The shop had four people including the owner, and had grown enough that a fifth was overdue. The hire started in March and left, by agreement, in October. Nobody behaved badly at any point, which is worth saying, because most stories of this kind are told as though someone did.
The figures below are rounded and reconstructed, and the useful part is the ranking rather than the totals.
The wage, which is the part everyone counts
Seven months of pay for a mid-level technician, plus payroll taxes and the workers' compensation premium on the added payroll. Call it the largest single line and roughly a third of what the episode cost.
That proportion is the point of the whole exercise. Most owners, asked what a failed hire costs, name the wage. It was not close to the whole bill.
Recruiting, twice
The first search ran six weeks: the posting, the reading, nine phone screens, four interviews, two reference calls. Say twenty hours of the owner's time, spread across evenings, plus the cost of the listings.
Then it ran again in October, at the same cost, in a worse month, with a shorter pipeline because the good candidates from the spring had jobs by then.
Twenty hours is not an abstraction in a four-person shop. It is quoting that did not happen, which shows up as a thinner October than the year before.
Training, which is mostly the owner's billable time
Three weeks of close supervision, then two months of frequent interruption. In a small shop the trainer is the most productive person in the building, which means training is paid for twice: once in wages and once in the work the trainer did not do.
Count it honestly and the training period cost more than the wage during the same weeks. This is the line that convinces people to keep a marginal hire, incidentally, and it is a sunk cost. The training is spent whether or not the person stays.
Rework and the two customers
Four jobs had to be revisited. Three were small and fixed inside a morning. The fourth was an installation that failed at a customer's premises, on a Friday, in front of a property manager who handled six buildings.
That property manager stopped calling. The other loss was a residential customer of nine years who had recommended the shop to neighbors regularly and did not after that spring.
Neither of those appears anywhere in the accounts, and between them they were the largest number in this list. A repeat customer's value is not one invoice, it is the invoices they would have sent for another decade plus the ones they generated by talking.
The overtime nobody planned
From June, when it was clear the work was not being covered, the two experienced technicians picked up the slack. Six weeks of consistent overtime at premium rates.
Then the quieter cost: one of them raised it, twice, and was told it was being handled. Being told a problem is being handled while nothing changes is the specific thing that makes good employees start looking, and the shop came close to losing one of them in August.
The cost of waiting from June to October
This is the line worth isolating, because it is the one an owner controls.
By early June there was enough evidence to decide. The decision was made in October. Those four months contain most of the rework, all of the overtime, both customers, and the near-loss of an experienced technician. The wage during that stretch is almost incidental next to what accumulated around it.
The reasons for waiting were entirely ordinary: the training was already paid for, the person was pleasant and trying, summer was busy, and firing somebody is unpleasant. Every one of those is real, and together they cost more than the hire did.
The costs that arrived after the last day
Two of them, and neither is obvious in the moment.
The separation itself produced an unemployment claim, which is ordinary and was not contested. What people forget is that unemployment insurance is experience rated: an employer with claims against its account pays a higher rate on future payroll, for years, across everyone on the books. The amount per employee is small. Multiplied by a payroll and a few years, it is not nothing.
Then the paperwork. Final pay rules vary by state and some require the last check within a short window of the end date, along with accrued vacation where the state treats it as wages. Getting that wrong turns a clean parting into a penalty, and it is one of the few parts of this whole episode with a deadline attached.
What the shop changed
Three things, none expensive.
A paid working day as part of hiring, arranged properly with the pay run, doing real work alongside someone. Two candidates in the next search declined it, which was information. The one hired was visibly comfortable by lunchtime.
Written checkpoints at four, eight and twelve weeks, twenty minutes each, with two lines recorded afterward. The point is not the record. It is that a scheduled conversation forces a judgment while it is still cheap.
And one sentence agreed in advance: if two experienced people independently raise the same concern, that goes in the calendar as a decision date rather than as a note to keep an eye on it.
What it looked like the second time
The replacement started in November and was working unsupervised by the end of January. The property manager came back the following year, after a call from the owner that acknowledged what had happened without making excuses for it.
Counted across both hires, the episode cost the shop the better part of a year of growth. What it bought was a hiring process that a four-person business can actually run, and the knowledge that the expensive part of a bad hire is never the hire. It is the four months after everyone already knows.