Last updated: September 12, 2026
How to calculate call center shrinkage
Shrinkage is what turns a headcount into a staffing requirement. It is the share of paid hours that cannot go to calls, so divide the hours lost to time off, training, meetings, and absence by total paid hours, then divide the agents you need on calls by one minus that figure.
Say you need 40 agents answering calls at two in the afternoon and shrinkage runs at a quarter. Hire 40 and you have about 30 on the phones, ten short every afternoon, in a building whose roster says it is fully staffed.
That gap is the whole reason the number exists.
The afternoon is not decoration either. Most floors protect the morning peak and push everything else into the back half of the day, so training, huddles, and coaching land between one and four. A monthly figure of 25% can sit on top of an afternoon block running past 35%, and the service level you promised a client gets missed in the same two hour window all month while the month end report reads fine.
The formula, and the division people skip
Step one is the ratio. Add the paid hours that did not go to calls in a period, then divide by total paid hours in that same period. A 20 seat floor paying 3,200 hours in a month, with 790 of them going to leave, training, meetings, breaks, and absence, is running 24.7%.
Turning that into seats is step two. Take the agents you need answering calls and divide by one minus the shrinkage. Forty over 0.753 is 53.1, so you schedule 54 and carry those extra fourteen people for as long as the floor exists.
The common error runs the other direction. Somebody takes 40, adds 25%, gets 50, and schedules 50. Fifty seats at a quarter shrinkage put 37.5 agents on calls, two and a half seats light against a 40 seat requirement, and the arithmetic looked like it accounted for shrinkage, which is what makes it hard to find in a spreadsheet somebody else built. At 35% the multiply version lands about five seats short instead of two.
Planned and unplanned shrinkage get forecast differently
Planned shrinkage sits on a calendar somebody controls: vacation approvals, holidays, class dates for a new hire group, the recurring Tuesday huddle, your coaching cadence. Next month's planned hours can be read off the schedule before the month starts, and they can be moved, which is the part most centers never use.
Unplanned shrinkage is a probability. Sick calls, no shows, an agent who logs in eleven minutes late, a carrier issue that takes the queue down for forty minutes. None of it is on a calendar and all of it has a history, so it gets forecast from your own attendance records rather than from a target somebody picked.
Treating the second one as a single average is where this usually goes wrong, because absence is not flat across a week and a floor staffed to the mean Tuesday runs short every Monday and every day after a holiday. Pull twelve months of your own attendance by day of week and buffer against the bad end of that spread.
There is a fair argument that the split is bureaucratic, since an hour off the phones costs the same whichever bucket it sits in. That holds for the ratio and falls apart for the plan, because planned hours are a decision you can reschedule out of the afternoon peak and unplanned hours are weather.
What belongs in each category
Seventeen rows, in the order a worksheet wants them. Put your own hours against each one for a recent month, add the column up, and divide it by the paid hours payroll reports for the same month.
| Category | Planned or unplanned | Where the hours come from |
|---|---|---|
| Vacation and paid time off | Planned | Approved requests plus accrued balances people have not used yet |
| Company holidays | Planned | The holiday calendar, adjusted for whoever covers the day |
| New hire training | Planned | Class dates and class size from the hiring plan |
| Ongoing training and product updates | Planned | Whoever books the sessions, usually with no backfill on the queue |
| Team meetings and shift huddles | Planned | The recurring calendar entry times everyone in the room |
| Coaching one to ones | Planned | Your coaching cadence times the agents who get one |
| Calibration sessions | Planned | Session length times the reviewers sitting in it |
| QA review time, where a reviewer also takes calls | Planned | Calls scored times minutes per call |
| Paid breaks | Planned | Break policy times shifts worked |
| Lunch, where lunch is paid | Planned | Counts only if your paid hours include it, which is a definition to settle first |
| Offline work, email, chat backlog, and admin | Planned | Whatever comes off the queue on purpose, plus what supervisors hand out ad hoc |
| Sick time and same day call outs | Unplanned | Your own attendance history, broken out by day of week |
| No shows | Unplanned | Attendance records, kept apart from sick time because the pattern differs |
| Late starts and early departures | Unplanned | The gap between scheduled start and first login, summed across the month |
| Unscheduled away time | Unplanned | Not ready states outside the break policy |
| System outages and downtime | Unplanned | The incident log, if anybody keeps one |
| Attrition gaps | Unplanned | Days between a resignation taking effect and a replacement taking live calls |
That table downloads as a CSV with a blank column for your own hours, rows in the same order.
