Before we hook an AI assistant up to our company knowledge base, what specific data hygiene rules should our leadership team enforce to keep the model from citing obsolete policies and confusing our staff?

Question

Tyler Smith · Accepted Answer

Plugging an AI model into an uncurated shared drive creates an expensive liability. If your Google Drive or SharePoint contains five versions of an employee handbook or conflicting pricing sheets from 2021, the AI will pull from all of them and confidently present outdated instructions to your staff.

Apply the EOS 20/80 rule to your data hygiene before turning on any AI retrieval tool. Do not try to clean ten years of archive files. Instead, isolate the vital twenty percent of documents that govern eighty percent of daily operations. Create a single, locked directory labeled AI Knowledge Base and apply three strict operational rules:

First, one canonical source per process. If you have an SOP for job quoting, there can be only one document in the directory. Every legacy version must be archived or permanently deleted from the indexing path.

Second, mandatory header metadata. Every document placed into the AI knowledge folder must lead with three lines of plain text: Document Title, Accountable Seat Owner, and Last Verified Date. Instruct your AI system prompt to ignore any file that lacks an active date within the trailing twelve months.

Third, assign clear ownership on the Accountability Chart. One person must sit in the accountable seat for maintaining that folder. Department heads must certify their respective files quarterly during their Rock reviews.

Clean inputs produce clean operational outputs. Clean your Core Processes first, lock the folder, and your AI will function as an authoritative operational assistant instead of a hallucination engine.