Why information architecture comes before any AI tool

If you run a growing company on Microsoft 365, fix how your files are organized before you buy an AI assistant. The assistant will answer from whatever it can find, and it will not tell you which version is right.

What an assistant actually does

Tools like Microsoft 365 Copilot do not know your business. They search the sites, libraries, chats, and mailboxes a person can reach, pull out what looks relevant, and write an answer from it.

That works well when the source material is clean. There is one current pricing policy, one approved safety procedure, and one home for each client’s contract.

It works badly when the source material is a pile. If three folders each hold a “current” pricing policy, the assistant may quote any of them, in confident and well-written sentences.

How a tenant gets messy

Nobody sets out to make a mess. It builds up one reasonable decision at a time.

• A project gets its own site, and the site outlives the project.
• A migration copies everything across “just in case.”
• A folder called Final sits next to Final v2 and Final v2 REAL.
• People leave, and the sites they owned belong to no one.
• Every team names things its own way.

Search used to hide most of this. People learned which folder to trust. An assistant has no such instinct. It reads all of it.

Three symptoms to look for

1. Several versions of the same document. Ask three colleagues where the current version of a policy lives. If you get three answers, you have this.
2. Sites nobody has touched in years. Old content is not harmless. It is still searchable, and it is often wrong.
3. No named owner. If nobody is responsible for a site, nobody removes what is stale.

A test you can run this week

Pick three documents that everyone should be able to find: a current policy, a price list, and a procedure. Ask five people, from different teams, to find each one and time them.

A new hire should find the right file in about a minute. If it takes much longer, or people land on different files, an assistant will struggle in the same way.

What good looks like

You do not need a perfect structure. You need a few plain rules that people actually follow.

• One home for each type of document. Policies live in one place. Contracts live in another.
• A named owner for every site. One person, not a team alias.
• Naming and labels people can explain. If a new hire cannot guess how a file is named, the rule is too clever.
• Stale content retired or archived. Remove it from search, not just from view.

The order to fix things in

Start small and work in this order.

1. Take inventory. List your sites and libraries, who owns each, and when each was last used.
2. Decide the homes. For your ten most important document types, name the one right place.
3. Move and redirect. Put the current version in its home. Leave a pointer where people used to look.
4. Retire the rest. Archive what nobody has used in two years, with the owner’s agreement.

Do the priority libraries first. A tidy set of a hundred important files beats a half-finished cleanup of a hundred thousand.

Where this shows up in the assessment

In our AI Readiness Assessment this is sub-score 1.2, information architecture and findability. It sits in the first pillar, Information & Data Foundation, because everything else depends on it.

A low grade here does not mean you cannot use AI. It means the first project should be the structure, not the tool. If you want to see how your organization would grade, read how the assessment works.