Shrinkage is the most quoted and least examined number in workforce management. Everyone knows the typical range, roughly 30 to 35 percent of paid time in most contact centers, and nearly every capacity plan has a single shrinkage cell somewhere that feeds the roster. That one cell is where two very common, very expensive mistakes hide.
The first is arithmetic. The second is averaging. Both look harmless on a spreadsheet and both show up as an unexplained service level miss on the floor.
Mistake one: multiplying instead of dividing
Shrinkage is the share of paid time agents are not available to take contacts. If shrinkage is 33 percent, only 67 percent of rostered time is productive. To keep 10 agents live, you do not add 33 percent to 10. You divide by 0.67, which gives about 15 agents.
Here is what that error costs on a real interval, using a standard Erlang C model. Take 1,000 contacts in a 30-minute interval, an average handle time of 276 seconds, and an 80/20 service level target. The model needs 163 agents on the phones. With 30 percent shrinkage:
| Correct method (divide by 0.70) | Common error (multiply by 1.30) | |
|---|---|---|
| Agents needed live | 163 | 163 |
| Agents rostered | 233 | 212 |
| Agents actually available after shrinkage | 163 | 148 |
| Service level | 83% | Collapses toward 0% |
The multiply method looks reasonable, 212 versus 233 is only about 9 percent fewer, and that is exactly why it survives in so many spreadsheets. But at a busy interval, being 15 agents short does not cost you a few points. Occupancy goes past 100 percent, the queue becomes unstable, and wait times grow without bound.
Mistake two: one flat number for the whole day
A blended shrinkage figure treats every interval the same. Real shrinkage does not behave that way. Breaks and lunches cluster around the middle of the day. Meetings and coaching get dropped into whatever slot looks quiet on the schedule, and that slot is often just ahead of a rush. Unplanned absence tends to bunch on Mondays and after holidays. The average can be exactly right while the peak interval is quietly worse.
Here is what that does. Start from a roster of 233 built correctly on 30 percent shrinkage, and let the actual shrinkage at the peak interval drift above plan:
Two extra points of shrinkage at the peak take service level from 83 percent to the mid 50s. Four extra points and the interval is effectively lost. This is the cliff that makes peak intervals so unforgiving: the closer the roster runs to full occupancy, the less room there is for any drift at all.
Why the average hides the problem
A weekly shrinkage report that says 31 percent looks like a rounding difference from a 30 percent plan. Look at it by interval and it can be 28 percent in the early morning, when almost nobody is off the floor, and 36 percent through the midday peak. The report averages those together and the problem disappears into the total.
The same logic applies to the split between planned and unplanned time. Meetings, training, and coaching are discretionary, you choose when they happen. Absence, lateness, and system downtime are not. Reporting one number hides which part you can actually move, and the discretionary part is where the fastest fix usually is.
What to do instead
Always divide, never multiply. Rostered agents equal required agents divided by one minus shrinkage. Check every template and every spreadsheet in your operation for this, because the multiply version is easy to miss and expensive to leave in.
Plan shrinkage by interval, not by day. Build an interval-level shrinkage curve from the last eight to twelve weeks. Breaks and lunches belong to the intervals where they actually land, and the plan should show it.
Split planned from unplanned in every report. Planned shrinkage is a scheduling decision. Unplanned shrinkage is a forecasting problem. They need different owners and different fixes.
Protect the peak. Keep meetings, coaching, and training out of your highest-volume intervals. Moving one coaching session from the busiest hour to a quiet one costs nothing and is often the cheapest service level gain available.
Add an unplanned buffer by day of week. If Mondays consistently run several points above plan, say so in the plan instead of absorbing it as variance.
Watch occupancy next to shrinkage. Industry operating guidance generally treats the 75 to 82 percent occupancy band as sustainable, with attrition risk climbing quickly once floors run above 85 percent for long. A roster that survives only by pushing occupancy to the top of that range is already telling you the shrinkage assumption is too low.
The bottom line
Shrinkage is not a single number, it is a curve, and the interval where it runs highest is usually the interval where you can least afford it. Fix the arithmetic first, then fix the averaging. Both changes take an afternoon, and both protect service level at the exact point where it is most fragile.
Test your own numbers with the free Erlang C Calculator, and use the Capacity Planning Calculator to build shrinkage into your roster correctly.