Lesson 4 of 6
Grounded in real usage
6 min read
You're certain users struggle at checkout. You scope a whole AI feature around it. Then someone opens the analytics — and it's a different screen entirely.
Assumptions vs. real usage
The most common reason AI projects fail isn't the model — it's the data. Teams scope against an imagined user: what they assume people do, struggle with, and want. Grounding your scope in real usage — what people actually do — is what separates a feature that lands from one that solves a problem nobody had.
A hunch and the data can point at different screens. Scope from what users actually do, and you build for the real problem.
Check the data you'd build on
Before you scope, ask what you're deciding from. Do you have real usage data, or a confident guess? The strongest scoping is grounded: pull the analytics, read real transcripts, watch a few sessions. It's slower than a hunch — and it routinely sends you to build somewhere you didn't expect.
You also need data to build it, not just to scope it. If the feature needs examples or history you don't have, that's a feasibility problem — surface it now, not after launch.
The shape of it
- —The top cause of AI project failure is data, not the model.
- —Scope against real usage — analytics, transcripts, sessions — not assumptions.
- —No data to build on is a feasibility flag; catch it before you commit.
You're sure you know where users drop off. Before scoping the feature, what's the grounded move?
Continue in the app
Take the whole Discovering & Scoping AI Opportunities course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.