Brief sections
The leak
When a twenty-year service manager leaves, nothing breaks on day one. It breaks for a year in small ways: the customer whose invoices go to a different address in summer, the supplier who’ll expedite if you call before 8am, the machine that throws a false alarm when it rains — a thousand facts that were never secrets, just never written down. The costs are repeated mistakes, re-burned lessons, and relationships reset to zero. Unlike the other leaks, this one is invisible until the departure — and then it’s unfixable, because you can’t interview someone who’s gone.
What AI realistically does here
It captures knowledge as a by-product of normal work instead of as a documentation project nobody finishes. Calls, messages, job notes, and documents already contain most of the facts; the system extracts and organizes them into an answerable memory — ask “what do we know about this customer’s delivery constraints” and get the institutional answer with its sources, not a folder to dig through. Two honest boundaries. It can’t recover what already left, so the time to start is before the next departure, not after. And it captures facts, not judgment — knowing the supplier answers before 8am is capturable; knowing when to spend a favor with him is not. The goal is that facts survive turnover, not that experience stops mattering.
What has to be true first
- The work leaves traces in channels that can be captured — calls, email, job systems, messages. Knowledge exchanged purely over coffee needs a light process change first, like debriefs that get recorded.
- The team knows what’s captured and why. Agree on what is captured, who can see it, and how contributors can review or correct it.
- An owner for answer quality, because a memory that returns confident wrong answers is worse than a filing cabinet.
What we'd install
This is the leak where the system of artifacts matters most, because the value compounds across everything else: captured knowledge feeds the SOP library, sharpens every agent on the roster, and shortens every onboarding. It gets scored in the opportunity portfolio — usually paired with the tribal-SOP fix, since they’re two halves of one problem — and the value ledger tracks lookup outcomes: questions answered from the system versus questions that still require finding the veteran. Sequence on the AI operating partnership.
Worth it?
Estimate the hours per week your team spends re-answering questions the business has already answered once, and put your turnover next to your veterans’ tenure. The calculator below prices the recurring interruptions as an illustrative annual figure — the walk-out event itself is the real cost, and you can probably already name the departure that proved it. The scorecard tells you how exposed you are and whether the capture channels exist.