A partner at a firm asks the firm’s AI assistant to pull together everything the firm knows about a prospective client. The assistant checks whether the partner has access to the relevant files, finds that they do, and returns an answer built in part from an engagement the partner should never have seen, because a personal holding disclosed months ago should have kept them off it entirely. Nobody bypassed a permission. No file was shared that shouldn’t have been. Microsoft Purview did exactly what it was built to do. What was missing was the layer built to catch what Purview was never designed to check.
It’s not a one-off — it’s the pattern compliance leaders describe the moment a firm moves AI past a pilot and into real engagement work.
What Microsoft Purview does well
Microsoft built Purview to answer a specific, enterprise-wide question: is this content protected, labeled, and retained the way policy says it should be. “Purview was always designed to prevent you gaining access to information,” says Richard Bowes, Intapp’s Senior Compliance Growth Director. It does that job well, and firms are right to rely on it for that job.
What Purview isn’t built to know is engagement-specific: whether a given person has a live conflict, an independence restriction, or a personal relationship that should override access they’ve technically been granted. That’s not a gap in Purview. It’s a different question, and one that sits with a firm’s compliance systems rather than its content platform.
Where Microsoft and Intapp work together for firm compliance
For a document sitting quietly in SharePoint, that difference rarely surfaces. For AI reasoning across content in real time, it surfaces any time the answer draws on an engagement the person shouldn’t see. The two layers are answering different questions entirely:
| Microsoft Purview | Intapp Walls and Collaboration | |
| What it answers | Is this content protected, labeled, and retained correctly across the firm? | Should this specific person have access to this specific engagement right now? |
| Where it operates | Firm-wide, at the tenant and label level | Engagement by engagement, updated as teams change |
| What it catches | Unauthorized access to classified or labeled content | Conflicts, independence restrictions, and personal holdings that override a technical permission |
| How it updates | Applied at the point content is labeled | Continuously, as engagement teams change |
That difference is what shows up in the scenario at the top of this post: an AI assistant that was technically permitted and still wrong. Intapp Collaboration keeps engagement data structured and current, so there’s a governed body of information for policy to apply to in the first place. Intapp Walls checks every access request against a firm’s actual compliance position, rather than at the point someone was first added to a folder. Paired with Purview’s enterprise controls, that gives a firm both halves of the picture: content protected the way Microsoft 365 already protects it, and access governed the way an accounting engagement actually needs it to be.
A pilot won’t expose the gap either. Pilots run on a handful of test queries against a narrow slice of content, so the access patterns that would reveal a live conflict don’t come up until AI is running at scale, against real engagement data. That is the point at which a firm can least afford to find out.
Trust is what makes the pairing worth building
“Trust is really what determines whether or not AI can scale,” says Brant Hollenkamp, Microsoft’s Professional Services Industry Leader. “If leaders can’t see what AI is doing, they can’t govern it and they can’t measure the outcomes. If this is the case, adoption will stall.” A firm that can demonstrate, engagement by engagement, that AI only ever surfaced what a given person was actually cleared to see is a firm that can expand AI use with confidence, on the Microsoft 365 investment it has already made. It’s also why several top accounting firms already run Intapp alongside that same Microsoft 365 environment.
That confidence isn’t a policy. It’s an operational condition that has to hold every day.
Missed part one? Read Compliant collaboration: a maturity model for AI-ready professional firms in our Compliant Collaboration Series. Then see how professional firms can build the foundation needed to govern AI in our engagement context white paper.
