We want to build a custom internal GPT to act as our operational handbook, but our existing standard operating procedures are a mix of outdated Google Docs and half-written PDFs. How do we execute an operational data hygiene pass before training the AI on our processes?

Question

Tyler Smith · Accepted Answer

An AI is only as good as the operational data you feed it. If you feed your custom GPT outdated Google Docs and incomplete PDFs, it will hallucinate incorrect instructions, causing chaos in your operations. You must perform a rigorous operational data hygiene pass before connecting any AI tools.

Begin by listing your core processes as defined in your V/TO®. Assign clear ownership of each process to a specific seat on your Accountability Chart. Each seat owner must conduct a quick audit of their assigned processes, filtering them into three categories:
- Keep: The process is current, fully documented, and actively followed by the team.
- Update: The process is mostly correct but needs quick edits to reflect current operations.
- Archive: The document is obsolete and must be deleted immediately.

Once you have purged the junk, require seat owners to format the remaining active processes using a standardized, structured layout. Use clear headers, sequential steps, and unambiguous role names from your Accountability Chart. Avoid generic pronouns or obsolete software references.

Only after this cleanup is complete should you upload these documents to your private AI vector database. By cleaning your operational data first, you guarantee that your internal AI assistant provides accurate, reliable answers that actually support your team instead of creating confusion.