Six stages, each with a deliverable you can read
The order matters: constraints are settled before code, and quality is measured before rollout. Lengths are the usual range for one process with a cooperative data owner — the framing stage replaces them with a plan for your case.
01 · 1–2 weeks
Framing
One process, written down as it happens today: the volume, the exceptions, the people involved and the cost of a mistake. The stage ends with a decision on whether AI is the right tool at all — sometimes it is not.
What this stage produces
- A written scope with the process map
- A first list of the data involved
- A go / no-go recommendation
What we need from you
- The process owner
- Two or three real examples, end to end
- The list of systems involved
02 · 1–2 weeks
Data and constraints
Where the documents live, who may see what, what may leave the perimeter, which model providers are acceptable, what the auditors and the regulator will ask for.
What this stage produces
- A data inventory with classifications
- A hosting recommendation
- A data-flow description for review
What we need from you
- Access to a representative sample
- The security or compliance contact
- Any existing AI or data policy
03 · 3–6 weeks
Pilot
A narrow, real slice in production conditions: retrieval, prompting, an evaluation set from your own cases, and an interface inside the tool people already use.
What this stage produces
- A working slice on your data
- Measured quality on your own cases
- The first list of failure modes
What we need from you
- A named user group
- A test environment
- One person who can approve access
04 · 2–4 weeks
Evaluation and hardening
The failures found in the pilot become the regression suite. Permissions, logging, rate limits, refusal paths and hand-over to a person are tightened before anyone depends on it.
What this stage produces
- A regression suite that runs on every change
- Permission and logging review
- A written operations description
What we need from you
- Feedback from the user group
- A security review slot
- The hosting decision
05 · 2–6 weeks
Rollout
Wider release, training for the people using it, monitoring and alerting, and documentation for whoever runs it after us.
What this stage produces
- Rollout plan and training material
- Monitoring and alerting
- Runbooks for the operating team
What we need from you
- A rollout plan and dates
- A support contact
- Agreement on what is measured monthly
06 · ongoing
Support
A fixed monthly engagement: monitoring, re-evaluation after model or document changes, incident handling and small changes, with a named engineer.
What this stage produces
- Monthly quality report
- Re-indexing and re-evaluation
- Incident response
What we need from you
- The retainer decision