How it works
Assess the question. Build only what is needed. Run it as a managed service.
Each stage has a clear output and a decision before the next commitment. An assessment can recommend an existing tool, BI, a managed connection, prerequisite work, or stopping.
Assess, Build, Run
Three stages. Each ends with a review before the next begins.
Production work begins only when the assessment supports it. Launch follows agreed answer, access, and operating checks.
01 Assess
02 Build
03 Run
The customer approves the business purpose, users, data, provider settings, and acceptance decision at the relevant stage.
Assess
Choose the simplest suitable option before building.
The assessment starts from the business question and compares the tools and access paths already available.
Inputs
One recurring question, its owner, likely source data, the intended AI assistant, and a trusted comparison result.
Review
Business definitions, source and access options, assistant compatibility, trusted comparisons, operating ownership, and exit.
Output
A written recommendation, source and access review, evaluation plan, and scope-based fee model.
Decision
Use an existing tool, use BI, build a managed connection, fix a prerequisite, or stop.
Acceptance before launch
Check answers, access, failures, and operating readiness.
The exact acceptance set follows the approved use case and is refined during the build.
- Representative answers are consistent with agreed reports or known results.
- Unauthorised users, operations, fields, and sources are rejected as designed.
- Unavailable data and unsupported questions fail visibly.
- Monitoring, support, incident contacts, responsibilities, and known limits are documented.
Answers
Reviewed against known results
Access
Approved and rejected paths tested
Operations
Monitoring and failure paths tested
Decision
Accept, revise, or stop
Acceptance applies to the agreed question, systems, users, assistant, and environment.
Run
A managed service includes maintenance, change, and exit.
Source systems, authentication, definitions, AI assistants, and provider policies can change after launch.
01
Monitor and support
Observe availability, authentication, tool outcomes, and relevant failures; investigate agreed service incidents.
02
Recheck answers
Run representative questions after material changes or on an agreed review schedule.
03
Manage change
Assess, test, approve, and document changes to sources, definitions, tools, access, or assistants.
04
Hand over or close
Document dependencies, revoke access, remove services, and retain or delete agreed records when the service ends.
- A new use case needs its own data and answer review
- Write actions require separate approval and controls
- A new assistant or source may change the data flow
First step
Start with the question, the owner, the data, and a known result.
That is enough to assess the available options.