Constraint test · evidence partial
A five-person consultancy tests a controlled internal AI pilot
A worked scenario showing how partial evidence becomes a reversible decision rather than an automatic purchase.
The team
Five people prepare meeting agendas, draft outlines, and internal research summaries for client engagements.
- partial
This team is a scenario, not a claim about a measured population.
MS-LEARN-COPILOT-PRIVACY · Data, Privacy, and Security for Microsoft Copilot · external-ms-learn-privacy
The pressure
Preparation time is increasing, but the team cannot trade client trust for speed.
- partial
Sector adoption growth explains relevance, not the correct decision.
UK-DSIT-AI-PBS-2026 · AI Adoption Plan for Professional and Business Services · external-uk-dsit-ai-adoption
The hard constraint
No client information enters the pilot until contract and privacy authority are explicit.
- partial
Product documentation cannot supply contract-specific authority.
MS-LEARN-COPILOT-CONTROLS · Security and data controls for small business · external-ms-learn-controls
The baseline
The team records current preparation time and correction effort for four weeks.
- partial
No reviewed source supplies this team-specific baseline.
MS-LEARN-COPILOT-PRIVACY · Data, Privacy, and Security for Microsoft Copilot · external-ms-learn-privacy
The pilot
Two users test one outline workflow with public or synthetic inputs for a fixed period.
- partial
A narrow pilot is a proposed risk-control response, not an evidence-backed guarantee of benefit.
UK-DSIT-AI-PBS-2026 · AI Adoption Plan for Professional and Business Services · external-uk-dsit-ai-adoption
The controls
Named accounts, restricted sharing, least privilege, logging, retention review, and administrator recovery are tested.
- partial
Vendor documentation describes relevant controls but local configuration remains unproven until tested.
MS-LEARN-COPILOT-CONTROLS · Security and data controls for small business · external-ms-learn-controls
The reviewer
One named person verifies every output against the source material before it is reused.
- partial
The evidence reviewed does not justify removing human accountability.
UK-DSIT-AI-PBS-2026 · AI Adoption Plan for Professional and Business Services · external-uk-dsit-ai-adoption
The measurement
The team records time, corrections, rejected drafts, boundary breaches, and reviewer confidence.
- partial
Outcome measurement is needed because public adoption data does not prove local value.
MS-LEARN-COPILOT-PRIVACY · Data, Privacy, and Security for Microsoft Copilot · external-ms-learn-privacy
A failed gate
If access, logging, or data restrictions cannot be demonstrated, the pilot stops before live information is introduced.
- partial
Described controls are not equivalent to verified controls.
MS-LEARN-COPILOT-CONTROLS · Security and data controls for small business · external-ms-learn-controls
The trade-off
Even useful drafts may create more review work or new governance overhead than the manual process.
- partial
No reviewed source establishes net benefit for this scenario.
MS-LEARN-COPILOT-PRIVACY · Data, Privacy, and Security for Microsoft Copilot · external-ms-learn-privacy
The conditional outcome
The team may continue the internal pilot only if controls pass and measured benefit exceeds review burden.
- partial
Client-data use remains separately blocked pending explicit authority.
MS-LEARN-COPILOT-CONTROLS · Security and data controls for small business · external-ms-learn-controls
The next action
Document the workflow and prohibited data first; do not buy, expand, publish, or deploy from partial evidence alone.
- partial
The public evidence supports investigation, not automatic commercial action.
UK-DSIT-AI-PBS-2026 · AI Adoption Plan for Professional and Business Services · external-uk-dsit-ai-adoption