What Happens When Your Best Operations Employee Leaves?
Your most reliable process probably lives in someone's head. How tribal knowledge becomes operational risk, and how to turn it into an asset you own.
Every operations leader has one: the person who just knows. The reviewer who can glance at a document packet and spot the problem in seconds. The team lead everyone routes the weird cases to. The one whose two weeks of vacation make the whole queue slow down.
That person is an asset. They’re also one of the largest unpriced risks on your operation, because when they leave, the process leaves with them.
The process you run is not the process you documented
Most operational processes exist in two versions. There’s the documented version: the SOP, the checklist, the training deck. And there’s the real version: which document variants actually show up, which discrepancies are deal-breakers versus formalities, what to do when two systems disagree, which edge cases are safe to wave through and which ones burn you.
The documented version might cover 60% of what actually happens. The rest is rules, experience, and accumulated judgment. It lives in people’s heads.
We saw this concretely at a leading automotive marketplace processing around 1,000 vehicle-purchase evaluations per week, each involving roughly five documents. The review process was nominally documented. But the operational reality (the state-by-state rules, the fraud patterns, the judgment calls) lived largely in the heads of the Deal Support team. When a senior reviewer left, their replacement didn’t inherit that knowledge. They rebuilt it, slowly, by making the same mistakes their predecessor had already learned from years earlier.
What tribal knowledge actually costs
The cost doesn’t show up as a line item, which is why it rarely gets managed. It shows up as:
Training time. Every new hire in a judgment-heavy review role takes months to reach full productivity, and reaches it by absorbing knowledge from the people around them, one escalation at a time.
Error regression. When an experienced reviewer leaves, error rates on their queue go up before they come back down. In document-heavy operations we’ve measured, teams running mostly on experience sat near a 7% error rate. Every departure reset part of the learning curve.
Key-person bottlenecks. The best people become escalation points for everyone else, which caps throughput on exactly the cases that matter most.
A hiring treadmill. If the process only exists in people, scaling the process means scaling the people, with every new head carrying the same ramp-up cost and the same eventual departure risk.
The alternative: make the knowledge executable
There’s a different way to treat this problem, and it starts with a discovery process rather than a technology decision. The goal is to take what your best operators know and make it explicit enough to execute, whether a human or an AI is doing the executing.
In practice, that discovery looks like this. You sit with the people who run the process and walk through real transactions: not the SOP, real cases. You catalog the document types that actually arrive, including the messy variants. You extract the verification checks they perform, the rules they apply, and the order they apply them in. And you pay special attention to the exceptions: the cases where they stop and think, because those boundaries between “routine” and “needs judgment” are the most valuable knowledge in the building and the least likely to be written down anywhere.
What comes out the other side is the process, externalized: document types, extraction fields, verification logic, business rules, decision criteria, and a defined exception path back to humans. At that point it can be operated by AI: documents in, structured data extracted and verified, rules applied, decisions out, exceptions escalated with context.
At the automotive marketplace, that’s exactly what happened. The captured knowledge became a production decision system. Around 50% of deals now clear automatically in under two minutes, about 70% of volume is AI-managed, and the error rate fell from roughly 7% to roughly 1%. That’s better than the experienced team it learned from, because executable rules don’t have off days and don’t forget the edge case from eighteen months ago.
The strategic shift
The headline benefit is efficiency, but the deeper one is ownership. Before, the company’s evaluation capability was a property of specific employees. After, it’s a property of the company: versioned, auditable, improvable, and immune to resignation letters. The remaining human team (six people, down from twelve, with the rest redeployed to growth work) handles genuine judgment calls, and everything they learn from new edge cases gets folded back into the rules instead of accumulating in one more head.
So the question worth asking isn’t “should we automate?” It’s simpler and more uncomfortable: if your three most experienced operations people resigned this quarter, how much of your process would still exist?
If the honest answer is “not enough,” the discovery process above is worth running even before any automation decision. And if you want to see where it can lead, the anonymized numbers are in our AI Deal Engine case study.