientists ran the engine over what people were actually saying and concluded Trump would win handily. This was against every major poll and essentially the entire press. We assumed we had a bug. We went looking for it and could not find one. We never published the prediction, which I still think was the wrong call, and we did eventually publish a paper showing how the sentiment moved over time.

The forecast landed within one electoral vote.

Set aside the nerve lesson, which is real but is a different essay. Look at what the episode actually proved: the engine could ingest a firehose of unstructured public text and extract a signal that professional instruments had missed, at national scale, months ahead. As a technical proof point it is the best thing we ever produced.

It was also a consumer-data artifact pointing at a consumer market, and consumer sentiment analytics in 2016 was a market full of people who wanted to talk to you about it and nobody who wanted to write a cheque for it. The demo proved the technology. It did not locate the buyer. I had quietly assumed those were the same problem, and they were not even adjacent.

If your most impressive artifact lives in a market that will not pay, it is not traction. It is an expensive advertisement for a capability you have not yet aimed anywhere.

Lesson 02

Runway is what makes an ICP sharp

I would love to tell you we ran a rigorous segmentation exercise, scored the options, and converged on a defensible ideal customer profile.

What happened is that we started running out of money.

Every founder has heard “narrow your ICP” and most of us nod along and then do not do it, because narrowing feels like subtraction and the deck already says the market is enormous. Breadth is comfortable. As long as five markets are theoretically open, none of them has told you no yet.

Runway removed the comfort. When you can count the months, you stop optimizing for the size of the opportunity and start optimizing for the shortest distance to a signed agreement. That is a worse way to build a category and a much better way to find out whether anyone actually wants what you have made.

The uncomfortable conclusion I have come to since, having now watched this from the inside at a number of companies: most teams do not focus until something forces them to. If nothing is forcing you, the honest question is not “what is our ICP” but “what would we do if we had five months of cash,” and then whether there is any reason not to do that now.

Lesson 03

A sharp ICP has a name, not a segment

One of our investors introduced us to Interos. They had a concrete problem in supply chain risk: understanding exposure across a network of suppliers, from sources that were mostly unstructured text. Our engine was, structurally, already good at exactly that. Nobody had aimed it there.

So we aimed everything at their use case. Not at “enterprise risk intelligence” as a category. At the specific problem in front of one named organization that would take our call.

This is where I part company with how ICP usually gets taught. An ideal customer profile is presented as a smaller circle drawn on a market map: firmographics, a segment, a persona, a total addressable market you can still defend to a board. That version is comfortable because it stays abstract, and abstraction is what lets a team keep believing five markets are open.

Sharp and pointy meant n equals one. A company, not a segment. A named buyer with a real problem, a budget, and the ability to tell us we were wrong on a specific Tuesday. Every product decision for the next stretch could be settled by asking whether it moved that one relationship forward, which is a question a team can actually answer, unlike whether something serves the enterprise risk segment.

The thing nobody tells you is that this is also how you find out what you have built. Serving one buyer completely forces every vague capability into a specific shape. Interos eventually acquired us, and the reason the engine was legible enough to acquire is that we had spent the preceding period making it legible to exactly one demanding reader.

Lesson 04

The bill arrives before the validation does

We walked away from consumer entirely.

Not deprioritized it, not put it in a later phase of the roadmap. Left it. That was our primary market, the one the team had been building toward, and the one where our single most impressive result lived.

This is the part of narrowing that the advice consistently understates. The cost is real, it is paid immediately, and the validation that it was the right call arrives much later, if it arrives at all. On the day you make the decision you are giving up a known thing your team believes in for an unknown thing one customer might want. Everyone can see the subtraction. Nobody can yet see the addition.

Which means narrowing is not really an analytical exercise. Analysis will happily support waiting. It is a decision about what you are willing to stop doing while people you respect think you are wrong, and the reason most teams fail at it is not that they cannot identify the sharp ICP. It is that they cannot pay for it.

What I would not conclude from this

Our one named buyer became our acquirer. That is a good outcome and it is also survivorship, and I would be selling you something if I let it stand as the lesson.

Concentrating on a single customer is genuinely dangerous. You can build precisely the thing one organization wants, discover it generalizes to nobody, and have narrowed your way into being a contractor with a cap table. That risk was live for us and it happened to break our way.

What transfers is not “go find an acquirer.” It is this. Your best demo and your best market are different questions, and confusing them costs years. Focus usually arrives as a consequence of pressure rather than insight, so it is worth asking what you would cut under pressure before the pressure is real. A profile sharp enough to change what you build on Monday has a company name in it. And the cost of that sharpness is charged up front, in full, before anyone can tell you whether you were right.

That is the whole lesson, and I paid $6M and two years of a good team’s life to learn it.

At SproutVest I work with founders and funds on exactly this problem: deep technology that demonstrably works, aimed at nobody in particular yet, and a board meeting coming. If that is the shape of the thing in front of you, I am happy to compare notes.