What Handover Looks Like for an AI System

What you receive

1. All source code, in a repository you own
2. The prompts, in configuration rather than buried in code
3. The evaluation set and the harness that runs it
4. A runbook — what breaks, how to tell, what to do
5. Credentials in your accounts, not ours

If any of those are missing, you cannot leave your supplier, which is not a position to be in.

The AI-specific part of handover

Conventional handover is about running the software. AI handover is also about judging the output — reading the evaluation results, knowing when a threshold should move, recognising drift in a correction rate.

We run a session on exactly that with whoever will own it, and leave written guidance on what the numbers mean.

The runbook contents

• Each alert, what it means and the first three things to check
• How to disable the AI step and fall back to manual
• How to widen the review band temporarily if quality drops
• Who at each provider to contact, with account references
• The cost cap, where it is set and how to change it

Support options

Retained suits systems where quality matters commercially. Ad hoc suits stable, low-stakes tools.

The first ninety days

Whatever the arrangement, we watch the first fortnight closely and stay available through the first quarter. Real usage reveals edge cases no test set contains.

That period is included in the build. It is not an upsell and it is the part that determines whether people trust the system.

Frequently asked questions