What a good AI discovery actually looks like
If your discovery is two Zoom calls and a slide deck, it's not a discovery. It's a sales pitch with extra steps.
Discovery is the first 2–3 weeks of an engagement, before the build starts. Done well, it saves the next 6 months. Done badly, it guarantees the build will miss.
What a good discovery includes
- Read access to your real data — CRM, job-management system, support inbox — for 14 days. Not a demo dataset. Yours.
- 1:1 conversations with the people doing the work, not just the founder. The receptionist will tell you more about the leak than the founder will.
- Two weeks of measurement on the three numbers that matter, before any build decisions are made.
- A written one-page brief at the end. What we'll build. What we won't. What success looks like. What we'd turn off if it goes wrong.
- A line-item budget. Build cost. Run cost. Maintenance cost. No bundled hand-waving.
What a bad discovery looks like
- Two Zoom calls and a deck.
- No access to real data.
- No conversation with anyone except the founder.
- No written brief — 'we'll figure it out as we go'.
- Bundled pricing — 'enterprise license' or 'transformation package'.
If a vendor's discovery looks like the second list, you'll know within a month why the build is missing.
A good discovery feels slow. A bad one feels exciting. Pick slow.
— Aaron Manton, AI Operator Club