Most healthcare AI engagements get stuck in the gap between the pilot demo and a real go-live. The pilot is technically successful and politically warm; nobody disagrees with anything; and three months later there’s still no signed scope for production. We’ve started running every engagement on the same 60-day rhythm to close that gap.
Days 1–15 — Discovery
We embed with the team. Two clinicians, one engineer, one operations lead — those are the four people we need in the room. We shadow workflows, audit the existing tech, and inventory three to five problems the model could realistically solve in the next six months.
- Day 5 — first written readout, with what we’re hearing back to the team verbatim.
- Day 10 — narrowed problem set, with the clinical lead’s explicit prioritization.
- Day 15 — agreed scope and integration plan, signed by clinical and IT.
Days 16–35 — Deployment
We stand up the first agent against a real workflow on a small population. Not a sandbox. Not a synthetic dataset. The actual tool against actual patients, with a clinical safety officer in the loop.
The deliverable at day 35 is not a slide deck. It’s a working integration with three named clinicians using it daily and a Slack channel where they’re reporting issues.
Days 36–60 — Measurement
This is the part most engagements skip. We instrument acceptance, modification, and override-with-reason from day one of deployment, so by day 60 we have four full weeks of clean data. The day-60 readout has three sections only:
- What’s working. The metrics, with the clinician quotes that triangulate them.
- What’s not. The override patterns, the workflow steps that broke, the integration edges we missed.
- The next 60 days. Either expand to the next site, retune for the current one, or sunset.
We’ve run this rhythm five times. Three engagements expanded. One retuned. One was sunset on day 60 with a clear conscience and no political damage. That last outcome is the one most consultancies don’t know how to deliver.