From pilot to production: 21 checks
What sits between a working demo and a system you dare to let run
Free · checklist · 16 pagesOne hour per feature
Within an hour you establish, for every AI feature on your roadmap, what still stands between it and production, and which three points block go-live.
The document is in Dutch.
Why you won't find this anywhere else
The MLOps checklists doing the rounds are written for teams with a platform department. This one is written for a team of five that has to get an AI feature into existing software, with a security officer looking over their shoulder and a budget that doesn't move with token usage. Every check comes with the failure mode that produced it, and three checks are in there because we learned them too late ourselves.
What's inside
- Six checks on behaviour: evaluation set, threshold, behaviour at low confidence, version control and regression after a model update
- Five checks on boundaries, each linked to what a security officer or a DPIA will ask
- Four checks on cost and load, with the rule of thumb for estimating in advance and where it always comes out too low
- Six checks on visibility and intervention, including the two most often missing and most expensive to miss
- A score sheet with three outcomes, and the three checks that block go-live whatever the score
Who it's for
- CTOs, IT directors and Heads of Digital at organisations of fifty people or more
- Teams that need to get an AI feature into existing software, not next to it
- Anyone with a pilot that demos well and doesn't go live
Not for anyone still working out what AI is. This starts with a team that is already building.
What this document doesn't solve
The list tells you whether it is ready. It says nothing about whether it is the right feature, how it fits your landscape, or what running it costs per year.
That's what the AI Opportunity Audit is for: we map your processes, show where AI and automation deliver the most, and set out what to do first.
See the AI Opportunity Audit