Playbook · updated September 1, 2026

Escaping pilot purgatory

In shortPilots die from design, not technology: no dated success criteria, no plant-side owner, alerts outside the daily workflow, no model-health plan, and per-line economics that don't scale. The fixes are contractual and organizational — dated first-insight commitments, a named process-engineer owner, CMMS/QMS integration from week one, vendor-side MLOps with visible health indicators, and a pricing path to the full plant.

The five design choices

1 · Dated, money-shaped success criteria. 'Find €X of confirmed causes by date Y' survives budget season; 'explore the technology' doesn't. Write it into the pilot agreement.

2 · A named plant-side owner. A process engineer with hours allocated, not a steering committee. Tools owned by plant engineers survive year two; babysat tools don't — it's the strongest pattern in deployment data (45% of manufacturers cite internal expertise as their top barrier; ownership is how you beat it).

3 · Workflow wiring from week one. Alerts land where work happens — CMMS for maintenance, QMS/8D for quality — with a feedback field. A separate dashboard is where pilots go to be forgotten.

4 · Model health made visible. Drift monitoring, retrain policy, versioning — the vendor's job, surfaced as indicators you can see. Ask what happens when you introduce a new product; grade the concreteness of the answer.

5 · Scale economics agreed up front. Per-line pricing that's tolerable for one line and impossible for forty is a pilot-purgatory machine. Negotiate the plant-wide path before the pilot succeeds, while you still have leverage.

The meta-rule

Every one of these is settled before go-live. Pilots are won at design time — the algorithm was never the constraint.

Run the numbersPrice this loss with the What is one OEE point worth on your line? calculator, then take the ROI workbook into the budget meeting.