GlossaryKind of term: Manufacturing termAI pilot
What is an AI pilot?
An AI pilot is a bounded trial of AI on one real task, with real data and users and a measure of success chosen up front.
The pilot is temporary, and its users are the people who do the work every day. The task, the measure of success, the comparison with the current way of working and the end date are fixed beforehand. A pilot usually runs for weeks to a few months and ends with a decision: roll out, adjust or stop.
How does an AI pilot work in practice?
A pilot tells you something only when it contains the ordinary work, including the awkward requests with revisions, missing quantities and subcontracted finishing, not just hand-picked examples. Run the same requests the current way too, or there is nothing to compare. Useful measures include quote turnaround time, corrections per drawing or the gap between the proposed cost and your own estimate or, where it exists, the actual cost. Agree beforehand what happens if the measure is met: who will work with it, with which data and at what cost. Settle as well where the files are stored, who can reach them and whether they are used for training.
What is an AI pilot often confused with?
A proof of concept shows that something is technically possible; a pilot shows whether it holds up in daily work. With a demo, the difference is the material: a demo runs on the supplier's examples, a pilot on your own work. And a successful pilot is not yet a rollout; the decision to roll out follows, with the data, connections and agreements that go with it.
How do you work with Blake in early access?
Blake is available in early access, where you work with the STEP and PDF files of real requests from your own shop. Drawing values are set out as proposals to confirm or correct, every cost is labelled as an estimate, and a correction to an estimate line stays on record with who, when and why. That lets you lay the result beside your own estimate, line by line.
Frequently asked questions
How do you start an AI pilot in a manufacturing company?
With one task where time is lost and whose outcome you can measure, such as taking over drawing requirements or building up a cost. Use real requests, agree a measure and involve someone who does the work every day.
How do you measure whether an AI pilot worked?
With a measure from the work, such as quote turnaround time, corrections per drawing or the gap between the proposed cost and your own estimate or actual cost. Run the same requests the current way too, so there is a baseline to compare against.
What happens to my drawings during an AI pilot?
Agree it in writing beforehand: where the files are stored, who can reach them, whether they are used to train models and when they are deleted. Settle it before the first drawing is uploaded.
Why does an AI pilot sometimes stall after the trial?
One cause often cited is that the pilot ran beside the work rather than in it: different data, a different team, no agreed measure, no decision on next steps. A pilot on real requests with the people who do the work makes that decision possible.