The pattern is so common it's almost boring. A digital health company runs a clinical AI pilot. It gets to a demo. Maybe to a small clinical sample. Then it stalls. Three months pass. The team is frustrated, the executive sponsor is quiet, and the clinician audience has lost interest.
The same five things kill pilots:
- No success criteria written down before the pilot started. The pilot is "let's see what AI can do," not "let's see if AI can do X measurably better than the current process."
- Scope expanded mid-pilot. What started as one workflow became four. Each one half-built. None defensible.
- No evaluation harness. Outputs are reviewed by feel, not by golden cases. Disagreement about quality is unresolvable.
- No clear governance owner. When a question comes up about a borderline output, no one knows whose call it is. Decisions don't get made.
- Clinicians weren't in the design room. They're invited to use the pilot, but they didn't shape it. Their feedback comes late and feels like rework.
The fix is unsexy but reliable. Before the pilot starts:
- Write the success criteria. One page. Sponsor signs.
- Lock the scope to one workflow. Resist heroically.
- Build a small evaluation harness. 30 cases. With a clinician.
- Name the governance owner. They get a fixed weekly slot.
- Put a clinician in the design room from week one. With a real seat, not as a reviewer.
This is the checklist I run on the first call of every Build engagement. If a pilot is currently stalled and you want a cold-eyed look, book a call — sometimes the problem is none of the above and it's worth knowing before you spend another quarter on the same approach.