Bespoke Work Is a Bug Report
Somebody builds something for one customer. A connector nobody else needs. A report with the customer's column names hard-coded into it. A prompt that only makes sense if you already know how that particular company files its invoices. It ships, the customer is happy, the engineer moves on.
Then what?
Most companies don't have an answer. The thing sits in a customer branch or a folder called custom/ and nobody looks at it again until it breaks. I've seen custom directories with hundreds of entries and not one person who could say what half of them did. People call that a services problem. It isn't. It's a product problem that never got written down.
The way I've come to think about it: every piece of one-customer work is a statement. The product could not do this on its own. That's a bug report. It just happens to be written in code instead of a ticket.
A normal bug report says I clicked this and it didn't work. A bespoke integration says I needed this and the product had no way to do it, so a person built it by hand. Same information. The second version is far more expensive to produce and, oddly, less likely to get read.
Why nobody files it
Filing a bug takes ten minutes and the filer expects nothing back. Building a one-off takes two weeks with the customer waiting. By the time it works the engineer is on a plane to the next site. There's a demo Thursday. Nobody's job is to look at what just got built and ask what it says about the product.
There's a quieter reason too. The one-off worked. The customer paid. Calling that a bug feels ungrateful, like complaining that the fire extinguisher was needed. But the extinguisher being needed is the whole point. That's the information.
When this work sat at a consulting firm none of it mattered. Custom work was the business. Now the same people are on the vendor's payroll and every hour spent on one customer is an hour the product didn't cover. That was the first essay. What I didn't say there is that the cost isn't really the hours. The cost is throwing away what the hours found out.
Two things it can be
Anything built for one customer ends up in one of two places, whether or not anyone decides.
It gets promoted. Someone looks at it, strips the customer's name out, and it becomes a feature, a config flag, a documented pattern. The next customer gets it for free. The engineer never builds it again.
Or it doesn't, and then it's evidence. Evidence that the product has a shape that doesn't fit this customer, probably doesn't fit others like them, and somebody decided that's acceptable. Sometimes it is acceptable. But it should be a decision. Not a default.
Most companies have a third bucket, which is neither. The work exists. Nobody promoted it, nobody declined it. It's just there. That bucket is where products go to become services businesses, one customer at a time.
The triage
Bug reports get triaged. Someone reads them, decides severity, assigns or closes. Bespoke work almost never does, and it's the same discipline. Same skills, even.
What I'd want, and have mostly not seen: every field engagement comes back with a short list. Here's what I built that only works here. For each item, could another customer want this, yes or no. If yes, what would it take to make it general. If no, why not, one sentence.
That's the whole thing. It isn't a process. The form doesn't matter. What matters is that someone on the product side reads it, within a week or two, and makes a call. Promote, decline, or hold. Decline is fine. Hold is fine for a while. Never looked at is the failure.
The reading matters more than the writing. An engineer can write the list in twenty minutes. If nobody reads it the second time they'll stop writing it, and they'll be right to stop.
What to measure
If you want one number: what fraction of bespoke work, by hours, gets a promote-or-decline decision inside a month. Not what fraction gets promoted. Just decided.
I'd rather see 30 percent promoted and 70 declined with reasons than 10 percent promoted and 90 never seen. The first company is learning something about its product. The second one is guessing, and paying field rates to guess.
There's a second number that's harder to get and tells you more. Of the last ten things that got promoted, how many customers used them afterwards? If the answer is one, the promotion was cosmetic. If it's four or five, the field team is finding real gaps.
Where this goes wrong
Some things should never generalise. A regulatory quirk that exists in one jurisdiction. A migration off a system nobody else runs. The triage should say decline, mean it, and nobody should feel bad about it. Promoting everything is as bad as promoting nothing; you end up with a product full of features one company needed once.
The metric can be gamed. An engineer who knows bespoke work gets counted will do less of it, or bury it in configuration where it doesn't register. A field team judged on promotion rate stops taking hard customers. Both are worse than not tracking at all. So the number belongs to the person running the function. It is not for anyone's performance review. The minute it shows up there it stops meaning anything.
And the analogy has a limit I should own. A bug is the product doing something wrong. A lot of bespoke work is the product doing nothing, in a place it never claimed to cover. That's still information. It's a different kind, though, and I've flattened it here.
The ask
Go look at your custom directory. Or your customer branches. Wherever the one-off work lives. Take the last twenty things in it and ask, for each one, whether anybody ever decided what it was. Not whether it works. Whether it was read.
If most of them were never read, you don't have a field engineering team. You have a consulting firm that forgot to invoice.