By Thursday, you have lost most of what you learned Tuesday.

It is 8pm in Lagos and a PM I know is sitting with three tabs open. An Otter transcript she has not read. A Notion page with nine bullet points. A Linear ticket that records a decision but not the reason for it. She ran a discovery call this morning. It was a good one. The customer said something about the approval flow that changed how she thinks about the quarter. She is now trying to remember what it was.

She will not find it. It is not in the transcript, because the customer never said it directly. It is not in the Notion page, because the bullet points capture actions, not interpretation. It was in her head at 9:47 this morning, sharp and specific. By Thursday it will be a feeling. By the following Tuesday it will be gone.

The three-week evaporation

This is the three-week evaporation, and every product team runs on top of it. The half-life of a discovery insight is measured in days. Not because PMs are careless. Because the tools they use capture the wrong layer. A customer conversation has four dimensions. What she said. What she meant behind what she said. What she did while she said it. What she felt. Every tool on the market captures the first one, some capture the third, none capture the second or fourth as structured data.

The Thinks column, unclaimed for two decades

The second dimension is the one that matters. We call it the Thinks column, and it has been unclaimed white space for two decades. The empathy map has been product craft since Dave Gray drew it at XPLANE in the early 2000s. Every design school teaches it. Every UX lead has sketched one on a whiteboard. And no tool has ever shipped it as first-class, queryable data. Dovetail modelled transcripts and tags. Says only. Maze modelled behaviour. Does only. Granola summarises. Notion AI paraphrases. Nobody carved out Thinks as a separate field with its own author, its own confidence level, its own evidence link back to the moment in the call that triggered it.

The reason is conceptual, not technical. Thinks looks like commentary, so tools treat it as a note. It is not commentary. It is the load-bearing artefact of the call. It is the moment a PM turns 'yes, this looks good' into 'she means only if the fees drop below 1.5%'. Without that layer, the transcript is inert. With it, the transcript becomes something you can build a roadmap on.

Four dimensions against four risks

There is a second frame we borrow, from Marty Cagan and SVPG. Every product decision carries four risks. Will the customer choose it (Value). Can she figure out how to use it (Usability). Can we build it (Feasibility). Does it work for the business (Viability). Value is the question asked in every standup, and it has never been organised as a first-class query. Ask your team right now: show me every insight tagged Value from the last 90 days. Nobody can answer it. The 4x4 grid of four dimensions against four risks is sixteen cells, and every decision your team makes lives in one of them. That grid is the moat. Any one cell is a feature. The grid is a category.

Why African product teams first

We are building this for African product teams first, and I want to be direct about why. The knowledge-decay problem is sharpest here. Teams are smaller, so each PM carries a bigger share of institutional memory. Turnover is faster. Budgets are tighter, so the per-seat pricing of the established research platforms puts them out of reach. Connectivity is variable, so tools that assume fibre fail. Building for Lagos, Nairobi, Cape Town, Cairo, and Accra forces a product that is cheap, fast on a 10Mbps upload, WhatsApp-native, and honest about what it stores. That same constraint produces a sharper product in London and San Francisco. Knowledge decay is not an African problem. It is a product-team problem. The African version is simply the hardest test, so we are taking the test first.

Where we are today: the substrate works on our own calls, and the briefs shown on this site are illustrative examples of its output. V1 is four items. The interview brief generator, under 90 seconds from upload to a decision-ready brief. The shareable brief URL. The corpus query interface. The MCP server, so every AI agent your team already uses can read the same substrate. We ship to design partners in weekly cohorts of 50, starting Q1 2027.

If any of this reads like something you have been arguing for privately for two years, join the waitlist. Reply to my launch email. Send me a Loom of your discovery-call workflow. We are building the substrate you would have built if you had six months and two co-founders. We do. So we are.