AI readiness assessment for a data centre operator

The challenge

A UK data centre operator had Microsoft 365 Copilot in use across the business. People were using it for their own work, drafting emails, summarising documents and building presentations, and the leadership team wanted to move on to AI that runs whole business processes. Before that could happen they needed to know what Copilot would find in a tenant that had grown over several years, and what had to be put right first.

What we did

• AI Art of the Possible workshop. A two hour session with the leadership team covering where Copilot and AI agents are heading, live demonstrations on their own kind of work, and a review of the specialist AI tools they already paid for.
• Tenant assessment. We used Orchestry to score the Microsoft 365 tenant for Copilot readiness, covering Secure Score, permissions, sharing, guest access and workspace health.
• Copilot governance policy. Data protection, sensitivity labels, oversharing controls, agent governance and acceptable use, written for how the business actually works.
• Data governance policy. A new SharePoint structure by department, permissions, retention periods by type of record, file naming and content ownership.
• Remediation plan. Every action from the assessment, the two policies and the workshop in one tracker, with a priority, an effort estimate and a week to do it in.

What the assessment found

As with most tenants that have grown with the business, there was a fair bit for Copilot to trip over:

• 46 sites with broken permission inheritance, plus 296 folders with their own permissions and 141 with direct sharing links
• 120 inactive workspaces, and 91 Teams with no conversations in them
• 220 guest accounts, 198 of them with no current access
• a default sharing link set to Anyone, which on its own was the biggest single drag on the Copilot readiness score

The two urgent fixes, changing the default sharing link and tightening external sharing, take less than a day between them.

The first AI agent

The workshop agreed the first process to hand to an AI agent: the monthly management pack, which draws on 12 to 15 contributors and is put together by hand each month. The agent will chase contributors, draft slides from the previous month's figures, pull the inputs together and keep the format and tone consistent. The board pack is next in line.

What happens next

The remediation runs as an eight week programme. Urgent fixes come first, then the permissions, workspace and guest clean-up, then the SharePoint restructure, Orchestry provisioning and archival, retention labels and the Copilot controls. After that a monthly health check keeps the readiness score where it should be, with guest and sharing reviews at the same time.