discovery — how much is enough
Every team starts somewhere on a confidence spectrum
Before deciding how to do discovery, decide where you are starting from. The same discipline looks very different depending on how much is already known — and applying the wrong depth wastes time or leaves dangerous gaps.
| Starting point | How to close the gap | What happens if you don't |
|---|---|---|
| New domain or approach | Watch before you name. Enter without a solution hypothesis. The gap that seems obvious is usually not the one that matters. Two to three days before any direction. | A confident solution to a problem described from memory. The real problem appears in production at maximum cost. |
| Existing client, new problem area | Name your assumptions first. Test the consequential ones — where being wrong changes the direction, not just a detail. | Assumptions from the last project that no longer hold. They appear as scope changes three sprints in. |
| Recurring pattern, different client | Targeted validation. Confirm the differences: this client's context, constraints, edge cases. The 20% that differs is often the 80% that matters. | Pattern built without checking fit. The edge cases that don't hold appear during QA or after release. |
| Internal tool, team is the user | You are a witness. Name what you are assuming about team members who use it differently. Talk to them specifically about the difference. | Built for yourself. Others don't mention edge cases for months because the tool mostly works for them too. |
How to apply this
- ✓ Before any session, write down what you currently believe and why. The why is what you are testing. If you cannot name why you believe it, you cannot test whether it is true.
- ✓ Is the depth proportional to the irreversibility of the decision? Expensive-to-reverse decisions warrant more discovery. Easily-changed decisions warrant less.
- ✗ Do not treat observation as the only form of discovery. Match the method to the unknown — watching, log analysis, metrics, targeted interviews are all valid depending on what is unknown.