Your most experienced person is going to retire. What they know doesn't have to leave with them.

Somebody on the floor has a question nobody else can answer. Which customer needs the certs printed with the PO number on top. Why the long trailer never goes to the north dock. What to do when the old press throws error 14, which isn't in the manual because the manual is for a different revision of the press.

So they walk it over to Carol. She's been here nineteen years. She answers in about eight seconds without looking up from her screen.

Now picture Carol's retirement party. It's a good one. There's a sheet cake and a card everybody signed. And on Monday, the same question walks over to an empty desk.

Where the cost actually lands

It never shows up as a line item. It shows up as a quote that sits a day because nobody's sure how that customer's freight works. A new hire who's still asking basic questions eight months in. The same mistake made twice, because the fix lived in somebody's memory. And for a while, phone calls to Carol at home. She'll pick up, the first few times.

The bigger cost is what it does to growth. Every time volume climbs, more questions route through fewer people who know the answers. The year you land the big new customer is the year you find out how much of the operation was running through one person. And you can't hire a replacement for nineteen years of context. That person isn't on a job board at any wage. The next hire can learn the job. What they can't learn is everything that happened before they got there.

What the numbers say

The Bureau of Labor Statistics put out its latest tenure survey last month. In January 2026, the median worker aged 55 to 64 had been with their employer for 9.6 years, against 3.0 years for workers aged 25 to 34. Put another way, about half of workers aged 60 to 64 had been with their current employer at least 10 years, compared with about 1 in 5 workers aged 35 to 39.

That's a national number across every kind of employer, so your roster will tell you more than it does. But the shape is hard to argue with. The long memory in most companies sits with the people closest to the exit.

There's a better number too. In a ZipRecruiter survey of 1,500 people aged 60 and up earlier this year, 61% said they'd work a year or more in a reduced-hours wind-down role before fully retiring. Most of your Carols would give you a window if you asked. The question is what you'd do with it.

What fixing it actually looks like

It isn't the documentation push. Every company has tried that one. You ask your most experienced person to write down everything they know, on top of their actual job, and you get a binder that's half done and out of date by spring. It fails for a simple reason. People don't know what they know until somebody asks them the question.

What works is capturing knowledge from the work itself. A build like this usually has three parts:

1. Gather what already exists. Job folders, past quotes, the emails where she explained something to a customer, the SOPs, the cheat sheet taped to the side of her monitor. There's usually more written down than anyone thinks. It's just scattered.
2. Fill the gaps from real jobs. Short recorded sessions where she walks through last month's actual orders and someone keeps asking why. Thirty minutes a week beats a two-day knowledge-transfer marathon.
3. Make it answerable. A custom AI tool built on your own material, where someone types the question the way they'd ask Carol and gets an answer that points back to the source document, so they can check it.

The part that makes it hold up is timing. Carol reviews the answers while she's still around. Every wrong answer she corrects in her last six months is one nobody has to rediscover after she's gone.

I'd start with a map of where the questions actually go, then a working prototype on your own documents in about three weeks, then the full build. Fixed price, and you own the code.

When it isn't worth building

Some of this shouldn't be software, and it's better to know that up front.

• If the knowledge is mostly in someone's hands, like the welder who can hear a bad bead, a tool won't capture it. Pair them with an apprentice and protect the time.
• If one process has one clear successor, a few months of side-by-side work and a short binder will beat anything you could build.
• If your ERP or service software already has a notes or knowledge feature nobody uses, try that first. It's cheaper to fix a habit than to build around it.
• If almost nothing about your jobs gets written down anywhere, start there. There has to be something for a tool to stand on.

Try this before Friday

Write down the five people whose leaving would hurt the most. Don't guess at anyone's plans. People leave for all kinds of reasons, and this is worth doing for anyone on the list.

Next to each name, write the three questions only that person can answer. Then, for each question, ask where the answer would live if they gave notice tomorrow.

If most of the answers are "in their head," and those questions come up every week, you don't have a documentation problem. You have a capacity problem with a date on it that you don't know yet.

One next step

If that list worries you but you're not sure what to do about it, that's what the half-day session is for. We trace where the questions go and who answers them, and you leave with a written capacity map that says what's worth building, what you should buy off the shelf, and what to leave alone. Plenty of those maps say "don't build anything," and that's a fine answer too.