Attrition is the disputable row. An open seat is not paid hours, so strictly it does not belong in the ratio, and it sits in the worksheet because every operator who has run a hiring gap writes it into the margin anyway.
Turning the number into a schedule
An Erlang calculation gives you agents required per interval from volume and handle time. Shrinkage sits between that answer and the schedule you publish, so run the division per interval rather than once for the month. The afternoon pattern above is what that difference looks like on a service level report.
A QA platform cannot tell you any of it. CallSniper scores the calls that were answered and knows nothing about the seats that were empty, so shrinkage belongs in whatever system builds your schedule. What scoring reaches is the damage, since rushed closes, skipped read backs, and callback numbers never confirmed digit by digit cluster in the intervals where staffing was short.
What goes in the denominator
Two bases circulate under the same word, several points apart. Paid hours includes vacation and holidays. Scheduled hours does not, because nobody schedules an agent onto the floor on a day they are on leave, so computing against it produces a smaller, tidier figure that works for a weekly roster and misleads an annual headcount plan.
Paid breaks and paid lunch are the other lever. Whether they fall inside your paid hours moves the figure by several points, and two floors running identical operations can report shrinkage six points apart on that choice alone.
The last base error is reaching for somebody else's number. No shrinkage figure we could trace leads back to a named publisher with a stated population, which is why the benchmarks page carries an empty cell there rather than a number. Figures do circulate, and every one we chased ended at a page quoting another page.
Write your definition down before pulling a single hour, then compare against your own trailing three months, the one comparison where the denominator belongs to you.
The hours your QA program takes off the floor
Quality time is shrinkage, and it is the category most QA programs get designed without.
Take a 20 seat floor and a modest program: five scored calls per agent per month, a 30 minute coaching conversation with each agent, and one 90 minute calibration session with two reviewers in the room.
The agent side comes to 10 hours of coaching. Against 3,200 paid hours that is about a third of a percent, a rounding error in the ratio, and if the agent side were the whole story this section would not be here.
Reviewer time is where the plan breaks, because a reviewer is one named person whose hours are already spoken for. A hundred calls at five minutes each is roughly eight hours of scoring. Add the ten hours of coaching conversations, which cost the supervisor the same half hour they cost the agent, plus three hours of calibration. Twenty one hours, about 13% of one supervisor's month, on a floor where that supervisor is usually scheduled on the phones too.
Push coverage toward the sample size the QA metrics page treats as defensible and the scoring line alone passes a hundred hours a month, most of a full time person nobody hired. That is why almost nobody runs that coverage, and why a QA plan belongs in the worksheet next to the training and the huddles before anybody gets promised a cadence.
Automating the scoring moves exactly one of those lines. A call review scores a batch of your own recent recordings against the scorecard you already use, which takes the review hours out of the row you just filled in and leaves the coaching hours on the schedule, where they were the point.
Frequently asked questions
What is the call center shrinkage formula?
Divide the paid hours that could not go to calls by total paid hours in the same period. To turn that into staffing, divide the agents you need on calls by one minus the shrinkage rate, so a 40 agent requirement at 25% needs 54 scheduled.
What is a good shrinkage percentage for a call center?
No published figure we could trace leads back to a named publisher with a stated population, so there is no benchmark worth quoting at you. Build the comparison from your own trailing three months, with your definition written down first.
What counts as planned shrinkage?
Anything already on a calendar somebody controls: approved vacation, holidays, training classes, huddles, coaching one to ones, calibration sessions, and paid breaks where your paid hours include them. Those hours can be forecast exactly and moved to a different interval.
Do breaks count as shrinkage?
They do when your paid hours include them, and that single choice moves the number by several points. Settle it before you calculate anything, then apply the same definition every month so the figures stay comparable.
How does shrinkage affect staffing calculations?
An Erlang calculation gives you agents required per interval, and shrinkage converts that into agents scheduled. Running it per interval rather than once per month is what keeps the afternoon from being short while the monthly average looks fine.
Sources
- No shrinkage benchmark traced back to a named publisher with a stated population as of 2026-09-12, so none is quoted here. Recorded as an empty cell on the benchmarks page.
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- The five minute per call review estimate. Editorial judgment, carried over from the QA metrics page.
- The 3,200 hour floor, the 40 agent worked example, the 20 seat QA program arithmetic, and the seventeen worksheet categories. Editorial judgment as of 2026-09-12.