tyler-smith.com · Questions & Answers

Our company shared drive is cluttered with duplicate spreadsheets, outdated SOPs, and conflicting client folders. What exact data hygiene steps must we take before connecting an internal AI assistant or search tool to our company files?

Connecting an AI assistant or retrieval-augmented generation system to an uncurated shared drive creates an expensive hallucination engine. The model cannot discern whether an SOP written in 2019 supersedes a revision saved in 2024. If your source data is contradictory, your AI output will be actively dangerous to operations.

Begin with quarantine rather than cleanup. Do not task your entire team with sorting through terabytes of legacy files. Instead, create a brand-new, read-only directory designated as the Single Source of Truth. Populate this directory with only your active, leadership-approved core processes, active client contract templates, and validated operational manuals.

Apply strict metadata and naming conventions to every document in that repository. Every file must include an explicit effective date, the name of the Accountability Chart seat that owns the process, and an explicit version number. Deprecate and archive all working drafts, raw meeting notes, and superseded spreadsheets to an offline archive folder that the AI crawler cannot index.

Next, convert unstructured documents into clear markdown or plain text files with consistent headers before indexing. AI models digest structured outlines, clear bullet points, and defined schemas far more accurately than complex multi-tab spreadsheets or PDFs with nested tables. Keep the ingestion footprint small. Feed the model only the top twenty percent of documented assets that drive eighty percent of daily operational execution.

Category: AI-Powered Operations

← All questions