Call Center Attrition Cost Index 2026: Withdrawn, and What Replaces It

By Miki Furman, Founder & CEO | Published July 27, 2026 | Withdrawn and rebuilt August 2, 2026 | Version 2026.2

What was wrong, precisely

The index was a derivation, not a survey. It chained three inputs, and every one of them has now been withdrawn.

1. The attrition bands by geography. This page attributed them to a combination of a contact center research firm, an industry association, and Call Force Global operational data. None of the three legs holds. The research firm publishes no attrition figures broken out by offshore geography and none for the Philippines, the Caribbean or Latin America. The association figure traces to a single uncited assertion in a guest newsletter article from 2015, framed around the US rather than globally, at a URL that is now dead. And Call Force Global does not hold a proprietary operational attrition dataset, so nothing here was ever our own measurement.
2. The recruitment cost band presented as SHRM-anchored. Its upper bound, $4,683, is a real SHRM figure: the mean cost per hire in the 2022 Talent Access Report across 472 organizations. Its lower bound has no SHRM provenance whatsoever and has been deleted rather than restated here. Presenting the pair as a single SHRM band was wrong twice over: wrong about where the floor came from, and wrong about reading a mean and an invented floor as a range.
3. The offshore full replacement cost band of $3,000 to $8,000. No source was ever found for it.

Because every derived output depended on those three, nothing downstream survives. That is why this is a withdrawal and not a correction.

What the published record actually supports

The 60 to 70 percent Philippine figure still circulating is CCAP's own description of the era before 2016 and is not a current rate. A 28 percent figure sometimes quoted for the first half of 2023 is a six month figure and must not be annualized.

Why published attrition figures disagree so much

Put two attrition surveys side by side and most of the gap between them is definitional rather than real. Three choices move the number more than any difference in the underlying labor market.

• What counts as a separation. Some surveys count voluntary resignations only. Others add involuntary terminations, which is why the Philippine figure arrives as a 31 percent voluntary component and a 14.1 percent involuntary one rather than as a single number. Almost none count no-shows or candidates who accept an offer and never start, even though on a hiring floor those consume the same recruiting spend as a resignation does.
• What sits in the denominator. Average headcount over the period, headcount at the start, and headcount at the end produce three different answers, and they diverge most in exactly the case you care about, a floor that is growing or shrinking quickly.
• The period, and whether it was annualized. A six month rate doubled is not an annual rate. Departures cluster in the early weeks of tenure, so scaling a partial year up by a multiplier inflates the result whenever the sample window happens to contain a hiring wave.

The practical consequence: comparing your attrition against another company's is close to meaningless unless both figures were built the same way. Your own rate, measured consistently across four quarters, tells you more than any industry average, because at least it is comparable to itself.

What replacement costs, and what nobody publishes

Cost per hire is the one replacement cost input with a real published anchor, and the useful thing about it is the spread rather than the headline.

Always state the edition and whether you are quoting a mean or a median. The roughly fourfold gap between the $4,683 mean and the $1,244 median is right skew: a small number of very expensive hires carrying the average. For budgeting, use a median. If you are making the case for retention spend, quote the mean and say that you are. The BLS annual figure is published directly by BLS and is not the hourly wage multiplied by 2,080.

What has no published figure at all: training cost per new hire, equipment and provisioning cost per new hire, and the length and depth of the ramp curve. Call Force Global works to roughly six weeks at roughly 70 percent lost productivity. That is a Call Force Global estimate, not a published measurement, and we are labeling it as one rather than dressing it up as a benchmark. Use your own payroll and your own ramp data instead.

Also worth saying plainly: BLS publishes no separations rate for SOC 43-4051, so any attrition percentage you see attributed to BLS or JOLTS for this occupation is misattributed. BLS is a wage source here, nothing more.

The method survives. The numbers did not.

The formula was never the problem. Attrition cost really is a rate multiplied by a cost per departure, and the cost per departure really does split into direct replacement spend plus the productivity lost while a replacement ramps. What made the old index wrong was filling those slots with numbers nobody had measured. Filled with your own numbers, the same model is sound.

Three rules for running it honestly:

• Use your own measured attrition rate, not a geography band. Count separations over average headcount over a stated period, and write down whether you are counting voluntary only, voluntary plus involuntary, or also no-shows and abandonment. Half the disagreement about attrition numbers in this industry is definitional rather than real.
• Use your own recruiting spend. If you have to borrow, borrow the SHRM median rather than the mean, and say which one you used. A mean that sits four times its own median is telling you the distribution has a long tail, not that the average is typical.
• Measure your own ramp. Ramp length and the depth of the productivity gap vary more between two floors in the same city than between two countries. Ours is roughly six weeks at roughly 70 percent, which is a Call Force Global working assumption and not a published measurement.

