Three months ago we lost our head of operations. Twelve years at the company, knew everything, and had spent the last eighteen months becoming the person who knew how to get the most out of every AI tool we had deployed.
The transition was painful in ways I expected. It was also painful in ways I did not.
The ways I expected: institutional knowledge walking out the door, process documentation that was out of date, a team that had depended on her judgment for things they now had to figure out themselves.
The ways I did not expect: she had built the prompting logic that made our internal AI assistant actually useful for operations queries. She had trained the document classification system through months of feedback. She had created the folder structure and tagging conventions that made retrieval work. None of that was documented because it had accumulated gradually and nobody had thought to write it down because it lived in her habits, not in any formal system.
When she left, the AI tools did not break. They just got worse in subtle ways that took us weeks to fully diagnose. Queries that used to return precise operational data started returning vague or outdated results. The assistant started giving answers that were technically correct but not calibrated to how we actually worked. The people using the tools started trusting them less without fully understanding why.
What I learned from this is that AI tool value in an organization is not just a function of the tool. It is a function of the accumulated configuration, prompting knowledge, and data hygiene work that specific people have invested in that tool. When those people leave, some of that investment leaves with them unless you have treated it as organizational infrastructure rather than individual knowledge.
We have since built a simple practice around this. Anyone who is a significant user of an AI tool documents, quarterly, what they have figured out about using it well. What queries work reliably, what queries need to be structured a certain way, what the tool does poorly and how they work around it. This documentation lives in a shared space and gets reviewed when someone transitions out of a role.
It is not a perfect solution. But the first time we used it during an offboarding, it cut the transition friction for the next person significantly. The new head of operations did not have to spend three months rediscovering what her predecessor had already figured out.
The investment in AI tools is not just the license cost and the setup time. It is also the knowledge that builds up around those tools over months of real use. Treating that knowledge as organizational property rather than individual knowledge is a small operational decision with compounding returns.
Top comments (0)