Data first, product second
Why I refuse to build the chat interface before the data layer is boring, typed and trustworthy.
- data
- product
- ai
There is a very tempting order of operations in AI products: build the chat box, wire it to whatever tables exist, demo it, get applause. It works for exactly one demo, on one dataset, with the person who built it holding the mouse.
I have shipped enough analytics work to know where that breaks. It breaks the moment two people define "active user" differently. It breaks when someone asks a question about last month and gets a number nobody can reproduce.
The order I actually use
- Get the data in. Boring connectors, idempotent loads, replayable history.
- Model it. Explicit grain, explicit keys, explicit definitions.
- Name the metrics once, in a typed layer that both humans and models read.
- Only then put a conversation on top of it.
- Only then let agents act, and score every action against outcomes.
A conversational interface is a multiplier. If the layer underneath is wrong, you have built a machine for producing confident mistakes.
Why this is a product, not a consulting gig
Every creator with an audience and every small company with a marketing team is running the same play with worse tools: exports, spreadsheets, a monthly ritual of reconciling numbers. The foundation is the same. The questions are the same. That is what makes it productizable.
So the roadmap is unglamorous on purpose: ingestion, modeling, semantics, then the interesting part.