Making Company Knowledge Findable: Clean Up First, Then AI Search

"Where is the current version of the travel expense policy?" That question costs many companies hours every day. The knowledge exists, spread across drives, wikis, chats and inboxes. AI search promises to make it available with one question.

The actual problem

According to Bitkom, only 32 % of companiesU in Germany make use of their data potential. 48 %U use it a little, 13 %U not at all. AI search does not change that by itself. It finds what is there, faster. If five versions of the same policy sit in the system, it may well return the wrong one.

There is a second effect many underestimate. AI assistants such as Microsoft 365 Copilot see everything the person asking is technically allowed to see. If folders have been shared "with everyone" for years, salary lists or personnel files suddenly show up in answers. Not because the AI hacks anything, but because the permissions were never right.

In its 2026 Work Trend Index, Microsoft describes a company's own well-maintained knowledge as the next competitive advantage. The order follows naturally: clean up first, then search.

The workflow

Answer the question only from the documents found. For every statement, give the document title and date. If documents contradict each other, show both passages and the more recent date. If you find nothing, say so explicitly.

Limits and risks

• Oversharing: Put a permissions audit on the list before launch. Many vendors provide their own reports for this.
• Outdated content: Without archiving, the document with the most copies wins, not the right one.
• Sensitive areas: HR, health and contracts with personal data belong in separate areas with their own permissions.
• Ownership: Every knowledge area needs a person responsible for keeping it current.

How to start

1. Pick an area with many repeated questions, such as HR policies or IT how-tos.
2. Check permissions and archive everything older than two years that is no longer valid.
3. Launch AI search for that area and require source links in every answer.
4. After four weeks, count how many questions to the expert team have disappeared.

Sources

1. UBitkom: Digitalisation of the economy, almost every company is looking at AI, March 2026
2. UForbes / Moor Insights: Microsoft Work Trend Index 2026 Shows AI Productivity Is Not Enough, 19 May 2026