When this is the right conversation
You are preparing an internal assistant, retrieval-augmented generation application or automated workflow and need a defensible data foundation before deployment.
What the work involves
- Select one use case and define its users, decisions, information boundaries and unacceptable failure modes.
- Record source provenance, permitted uses, access groups and material restrictions with the relevant owners.
- Design permission-aware retrieval, refresh, removal and escalation requirements for the implementation team.
- Prepare representative evaluation scenarios and an evidence handoff to the people governing the AI system.
What we need from your team
An accountable use-case owner, technical lead and relevant data owners. Use rights, privacy and security questions may need specialist review.
What useful completion looks like
The pilot has a documented source boundary, access model, update/removal process and an explicit decision on unresolved gaps.
This service does not guarantee AI accuracy, model safety or regulatory compliance. AI system governance is a separate, connected workstream.
Before we begin
We agree the business objective, named delivery roles, access arrangements, fee, timetable and acceptance criteria in a written scope. Dependencies and out-of-scope work stay visible. We do not assume permission to change production systems or transfer sensitive information.
See the engagement processRelated field notes
AI-ready data is more than clean data
An internal policy assistant needs a source boundary.
A fictional pilot answers staff questions from approved policy documents. The governance work identifies the exact sources, their versions and owners, permitted uses and intended users. The team tests what happens when a source is superseded, a user loses access or a document is withdrawn. A clean document collection alone does not settle those questions.
Illustrative scenario, not a client case study or a claim of past results.
The working evidence
- Approved-source registerDocument or dataset, owner, version, purpose and unresolved rights.
- Access requirementsWho may retrieve which information, including negative test cases.
- Refresh and removal planResponsibilities for updates, corrections, indexes, caches and derived copies.
- Use-case handoffSource limitations, evaluation cases and decisions for wider AI governance.
Who does what
Source owners confirm the information and its intended use. The AI use-case owner sets the user and task boundary. Privacy, security and legal specialists review the issues within their remit. Engineers implement and test the agreed behaviour.
Use test accounts with different permissions. Rehearse a source replacement and withdrawal, and verify the affected retrieval paths in the agreed environment. Record failures as findings, not as a passed “AI-ready” score.
Data readiness is only one part of AI system review. It does not certify safety, accuracy, fairness, security or compliance, nor automatically authorise model training or commercial reuse.
Make the first workshop useful.
Bring roles, questions and approved examples—not an unrestricted data export.
- What precise use case, user group and information boundary define the pilot?
- Who can approve the source permissions and expected behaviour?
- How will restricted retrieval, source changes and removal requests be tested?