1. Purpose and scope
This stage does two things that must be finished before anyone runs a candidate tool on company design data. It narrows the field of possible tools to a shortlist worth piloting, and it fixes how the pilot will be judged. The second is the more important. AI-assisted outputs can look convincing when they are wrong, and a pilot judged after the results are known tends to be judged on what the tool happened to do well. Writing the reference designs, the known answers and the success criteria down first is what makes the pilot an experiment rather than a demonstration.
The stage also clears the conditions for a safe pilot. Information security approves exactly which data may be used and where it may go. Legal agrees evaluation terms with each supplier, including confidentiality and a commitment not to train on pilot data unless that is explicitly agreed. The sponsor commits budget and people. Machine learning and data engineering prepares an isolated environment that cannot reach production design data.
In scope: supplier and category screening, the build or buy question, the hosting question at a first level, the reference design set, the success criteria, the evaluation plan, the pilot data handling approval, the evaluation agreement, and the pilot budget and environment. Out of scope: running the pilot, formal qualification, and the production data, IP and licence review, which come later and go deeper.
12 more sections for registered members.