What It Actually Costs When a New Agent Quits

Replacement cost for a contact center agent is best modeled as 15–30% of fully-loaded annual cost per agent — call it 22% as a working default — rather than as an absolute dollar figure. That single change is the difference between a cost model you can use across your whole delivery footprint and one that's quietly wrong everywhere except the country it was built in.

The published figures for this are all over the map, and they're all American. A composite of 2026 turnover studies puts direct replacement cost — recruiting, screening, training — at $10,000 to $20,000 per agent, rising to as much as $46,000 all-in once lost ramp productivity is counted. Gartner put the US total, including training, at $21,000 back in 2022. Those are real numbers from real research and they are useful right up to the moment you cross a border.

Why an absolute dollar figure breaks at the border

Look at what an agent costs to employ for an entire year, fully loaded — wage plus statutory contributions, benefits, facilities and supervision overhead. In the US and Canada that's roughly $66,000 to $101,500, with a midpoint around $83,750. In the Philippines it's about $5,400 to $7,300, midpoint $6,350. India runs about $7,000 at the midpoint, Colombia $10,250, Mexico $10,800, and a broad EMEA blend about $30,000.

Now put the $21,000 replacement figure next to the Philippine number. It is more than three times what that agent costs to employ for a full year. Applied offshore, the industry's standard replacement cost isn't merely imprecise — it produces a business case in which firing everyone and never rehiring would be cheaper than staffing the account. Any CFO who checks it will throw out the whole model, and they'll be right to.

A percentage survives the trip. Recruiting, screening and training all scale roughly with what labor costs in that market, because they're mostly labor themselves. Fifteen to thirty percent of fully-loaded annual cost lands inside the credible US range and stays sane in Manila, and it's what the readiness assessment uses for exactly that reason. At the 22% default:

• US / Canada — about $18,400 per agent. Ranging $12,600 to $25,100 across the 15–30% band. Note that this lands right on top of the published US research, which is the point: the percentage model doesn't contradict the dollar research, it explains it.
• Philippines — about $1,400 per agent. Ranging roughly $950 to $1,900. Same program, same mistakes, one-thirteenth the number.
• India — about $1,540. Colombia — about $2,255. Mexico — about $2,376. Nearshore LATAM generally lands between $2,000 and $2,500 per agent at the default rate.
• EMEA — about $6,600 per agent. On the widest band in the table, spanning Eastern European delivery centers through Western European in-house operations. If you operate in one specific market rather than across the region, this is the figure most worth replacing with your own.

Replacement cost is only about half the bill

Here's the part most models skip, and it's frequently the larger half. Replacement cost covers finding and training a body. It says nothing about the seat that body was supposed to be filling, which sits empty and then sits unproductive while the replacement ramps. If a new hire takes eight weeks to reach full productivity, you don't lose one agent — you lose one agent plus eight weeks of a seat.

What those eight weeks cost depends on what kind of operation you run, and the three shapes are genuinely different:

• Outsourcer: lost billable seat-weeks. An empty seat is revenue you contracted for and aren't earning. At a Philippine bill rate around $11/hour and a 40-hour billable week, that's roughly $440 per seat per week. In the US it's about $28/hour, or $1,120 a week — with tier-1 urban delivery running $32 to $42.
• In-house service: coverage cost. Nobody's billing you, but the contacts still arrive. The gap gets absorbed as overtime plus service-level degradation — modeled conservatively as a 50% premium on the vacant seat's loaded weekly cost, which for a US center is about $805 a week.
• Sales or blended: lost contribution. The seat should be generating bookings. Modeled at 1.0× the seat's loaded weekly cost, on the deliberately modest assumption that a productive sales seat contributes at least what it costs. Real contribution multiples run higher, so this is the floor, not the estimate.

The same operation, in two geographies

Take a 500-seat operation that hired 300 people over the last twelve months and has already lost 4 out of every 10 of them. That's 120 early leavers. Say a replacement takes eight weeks to reach full productivity, and say the operation scores 55 on the readiness assessment — which, as the next section explains, means roughly 35% of those departures get attributed to how people were prepared. That's about 42 leavers whose exit traces back to readiness.

Run it as a Philippine outsourcer: 42 attributable leavers × $1,397 replacement cost = about $58,000. Then 42 × 8 weeks × $440 of unearned billable seat time = about $147,000. Total: roughly $205,000 a year.

Run the identical operation as a US in-house service center: 42 × $18,425 = about $769,000 in replacement, plus 42 × 8 weeks × $805 of coverage cost = about $269,000. Total: roughly $1.04 million. If it were a US sales floor instead, where the vacant seat costs contribution rather than overtime, the ramp half roughly doubles and the total lands near $1.3 million.

Two things are worth sitting with. The first is the five-fold spread between two operations with identical numbers of people making identical mistakes — which is the case against a single global dollar figure, made concrete. The second is subtler and more useful: the halves invert. For the offshore outsourcer, lost seat-weeks are roughly 70% of the damage and replacement cost is 30%. For the US in-house center it's almost exactly reversed. An offshore BPO that models only replacement cost is capturing under 30% of its exposure — and the part it's missing is the part that gets fixed by shortening the ramp gap, not by recruiting harder.

Not every early leaver is your training program's fault

This is where cost models usually lose credibility, by implying that every departure was preventable. People leave for reasons no onboarding program touches: a better offer, a commute, a family move, a job that was never going to suit them. Attributing all of it to readiness produces a number nobody in the room believes, and a number nobody believes is worse than no number.

So the share attributed to preparation is tied to the readiness score itself, on a straight line: a program scoring 0 has 65% of its early attrition attributed to how it prepares people, and a program scoring 100 has 10% — the residual that no training program can prevent. A center in the middle sits in the middle. That upper bound isn't arbitrary either: between 69% and 73% of all contact-center turnover happens in the first year, so attributing roughly two-thirds of *early* attrition to preparation at the very bottom of the scale stays inside what the published data supports.

The practical consequence is that the score and the dollar figure are one finding rather than two numbers sitting next to each other. Improve the program, and the same headcount and the same turnover produce a smaller attributed cost — which is exactly the sentence a business case needs.

How to use a number like this without getting it thrown out

• Show every assumption, and let people change them. A cost estimate whose inputs are hidden reads as a sales tool. One where a skeptic can adjust the loaded cost, the replacement percentage and the bill rate and watch the total move is one they'll defend to their own VP.
• Use a percentage, then convert to dollars at the end. It's the only form that survives a multi-geography footprint, and it makes the estimate re-checkable when wages or FX move.
• Count the seat, not just the person. For an outsourcer this is usually the bigger half. If your model stops at recruiting and training cost, it's roughly a third of the real number.
• Round to something sayable. "About $1 million" survives a meeting. "$1,038,459" invites an argument about the estimate instead of about the program — false precision on a model is a liability, not a strength.

Every figure above comes from a reference table stored with its own source and a last-verified date, and the free Agent Readiness Assessment runs this exact model against your operation — your headcount, your hiring volume, your geography mix, your survival rate. It takes about five minutes, needs no login, and shows the dollar figure before it asks for an email. Every assumption in it is editable.

One caveat on the input side. This model needs an early-tenure loss rate, and if the only attrition number you have is the annual one, you can't produce it — the annual figure averages the problem away, which is its own piece.