Essay · attrition
Attrition is not the weather.
The replacement cost is the part that reaches the budget. The rest of the bill is paid inside the operation, months before anyone resigns.
The replacement cost is real and it is defensible: recruiter fees, background checks, trainer time, the seat that sits empty between a resignation and a start date. It gets forecast, reported, and argued over on a schedule.
It is also the smallest part of the bill.
Underneath that number is an arrangement I call the Disposable Agent model:
"Hire fast. Train cheap. Expect churn. Replace. Treat attrition as a budget line, not a failure."
(excerpts from my working manuscript, here and below)
The last clause is the expensive one. Once attrition is a budget line, it has been priced rather than diagnosed, and the operation stops asking what produced it.
Below the waterline
Recruiter fees and trainer time are the visible line items. The full local calculation tests what happens underneath.
Ramp-up latency. A new agent generally produces less independent work while learning the role. How deep that gap runs and how long it lasts belong to the operation's own cohort data, not a published benchmark.
The quality dip. New agents make mistakes. They give wrong information, miscode tickets, route to the wrong escalation queue. Mistakes cause callbacks. Callbacks generate volume. The center starts manufacturing its own work, and the manufactured volume comes back into the forecast looking like demand.
The veteran two seats over. The cheapest escalation is to a peer: before the knowledge base, before the ticket, before the supervisor, an agent with a hard case turns to the veteran two seats over and gets an answer no system records. The veteran is other people’s performance. When they leave, those escalations do not stop; they get priced: longer holds, more transfers, a heavier supervisor queue. The cost of one exit surfaces as a drift in everyone else’s numbers, in reports that never mention the person who left.
There is a fourth the staffing model can price directly and almost never does. Ratios are not org-chart trivia. They are operating constraints, and the span a floor can carry depends on its tenure mix. A team full of nesting agents needs the 1:1s, side-by-sides and floor presence that a settled team can go without. Replace enough of a floor and the same headcount now requires a richer ratio than the one it is funded for, so the coaching capacity meant to develop the people who stayed is spent on the people who just arrived. Nobody books that as a cost of attrition. It is one.
The chain runs backwards from the exit
Attrition reads as weather because the decisions that produced it were made months earlier, in a different part of the operation, by people solving a different problem.
Shrinkage is where to look, because shrinkage is not one number. It is a stack of operating choices hiding inside one number. PTO is not the same as coaching.
"A 1:1 is not waste because the agent was not taking calls for thirty minutes. It is relationship infrastructure, performance infrastructure, and retention infrastructure."
Collapse all of it into a single percentage and the operation can no longer tell absence from investment. So it cuts the visible off-phone work first: the huddle, the coaching block, the calibration session, the development hour.
From there the sequence is easy to follow and easy to miss. Coverage comes up short. The coaching block is the only thing on the calendar that can be moved without a customer noticing. The 1:1 is canceled for queue pressure, and what the agent hears is that the queue matters more than they do. Repeated cancellations make the session look conditional and suppress the issues it was meant to surface. Development stops. Quality drifts. Two quarters later a resignation appears in a report, attached to a reason that explains nothing, with no trace of the Tuesday it started on.
That sequence is a hypothesis, not a law, and it should be treated as one. Track what was canceled, whether quality or coaching signals moved afterward, and whether the exit evidence points to those conditions rather than to pay, leadership, the labor market, demand, or a policy change. An operation that cannot separate those has the same problem in a different place: the week service level missed because the forecast was wrong needs a different answer from the week it missed because coaching was canceled and the queue ate the people.
The instrument for that test is already in the process, though it rarely works as built. The person leaving holds the most accurate diagnosis in the building and, out of professionalism, declines to file it: the stated reason gets edited into something everyone can live with, and the operation settles its own attrition record with the leaver’s polite help. The silence did not start at the exit. It started whenever speaking up stopped being worth it. One question gets past this, because it asks for a moment rather than an accusation: "In the last six months, what was the moment you decided you were going to start looking?" A moment is just a fact. Naming it indicts nobody, so it can be answered honestly on the way out the door. The answers (recurring shapes across many exits, not quotations from any one interview) sound like "my team leader canceled my 1:1 for the third week in a row", or "I asked for a day off six weeks ahead and was told no with no explanation." None of those is pay. None of them is the bots. Each points at a management decision the operation could have made differently, on a date it could name.
I have been on the wrong side of this. I canceled the 1:1 because the queue was hot and told myself that was leadership. The team noticed every time.
What follows
The conclusion is not that development is virtuous. It is that development is capacity, and capacity is either funded or it is not. If the center says development matters but does not allocate capacity for development, it does not have a development strategy. It has a wish competing with the queue, and the queue wins every Tuesday, quietly, in a way no report captures until the resignations arrive.
That is why attrition belongs next to the measurement question, not the recruiting budget. When someone says we hit the number, the question worth asking is what did we bend to get there. Attrition is one of the slowest answers that question has. The service level held. The coaching hour paid for it. The exit lands two quarters later in a different report owned by a different function.
An operation that keeps skilled people can outperform one that repeatedly rebuilds the same capability, even when experienced people cost more per hour. That comparison only works if it includes the full local cost, not the wage line. The investment side also runs ahead of the reading: coaching time and schedule flexibility are visible long before the return is mature enough to see, which is exactly when they are easiest to cut.
So stop calling attrition "the weather." It is not. It is an operating outcome with a P&L number attached. Weather is something you plan around. An operating outcome has inputs, and inputs can be watched, funded, protected, and owned.
Want to talk about this? Email me.