Call center shrinkage: What it is and how to reduce it

Call center shrinkage is the difference between paid staff scheduled and available, and it materially increases the headcount required to meet service levels. Enterprise operations often carry this percentage as a fixed assumption that leaders approve once a year in the budget cycle, and the staffing model divides every forecast by it.

An illustrative Monday service-level report shows 400 human agents on the schedule and, at once, 268 of them logged into the queue. Abandonment ran past target, and the workforce management (WFM) team has asked for 40 more heads. The 132 who were missing still received pay: they were in training or a one-to-one, on break, out sick, or waiting for a frozen desktop to restart.

What counts as unavailable agent time

Call center shrinkage is the share of paid human agent time when agents are not available to handle customer contacts. Planners calculate it from paid time only, so an unpaid lunch never enters it and a paid 15-minute break does. Annual leave, training, team meetings, sick days, and system outages all count, whether the planner saw them coming or not.

Planners remove shrinkage before calculating occupancy: it reduces the available hours that planners measure occupancy against, so one understaffed floor can post high occupancy per person and high shrinkage per team in the same interval. Shrinkage is a planning input and says nothing about how hard the people on the phones are working.

It matters in the staffing model because planners divide the demand forecast by what remains after shrinkage; they do not add a percentage on top. That distinction between dividing and adding is where the arithmetic starts.

How to calculate call center shrinkage

The core formula is straightforward: Staff Required = Staff Demand ÷ (1 − Shrinkage%). Planners divide; they do not add, and the whole staffing model rests on that single operation.

Consider an operation that needs 100 human agents in the queue in a given interval at 40% shrinkage. It must schedule 167, because 167 × 0.6 rounds to 100. The additive shortcut, 100 plus 40%, schedules 140, and 140 × 0.6 leaves 84 in the queue: a 16-agent shortfall in every interval the shortcut touches.

That difference between dividing and adding is the shrinkage trap. It shows up again when planners sum component percentages measured against different bases, as though they were all fractions of the same paid hours. Correct arithmetic on a single, consistent base of paid hours is what makes the resulting headcount defensible, and it also determines which activities feed each row of the calculation.

An example of call center shrinkage components

Shrinkage components split by who controls them, and the clearest way to see the impact is to stack them against a single base of paid hours. External components are absences the operation does not control day-to-day: annual leave, public holidays, sickness, and lateness. Internal components cover activities the operation schedules itself and time lost to system problems.

The example table below shows how ordinary scheduled and unscheduled time stacks up to roughly a third of paid time on an illustrative floor.

The totals are illustrative model inputs; an outsourced sales floor and an in-house claims desk will each produce a different table. The order of operations does not change between them: shrinkage comes off paid hours first, and planners measure occupancy against the hours that remain. The two sit side by side in any set of call center efficiency metrics a planner reports upward, and each row shapes what leaders can act on next.

What causes shrinkage and what a good rate looks like

The causes split by whether the planner can see them coming, and the unplanned share is where budgets break. The list below groups them by controllability, since that determines whether a planner can act on a row or only forecast it.

• Scheduled offline time: Planned development activities, meetings, and paid breaks. Planners schedule these activities and can move them within each site's notice rules.

• Unplanned absence: Sickness, short-notice leave, and lateness arrive interval by interval. Last year's average cannot predict these absences precisely.

• System and process loss: Technology failures and slow back-office tools can hold a logged-in human agent off the queue. This category can hide inside "other activities" with no clear owner.

• Attrition-driven staffing shortfall: Budgeted but unfilled seats create a separate shortfall, and paid classroom time for new hires counts as shrinkage. ContactBabel's Steve Morrell told No Jitter in November 2025 that US attrition runs about 30% (opens in a new tab) and has held there for seven or eight years, so vacant seats push occupancy up on remaining agents and drive the next resignation.

Treat published industry benchmarks as directional conventions, not operating targets. Manage instead to your own component data, tracked by half-hour interval and rolled up monthly, because that is the only number that reflects your sites and your absence patterns. Pushing the percentage down too aggressively raises burnout, and burnout is what turns a shrinkage rate into a headcount problem at enterprise scale.

What shrinkage costs at enterprise scale

At 3,000 seats, shrinkage is a headcount line the CFO reads. Start with 100 scheduled human agents at 35% shrinkage: 65 sit in the queue in any interval. Drop shrinkage to 30% and 93 seats deliver the same 65, freeing 7 full-time equivalents (FTEs) per 100 scheduled. Across a 3,000-seat operation, that recovers roughly 200 FTEs, the ceiling of what schedule discipline delivers. Removing 30% of human-handled volume takes about 900 seats out of the base staff requirement before shrinkage applies.

The blended daily average hides a second cost: shrinkage spikes in adjacent half-hour intervals, and one day's percentage understaffing covers the worst of them. Intraday fixes then carry their own cost, because Los Angeles County's predictable scheduling ordinance (opens in a new tab), along with rules in Seattle, San Francisco, Philadelphia, Chicago, Oregon, and New York City, triggers predictability pay when planners reshuffle to cover a spike.

