Five questions. Five short answers.

We lock the metric in week two of the audit and start the 90-day build against it on day 1. Mid-market clients see an early signal by week six — the build ships, and the metric you picked moves within the week. By day 31 we hand the system to your operators, and if the metric has not moved the invoice does not go out.

Every mid-market operator’s metric is different, so the quote is written against the metric you pick at target selection — claims cycle time, contracts drafted per week, hours reclaimed per close — not against a published band. The three engagement tiers cover most builds, but the quote binds against the metric, not the tier. The number is on the quote you sign before any work starts.

Your data, your environment, your credentials — at all times. Nothing of yours lives in a vendor account we control, and nothing we build on top of it depends on us being there. The handoff document is the same template whether you keep us on AI Ops or walk on day 31, and the system works without us at the end.

Operator-on-the-loop guards, retrieval scoped to your own context, and an audit log we ship with the metric dashboard — so every action the system takes is traceable to an operator on your side. We post one weekly note on what was retrained, what was flagged, and why. No black-box third-party calls, no vendor-controlled backend, no model that survives a contract review only in our hands.

The handoff document is written for the operators who pick it up Monday morning — not for the steering committee that approved the project. Weekly notes go to the people who act on them, and the first month is on-call so the failure cases do not become a Tuesday-morning adult onset. AI Ops is the optional hour-by-hour drift watch after that; most operators do not need it, and nobody is forced onto it.