The turnover cost calculator runs this formula live on your inputs, with every default editable and labeled. The attrition benchmark tool scores your rate against the published series where one exists and tells you plainly where one does not.

How to build this index for your own operation

Five steps, in order, using only numbers you can pull from your own records.

1. Divide separations over twelve months by average headcount over the same twelve months. Write the definition down beside the result: voluntary only, voluntary plus involuntary, and whether no-shows and abandonment are in or out. The definition is part of the number, not a footnote to it.
2. Split that rate by tenure band instead of reporting one blended figure. At minimum, separate losses inside the first 90 days from losses after them. Blending hides the shape, and the shape is what tells you what to fix. Loss inside the first 90 days is a screening and onboarding problem and is usually the cheaper one to solve. Loss around month nine is a wage and progression problem, and solving that one costs money.
3. Add up what a replacement actually costs you from your own invoices: recruiting spend including agency and referral fees, training hours priced at the trainer's loaded wage, and equipment and provisioning.
4. Measure your ramp against a tenured cohort. Track a new hire cohort's output per hour week by week beside a tenured cohort on the same queue. The weeks before the curves meet is your ramp length, and the gap between them is the productivity you lost.
5. Multiply, then sanity check per seat. Rate times cost per departure is your annual cost per seat. If it lands above a seat's annual wage, recheck the ramp assumption before presenting it.

When you ask a provider for their attrition instead, ask for the same things: the rate by tenure band, the definition behind it, the period it covers, and whether it is a mean or a median across their programs. A provider who answers with one blended annual percentage and no definition has told you almost nothing.

Every figure we removed

Listed in full so that anyone who quoted this page can find their number and see that it is gone. None of these should be republished, and none of them should be re-sourced to a different name.

What changed in how we publish

Three rules came out of this, and they apply to everything we put on the site from here.

• An attribution names an edition. Not a firm, an edition, with the year and the sample size where the publisher gives one. If we cannot name the edition, we do not have the source.
• A mean and a median are different claims. Every statistic on our pages now says which one it is. Where both exist, we publish both, because the gap between them is usually the most informative thing about the number.
• A Call Force Global estimate says so at the point of use. Not in a footnote, not in a methodology page, in the sentence itself. We hold no proprietary attrition dataset and will not present our working ranges as though we do.

People reasonably ask what it would take for cost figures to come back to this URL. The standard, rather than a promise about timing: a named collection period, a stated respondent count, one published definition of what a separation is applied across every row, results reported by tenure band and not only blended, and the underlying record open enough that a reader can recompute the output rather than take it on trust. Until a dataset clears that bar, this page stays a withdrawal notice and a pointer to the sources that do.

The broader picture on what the attrition data does and does not support by geography is in our call center attrition guide, which carries the same figures and the same admissions.

Download the data

The CSV at this URL has been rebuilt. It no longer carries any derived cost figure. It now holds one row per published statistic: market, metric, value, unit, reference period, whether the figure is a mean, a median or a component, its status, and a sources column naming the publisher and edition for every row. Markets with no published series appear with an empty value and their status recorded as such, rather than with an old number and a new caveat.

Download the published attrition and cost reference (CSV)

CC BY 4.0. Republish, syndicate, or build derivative datasets freely with attribution to Call Force Global and a link back to this page, and cite each external figure to its own publisher. Companion datasets: the Caribbean Nearshore BPO Wage Index 2026 and the AI agents in Search Console study, which is a first-party study with a published dataset and reproduction script.

How to cite this page

Published under a Creative Commons Attribution 4.0 license. Cite, quote, and republish freely with attribution. The external statistics belong to their original publishers and should be cited there. Suggested citation for this page:

Frequently asked questions

For deeper reading: the full call center attrition guide, which carries the same figures and the same admissions plus the root causes and the retention plays; the attrition benchmark tool, which scores your rate against the published series and says plainly where none exists; the turnover cost calculator for the model on your own inputs; the Caribbean Nearshore BPO Wage Index 2026 for the wage side; the 2026 call center metrics benchmarks; and outsourced call center services if you want to talk about a floor rather than a spreadsheet.