Lost a Good Hire at Eighteen Months? Five Numbers That Tell You Why Before the Next One Goes

Start with the arithmetic on a single departure, because almost every retention argument inside a company is really an argument about a number nobody has written down. Take a role paying $62,000. You use a contingency recruiter at a twenty percent fee: $12,400. If you post it yourself instead, call that $600 in job board spend and a great deal more of your own time. Either way, the hiring manager screens somewhere around forty applications, runs ten phone screens at half an hour each, and puts five candidates through a panel that consumes three hours of combined staff time per candidate. At a loaded internal cost of roughly $55 an hour across the people involved, that is about $1,400 of screening and $825 of panel time, give or take the way your calendar actually falls. Then the new person ramps. Three months at half effectiveness against a $62,000 salary is about $7,750 of payroll that does not yet produce full output, and that assumes a clean ramp with a manager who has time to teach.

Add it up and the recruiter route lands somewhere above $22,000 before you count a dollar of overtime to cover the gap. The self-posted route is cheaper in cash and more expensive in weeks. What neither version captures is the client who had worked with that person for two years and now has to explain the account again. That number is real and it cannot be calculated honestly, so the correct move is to name it as unknown rather than invent a multiplier for it.

1. Cost the departure before anyone argues about the raise

The conversation that goes badly is the one where a manager asks for a $5,000 increase to hold someone and a finance partner says the budget is set. Both people are right in isolation and the exchange produces nothing, because the $5,000 is being weighed against zero rather than against the $22,000 above.

Build the replacement cost model once, for three or four representative roles, and keep it in a spreadsheet with the assumptions visible in the cells. Fee percentage. Hours of screening. Length of ramp. Loaded hourly cost. When somebody disputes the total, they have to dispute an assumption, which is a much more productive argument than disputing a conclusion. After you have done this through a few cycles, you notice the assumption that moves the answer most is ramp length, not recruiter fee, and ramp length is the one you can actually shorten.

A caution learned the hard way: the model justifies a retention raise only when the person is genuinely replaceable at the same price. If the market has moved and you would have to pay $68,000 for an equivalent hire, the raise is not a concession. It is a correction you were going to pay anyway, later, to a stranger.

2. Compare your current staff against what you would pay a stranger tomorrow

Pay compression is the most predictable retention failure there is, and it is invisible unless you deliberately look. It happens because starting salaries move with the market every quarter while internal increases move once a year, in a cycle set eleven months earlier. Two years of that and your three-year employee is earning less than the person they are training.

The check takes an afternoon. List every person in a job family with their current base. Next to each, write the number you would actually offer today to fill that seat with someone of equivalent capability. Not the number in the approved band. The number you would really offer when the role has been open nine weeks and the hiring manager is covering the work personally. The gap between those two columns, summed, is your compression exposure.

The Bureau of Labor Statistics is responsible for tracking employment, wages, and job openings and labor turnover across industries, and its series are the right starting reference when you want to know whether a shift you are seeing in your own hiring is local to you or general. Pair that with what your own recent offers have actually closed at, since your offer history is the most specific market data you will ever have and most employers never look at it in aggregate.

3. Look at when people leave, not just how many

A turnover percentage is nearly useless as a management signal because it averages together events with completely different causes. Twenty percent annual turnover made up of people leaving in month four is a selection and onboarding problem. The same twenty percent made up of people leaving at month twenty is a progression and pay problem. The remedies share nothing.

So plot tenure at exit. Take the last two or three years of departures, bucket them by months of service (0 to 3, 4 to 6, 7 to 12, 13 to 24, 25 plus), and count. With a small team the counts will be small, and you should resist reading a trend into four data points. Say so out loud when you present it. What you are looking for is a cluster obvious enough to survive the small sample, and clusters usually are obvious when they exist.

Two patterns recur often enough to name. An early cluster, inside six months, almost always traces to a gap between the job as described and the job as performed, which means the fix sits in the job posting and the interview loop rather than in compensation. A cluster at eighteen to twenty-four months usually means the second role never appeared: the person finished learning the first job, looked up, and saw nothing above them that they could reach. That one is fixable but not quickly, which is why finding it early matters more than finding it precisely.

4. Sort turnover by manager before you sort it by department

Departmental turnover figures are an aggregation that hides the thing you can act on. Break the same departures out by the person each employee reported to, including managers who have since moved on, and the distribution is rarely even. One team of nine will account for a third of the exits.

Before you conclude anything about that manager, check three alternative explanations, because experience says at least one of them is usually operating. First, did that team carry a disproportionate share of nights, weekends, or on-call? Second, is it the team with the lowest-paid entry role, meaning the exits are market-driven rather than relationship-driven? Third, did that manager inherit the team mid-year from someone else, which puts the departures in their column but not their doing?

If none of the three explains it, you have a supervision problem, and the useful version of that finding is specific rather than evaluative. Span of control above roughly eight or nine direct reports tends to mean one-to-ones get canceled first when the week compresses, and canceled one-to-ones are where small grievances mature into job searches. Reducing the span costs the salary of another supervisor, which you can now weigh directly against the replacement costs from the first check.

5. Count the commitments that quietly did not happen

Every person you hired was told something. A training budget. A schedule posted two weeks out. A review at six months. A path to a senior title. Those commitments are made verbally by a manager under time pressure and then recorded nowhere, which is exactly why they go unmet and exactly why the employee remembers them precisely.

Write them down at the point they are made. A single shared document per hire, with the promise, the date, and the owner, is enough. Then audit it quarterly and count the open items. The count itself is the metric. If a team has fourteen open commitments averaging seven months old, you do not need an engagement survey to know what the exit interviews will say.

Schedule predictability deserves its own line for hourly roles. The cost of posting schedules three weeks out instead of three days is mostly planning effort rather than cash, and it is the rare retention lever where the expense is genuinely small and the effect shows up fast in the early-tenure bucket from the third check.

Keep a dated note each time you make one of these calls: what you decided, what you expected to happen, and what the deciding number was. Six or eight entries in, the log starts telling you which of your instincts holds up under your own conditions and which one you should stop trusting. That is the only reliable way the judgment compounds, and it is why the person who has run this cycle a dozen times reaches a defensible answer in an afternoon while someone doing it for the first time spends three weeks and still guesses.