There’s a pattern I see constantly in my consulting work. An organisation gets excited about Microsoft 365 Copilot, deploys it across the business, and then wonders why the results are underwhelming.
The answer, almost every time, is the same: the information architecture wasn’t ready.
What information architecture actually means
Information architecture isn’t about folder structures or naming conventions — though those matter. It’s about how your organisation’s knowledge is structured, connected, and made discoverable.
In a SharePoint context, this means:
- Site structure — How sites, hubs, and libraries relate to each other
- Metadata and content types — How documents are classified and tagged
- Navigation and findability — How people (and AI) locate what they need
- Permissions architecture — Who can see what, and why
When any of these are poorly designed, humans struggle to find information. And if humans struggle, Copilot doesn’t stand a chance.
Why AI makes this urgent
Before Copilot, poor information architecture was an inconvenience. People would spend extra minutes searching, or they’d rely on tribal knowledge — knowing that “Sarah keeps the budget files in her team’s channel.”
With Copilot, the stakes are higher. The AI is searching across your entire Microsoft 365 tenant. It’s pulling from SharePoint, OneDrive, Teams, and Exchange. If your information is scattered across subsites, duplicated in personal drives, or locked behind inconsistent permissions, Copilot will either:
- Return irrelevant results — because it found the wrong version of a document
- Miss critical information — because it couldn’t access or discover it
- Hallucinate — because the source material was contradictory or outdated
None of these outcomes build trust.
The three foundations
Based on my work with dozens of organisations, there are three things you need to get right before Copilot delivers real value:
1. Flatten your site hierarchy
Subsites were a reasonable choice in 2015. In 2026, they actively work against you. Modern SharePoint — and Copilot — works best with a flat hub-and-spoke model where each site is a top-level site collection connected via hub associations.
If you’re still running deep subsite hierarchies, this is the single highest-impact change you can make.
2. Invest in metadata
Column-based metadata beats folder structures every time. When documents have proper content types and metadata columns, Copilot can filter and reason about your content much more effectively.
This doesn’t mean you need to retag every document in your tenant. Start with your highest-value document libraries — the ones your teams actually use daily — and build from there.
3. Clean up permissions
Oversharing is the silent killer of Copilot trust. If someone asks Copilot about Q4 revenue and it surfaces a draft document that was meant for the executive team, you’ve got a problem.
Run a permissions audit. Use sensitivity labels. Establish clear ownership for every site.
Getting started
You don’t need to solve everything at once. The most effective approach I’ve seen is:
- Audit your current state — Understand what you have and where the problems are
- Pick your highest-value area — Usually the department or team that would benefit most from Copilot
- Fix the foundations there first — Flatten sites, add metadata, clean permissions
- Expand outward — Use that success story to drive change across the organisation
The organisations that get the most from AI are the ones that did the unglamorous work of structuring their information first. It’s not exciting. It doesn’t make for great demos. But it’s the difference between an AI that sounds impressive and one that’s actually useful.
If you’re planning a Copilot deployment and want to make sure your SharePoint is ready, get in touch. I work with organisations to build the information architecture that makes AI actually work.