The Problem
An enterprise client had a SharePoint full of policies, contracts, and SOPs that nobody could find anything in. Some docs were two years out of date. Critical SOPs were buried four folders deep. The fastest path to an answer for most employees was interrupting a supervisor and hoping they remembered correctly.
It's the kind of problem that doesn't show up on a P&L. But it drains productivity every day, and it creates compliance risk every time someone makes a decision based on what they think they remember from training.
What We Diagnosed
The chatbot wasn't the first move. We mapped what they actually had and found that ~40% of their internal policies were either out of date, contradictory, or genuinely missing. Pointing an AI at that document set would have produced confidently wrong answers at scale.
So we did the unglamorous work first: restructured policies, contracts, and SOPs into clean, current documentation with a clear owner per doc. Only then did we build the bot.
What We Built
- A RAG-powered knowledge bot that indexes the cleaned-up document set and retrieves answers sourced from official docs
- Source citations on every answer, so employees can verify
- Feedback loops that flag bad answers and missing knowledge — feeding back into the doc owners' queue
- Usage analytics for the policy team, so they can see which docs get asked about (and which are silent)
All of it monitored, with alerts when retrieval quality drops.
The Results
| Metric | Before | After |
|---|---|---|
| Time spent searching for policy info | Baseline | -40% |
| Errors from outdated/wrong info | Regular | Near zero |
| Internal documentation health | Stale, partial | Clean, owned, current |
| Compliance risk surface | High | Materially reduced |
The byproduct nobody expected: the documentation cleanup was more valuable than the bot itself. The bot just made the cleanup load-bearing.