Capacity that carries no shrinkage sits outside the gross-up entirely. HSE's AI agent handles 3 million automated calls a year and 600 simultaneous calls, taking no breaks and needing no advance notice to absorb a peak.

Five levers that reduce staffing pressure without adding headcount

The first two levers and the fourth trim shrinkage. The third reduces handling workload, and the fifth shrinks the base staff requirement that shrinkage multiplies. Together, they reduce staffing pressure through distinct parts of the capacity model.

1. Plan to 15- or 30-minute intervals

Forecast shrinkage for each interval and schedule training, meetings, and breaks against that forecast.

Interval planning surfaces the half-hours where scheduled offline time stacks up, so planners can move activities before they collide with a demand peak. It also removes most of the same-day changes that predictability pay penalizes, because activities land where the schedule already accounted for them rather than being reshuffled inside the notice window.

2. Enforce schedule adherence in real time

Real-time adherence tracking flags a human agent who returns from break four minutes late and lets a team leader intervene within the same shift. It also identifies after-call work (ACW) beyond the average handle time (AHT) model, though ACW contributes to handling workload and falls outside shrinkage. Recovered break minutes come off the internal rows without touching anyone's contract or notice period, so the gain lands in shrinkage without triggering predictability pay.

3. Cut after-call work with agent-assist summarization

AI-generated call summaries and disposition codes shorten ACW on every call, and on a floor handling tens of thousands of calls a week, the recovered minutes add up to whole FTEs. The calls still arrive, and each still needs a human, so this lever reduces handling workload rather than shrinkage. It compounds with interval planning by giving the same forecast a shorter per-call handle time to staff against.

4. Place training and coaching in forecast troughs

Move training, quality sessions, and one-to-ones into the intervals the forecast marks lowest, and publish those blocks inside the notice window so they do not trigger predictability pay. The training hours do not disappear; they stop landing on the intervals that produce abandonment.

Combined with interval planning, this lever converts development time from a service-level risk into a scheduled activity that fills troughs the operation would otherwise carry as idle time.

5. Route routine calls to AI agents

The fifth lever directly reduces human forecast and base staffing requirements. An AI agent recognizes the caller's intent from natural speech, authenticates against a backend system, and completes routine service without a human, including routing and appointment booking. Edge cases hand off to a human agent with the transcript and identified intent attached. Because the AI agent runs 24/7 across multiple languages, human staffing needs drop before planners apply shrinkage, making contact center automation a staffing decision.

BarmeniaGothaer reduced switchboard workload by 90% with its AI agent, which uses smart call routing on every call. That volume is removed from the base before the gross-up, and no break rotation or adherence dashboard reaches it.

Reduce call center shrinkage by shrinking the volume you staff against

Shrinkage is the multiplier a Head of CX inherits, but volume is the base they can move. Every routine call that reaches a human enters the gross-up and competes for the same limited capacity that revenue-generating sales conversations and high-stakes support cases depend on. Cutting the volume that requires human handling shortens the distance between a caller who needs an answer and a queue with someone free to give it, and it protects the intervals where a sales inquiry and an escalated support issue arrive at the same moment as a routine password reset.

Parloa puts AI agents on routine sales and support calls so those contacts never enter the shrinkage calculation. The AI Agent Management Platform covers Build, Optimize, and Observe across the AI agent lifecycle, runs 24/7 across 140+ languages, and can go live in weeks.

Book a demo to see how many of your calls you never need to staff.

FAQs about call center shrinkage

Is shrinkage the same as occupancy?

No. Planners remove shrinkage from paid hours first, then measure occupancy against the remaining hours, so the two use different denominators. A team can post high occupancy and high shrinkage in the same interval, and each number tells the planner something the other cannot.

How should shrinkage targets differ across sites in different jurisdictions?

Set the target and the contingency plan per site. A site covered by a predictive scheduling ordinance cannot move breaks or call in extra people the same day without paying predictability pay, so its planned shrinkage needs a larger interval-level allowance than a site with full intraday flexibility.

How often should shrinkage be recalculated?

Recalculate it monthly and by half-hour interval, using the previous month's actuals for every component. An annual assumption locks last year's absence and attrition pattern into this year's schedule.

Does BPO pricing already include shrinkage?

Yes. Providers price seats gross of shrinkage, so the client pays for scheduled offline time and absence inside the hourly rate. Ask for component-level shrinkage reporting in the contract, and audit it against the seat-hours the provider actually delivers on the queue.

Does shrinkage still matter once AI agents handle most calls?

Yes, for the human pool. Forecast the AI-handled stream and the human-handled stream separately, apply the gross-up only to the human stream, and expect the human stream's mix to shift toward longer, harder calls that change AHT and coaching load.