CX Operations · Cost
Agent attrition is a P&L line. Most scale-ups book it under HR
In most scale-ups, agent attrition gets discussed once a year, in a people review, next to recruitment cost. It is reported as a percentage. Nobody in the room converts it into money, so nobody in the room fights for the budget that would reduce it.
Meanwhile the same company will spend three weeks debating a tooling contract worth a fraction of it.
Why you should not use a published benchmark
Search for the cost of replacing an agent and you will find confident figures. Most are American. Most are published by companies selling either staffing or software. And most collapse two things that are not comparable: an internal employee in a European market with statutory notice and severance, and a seat on an outsourced floor where replacement is the partner's problem and the cost reaches you as churn in service quality.
The range that circulates sits between thirty and forty-five percent annually, with outsourced voice floors reported higher. I went looking for the primary source behind it. What I found was a chain of citations leading back to American industry publications, several of them quoting the identical range years apart. A number that does not move for that long is being copied rather than measured.
It is still useful for understanding the shape of the problem. It is not your number, and putting it in front of your CFO invites a debate about the source instead of a decision about the operation.
Build your own. It takes an afternoon.
The five buckets
One: recruiting. Agency fee or the loaded hours your internal recruiter spends. Advertising. Interview time from your team leads, which is real capacity taken off the floor.
Two: the non-productive period. Salary during training, plus the trainer, plus the floor time of whoever shadows them. Count this in weeks, honestly. Most operations quote their formal training length and forget the two weeks of supervised handling that follow.
Three: the below-average window. This is the bucket everyone skips. After go-live, a new agent resolves less and takes longer for a period that varies by product complexity. Look at your own data: plot resolution rate against tenure, in months. The curve tells you how long you subsidise each hire.
Four: coverage while the seat is empty. Overtime, temporary agency cover, or service level that simply degrades. If it degrades, that is not free, it is just uncounted.
Five: what leaves with them. The hardest to price and the largest. An agent at month eighteen knows the workarounds, the edge cases, and which customers need handling differently. None of that is in the knowledge base.
Measure first-year attrition, not annual attrition
An annual percentage hides the thing that matters. In the operations I have run, departures often clustered in the first year, though not consistently and not in every market, which is exactly the point. You cannot assume where your churn sits. You have to look.
Split your leavers by tenure cohort: under three months, three to twelve, over twelve. The three groups tell you three different stories. Departures under three months usually point at recruiting or onboarding. Between three and twelve months, at the daily experience of the job. After a year, at the absence of anywhere to go next.
You cannot fix all three with the same intervention, which is why a single annual number leads to the wrong action.
The link to your customer metrics
SQM Group research identifies agent attrition as a leading driver behind falling first contact resolution scores. The mechanism is not mysterious. A newer agent resolves less on the first attempt. Unresolved contacts come back. Repeat contacts add volume. Added volume increases pressure on a team that is already short. Pressure increases departures.
If your FCR is drifting down while headcount looks stable on paper, check the tenure distribution behind the headcount. The number of seats can be flat while the experience sitting in them halves.
This is also why cost-cutting through attrition tends to backfill itself. The savings arrive in the salary line and the cost reappears in repeat contact volume, where nobody is looking for it. The measurement problem is the same one I described in the piece on CSAT and FCR.
What actually moves it
Pay sets a floor. Below market, nothing else you do will hold. Above the floor, the reasons people leave are rarely primarily about salary.
In the operations I have run, the changes that moved retention were about control rather than compensation. Agents having input into how the work gets done before being asked to do more of it. A progression path that exists visibly rather than being mentioned once in the interview.
And removing the metrics that punish the behaviour you claim to want. Tell a team you care about resolution, then rank them on handle time, and the agent who asks the extra question that actually closes the case ends the month behind the one who closed fast and left the problem open. Nobody concludes the targets are badly set. They conclude that the speech about quality was decoration. After that they stop listening when you talk about quality, because they have already seen what you reward.
The psychology behind why the shouting approach produces compliance and kills ownership is in the piece on locus of control. It is the single mechanism I would want a GM to understand before approving any retention budget.
The number to bring to the board
One line. Your annual replacement cost, built from your five buckets, multiplied by your actual leaver count, split by tenure cohort.
Put next to it the cost of the intervention you are proposing. If the intervention costs less than the churn and you still lose the argument, the problem is not the finance case. It is that nobody in the room believes attrition is addressable, which is a different conversation and usually a more honest one.