A person gives notice on a Tuesday. The replacement starts six weeks later. In between, nothing on the books says "turnover." What the books say is overtime, a canceled install, a second trip to a job that should have taken one, a customer credit, and three hours of a supervisor's week spent rebuilding a schedule that was already built. The cost is real and it is fully paid. It is simply filed under other headings.
That is the part most owners miss. Not the recruiter fee or the job board spend, which are visible and annoying and usually small. The expensive part is diffuse, and it lands in departments that have no reason to connect it back to a resignation.
Where the money actually shows up
Start with overtime. When a shift has to be covered, it is usually covered by someone already at or near forty hours. That premium does not appear in a hiring budget. It appears in payroll, where it looks like demand, which is the most flattering possible misreading of it.
Then rework. A new hire in month two is not slower in a way that shows on a timesheet; they are slower in a way that shows on a callback. The second visit, the corrected invoice, the order that went out with the wrong finish. Those get logged as quality issues, or as customer service, or as nothing at all.
Then the supervisor's calendar. Every departure consumes a manager's attention in increments too small to invoice: the exit conversation, the handoff list that was never written, the account that needs reintroducing, the interview loop, the first weeks of correction. A manager running two open roles is not managing. They are staffing. The work they were hired to do waits.
Then the quiet one. Institutional memory. Which customer wants a call before the crew arrives. Which supplier will take a Friday order. Which piece of equipment needs the breaker reset before it will start. None of that is written down anywhere, and none of it transfers in an exit interview. It leaves with the person, and the business relearns it at full price.
Finally, the contagion effect. The people who stay absorb the gap. They work the extra shifts and answer the questions. If the gap lasts long enough, one of them starts looking, and the next departure is not random. It is caused.
The sequence that decides whether someone stays
Retention is usually discussed as culture, which is a word for everything and therefore a plan for nothing. It is more useful as a sequence, because the steps have an order and the order matters.
Before day one. The offer letter, the start date confirmation, the note telling them where to park and who to ask for. A candidate who accepts and then hears nothing for two weeks is a candidate who takes a counteroffer. This is the cheapest retention work there is and the most commonly skipped.
Days one through five. Equipment that works, a login that exists, a named person responsible for them, and actual work by the second day. The failure mode here is not hostility. It is vagueness. Someone sitting in a break room waiting to be useful is forming a permanent impression of how the place runs.
Weeks two through six. This is where the job either becomes legible or does not. The new hire needs to know what good looks like, who decides, and how they will be told when they are off. A single scheduled check-in at the end of week two, with a second at week five, does more than any amount of general goodwill.
Month three. The honest conversation about whether this is working, in both directions. Most early departures are visible by week nine to anyone watching for them. The ones that get caught are the ones where someone asked.
Months six through twelve. Pay and path. If the person is doing more than they were hired to do, and nobody has acknowledged it in money or title, the clock has started. This is the point where a delay stops being recoverable, because by the time the conversation happens, the comparison has already been made somewhere else.
You can be late on almost every step above and still recover. You cannot recover from being late on that last one. Once a good employee has an offer in hand, a raise reads as a reaction rather than a judgment, and most people who accept a counteroffer leave within the year anyway.
The rules are not the same in two towns thirty miles apart
Federal wage and hour law sets a floor, and the Department of Labor is responsible for enforcing it. What catches employers out is that states and, increasingly, cities and counties build well above that floor, and the differences bear directly on retention.
Accrued paid time off is the clearest example. In some states, earned vacation is treated as wages, which means it cannot be forfeited and must be paid out on separation. In others, a written use-it-or-lose-it policy is enforceable. Same policy document, two completely different liabilities and two completely different signals to an employee weighing whether to leave in March or wait until after the summer.
Paid sick leave is now a patchwork of state and municipal ordinances with differing accrual rates, carryover caps, permitted uses, and documentation thresholds. An employer with crews working in three jurisdictions can be subject to three accrual schedules for one payroll run. Handled well, this is an administrative problem with a known solution: write the policy to the most generous applicable rule and apply it everywhere. Handled badly, it becomes the thing an employee resents weekly.
Predictive scheduling ordinances exist in a handful of cities and apply mostly to retail and food service. They require advance notice of schedules and premium pay for last-minute changes. The retention effect is the point of the law and also its most useful feature for the employer: a schedule posted two weeks out is a schedule people can build a life around, which is one of the most frequently cited reasons hourly workers stay.
Pay transparency rules, which now require posted salary ranges in a growing number of states and cities, have a second-order effect that lands inside the building. Once ranges are public, existing staff can see where they sit. Compression that was invisible becomes a question with a date attached. Employers who audit internal pay before the first posting goes up answer that question on their own terms.
Noncompete enforceability varies enormously by state, and in several it is sharply limited or unavailable for most workers. Where the agreement will not hold, retention has to be built from the job itself. That is the better foundation in any case, and it is the one that survives a move across a state line.
What to do first, and what can wait
Run the numbers you already have before you change any policy. Pull twelve months of departures with start dates, tenure, role and stated reason. Tenure distribution tells you which part of the sequence is failing. A cluster under ninety days is an onboarding problem. A cluster between twelve and twenty-four months is a pay and progression problem. They have different fixes and spending on the wrong one is money gone.
Then confirm the rules that actually apply to each work location, not just to your headquarters. Accrual, carryover, payout on separation, scheduling notice, posting requirements. Write them down once.
Then fix the handoff. A written list of what each role knows and nobody else does, updated quarterly, is unglamorous and it is the single thing that keeps a resignation from becoming a six-week outage.
Pay review comes after all of that, because it is the most expensive lever and the easiest to pull in the wrong direction.
The businesses that hold onto people are rarely the ones paying the most. They are the ones where the sequence runs on time: the offer confirmed, the first week organized, the ninety-day conversation held, the twelve-month review that happens before anyone has to ask for it. That is schedulable work, and schedulable work gets done.
