When Your Best Person Leaves, What Leaves With ThemWhen Your Best Person Leaves, What Leaves With Them

A friend of mine who owns a distribution company had an operations employee with nineteen years under her belt. She knew, without checking anything, which customers would accept a partial shipment and which ones would treat it as a broken promise. She knew that one account's receiving dock closed at two on Fridays. She knew that a particular buyer always asked for a discount he did not actually expect to get, and that giving it to him early cost margin for no reason.

None of that was in the ERP. None of it was in a policy document. It was in her.

She retired in March. Her replacement is capable, has the same systems and dashboards, and has spent six months rediscovering things the company already knew. Two accounts got handled badly enough to notice. Nobody did anything wrong. The knowledge simply was not anywhere it could be handed over.

That is the most expensive thing happening in most businesses right now, and almost nobody has it on a budget line.

Nobody Is Staying Long Enough for the Old Model to Work

The way companies have always handled this is time. You put the new person next to the experienced one, and over a couple of years the knowledge transfers by osmosis. That model quietly stopped working, because the couple of years is gone.

The Bureau of Labor Statistics reports that median employee tenure is 3.9 years, the lowest reading since 2002, and its most recent JOLTS release puts quits at 3.1 million a month. At that turnover, an apprenticeship model of knowledge transfer is being asked to run on a fraction of the time it was designed for.

Meanwhile the daily cost of not having the knowledge is enormous and invisible. McKinsey found that workers spend nearly 20% of the workweek looking for internal information or tracking down the colleague who has it, which is a full day a week spent on retrieval rather than work. A Panopto and YouGov study put a number on the leakage: 42% of institutional knowledge is unique to the individual who holds it, and the average large US business loses $47 million a year in productivity to inefficient knowledge sharing. Onboarding does not close the gap quickly either, since SHRM reports that nearly a third of new hires leave within six months.

Read those together. Nearly half of what your company knows sits in one head each, the heads turn over every four years, and everyone else spends a day a week hunting for what those heads know.

It Was Never Written Down Because It Was Never a Document

The instinct here is to demand documentation. Write it down. Build a wiki. Make everyone contribute before they leave.

I have watched that fail enough times to stop recommending it, and the reason is structural rather than a matter of discipline. What that operations manager knew was not a document. It was a set of judgments distributed across nineteen years of specific situations. Nobody can sit down and write out which customers accept partial shipments, because the answer is not a list. It is a pattern formed from a few hundred instances, most of which she could not consciously recall.

Exit interviews cannot extract that, and neither can transfer meetings. The knowledge was created as a byproduct of doing the work, and the only reliable place to capture it is inside the work itself.

Your AI Has the Same Blind Spot Your New Hire Does

Here is where this connects to the thing everyone is spending money on.

Most companies deploying AI have given it exactly what they gave the replacement hire. Access to the systems, access to the reports, access to the record of what happened. And nothing at all about why any of it happened, which accounts are exceptions, or which of last year's decisions were later regretted.

A model reading your ERP sees that account 4471 received three partial shipments. It cannot see that two were fine and the third nearly cost you the relationship. It will be as confidently wrong as a competent new hire in week two, for the same reason and with less hesitation.

This is the operational version of the argument I made in Context Is the Whole Game. Data is what your systems recorded. Context is what your people know about what your systems recorded. Almost every disappointing AI deployment I have looked at had plenty of the first, none of the second, and blamed the model.

It is also why the assembly work I wrote about in Stop Automating Decisions keeps needing a human in the loop. The person is not there to gather the records. The system can do that. The person is there because they are the only place the interpretation lives.

Capture Has to Be a Byproduct

The version of this that works does not ask anyone to write documentation. It captures reasoning at the point of decision, because that is the only moment the reasoning exists in a form anyone can state.

Practically, that is a few small habits. When someone overrides what the data suggested, the override gets a one line reason attached to it, in the system, at the time. When a customer is handled as an exception, the exception is recorded against the customer rather than remembered by a person. When a bid is lost, the reason goes next to the bid within a week, while it is still true and not yet reshaped by hindsight.

None of that is a project. It is a habit with a place to put things, and it compounds in a way documentation never does, because it accumulates daily rather than in a panic during somebody's notice period. Two years of recorded overrides is a real map of how your business makes decisions, which is exactly what a new person, or a system, needs to become useful quickly. It is the same point I made in What Is Operational Intelligence: the problem was never a shortage of data, but the absence of anywhere for meaning to collect.

You can test whether you have such a place this week. Pick the person whose departure would hurt most, and you already know who that is. Ask them to walk you through one recurring decision, and every time they say "well, it depends" or "with that customer we usually," write the sentence down. Then ask where in your systems that sentence could live so the next person finds it without asking anyone. Usually there is no such place, and that is the actual finding.

The Takeaway

Turnover is not primarily a recruiting cost. It is a context cost, and it is the one nobody measures.

With tenure at 3.9 years and nearly half of what your company knows sitting in individual heads, the knowledge that makes your business specifically good at what it does is leaving on a four year cycle and getting rebuilt from scratch each time. Documentation projects do not fix that, because the knowledge is a byproduct of the work and has to be captured there.

The companies that solve this will not be the ones with the best retention. They will be the ones where the context stays when the person goes, which is also the only condition under which their AI will ever be worth what they paid for it.

Your best person is going to leave eventually. The question is whether what they know leaves with them.

Tracy Thayne is the founder of Expona, an AI consulting practice for smaller B2B companies in construction, manufacturing and the trades around them. Take the free Decision Line Assessment or subscribe to the blog (below) for plain spoken posts on putting AI to work in businesses that don't call themselves tech.