Every WFM team plans for unplanned absence. Most plan for the average: if 8% of agents are usually off, schedule enough extra people to cover 8%. On paper that works. In practice the service level still misses on the days that matter, and nobody can explain why, because the average absence report says the plan was right.
The reason is that absence is not an average. It is a spread. Some days 3% are off, some days 14% are off, and the days with more absence cost far more service level than the quiet days give back. This guide uses a simple Erlang C model to show what that does to your service level, and how many extra agents it really takes to be safe.
The setup
We took one half hour interval, a 360 second handle time and an 80/20 service level target. For each team size we worked out the agents needed on the phones, then scheduled enough people to cover an 8% average absence rate exactly. Each scheduled agent then has an independent 8% chance of being absent that day.
| Team size | Agents needed on the phones | Scheduled for 8% average absence | Average service level | Intervals that actually meet 80/20 |
|---|---|---|---|---|
| Small | 14 | 16 | 89% | 87% |
| Medium | 30 | 33 | 81% | 73% |
| Large | 57 | 62 | 81% | 62% |
| Very large | 109 | 119 | 82% | 65% |
Read the last two columns together. For the large team the average service level is 81%, which looks fine. But only 62% of intervals meet the target. More than a third of the time, the team is short.
This is the same effect that makes average numbers look safe while the daily experience is poor. For a similar trap in forecasting, see forecast accuracy: MAPE, WAPE and bias.
Why the average fails
Service level does not fall in a straight line when agents go missing. Losing one agent when you are exactly on target costs more service level than gaining one agent adds. So the bad days hurt more than the good days help, and the result is below what the average promises.
At the same time, the plan has no spare capacity. Scheduling for exactly the average means that half of all days will have more absence than planned, by definition. Without a buffer, half the days are short, and the ones that are short by 3 or 4 agents miss the target badly.
How many extra agents do you need?
We added extra scheduled agents on top of the "average absence" plan and measured how often an interval met the target.
| Agents needed on phones | Scheduled for average absence | Extra agents to reach about 90% of intervals | Extra cost |
|---|---|---|---|
| 14 | 16 | +1 (96% of intervals) | about 6% |
| 30 | 33 | +2 (94% of intervals) | about 6% |
| 57 | 62 | +3 (93% of intervals) | about 5% |
| 109 | 119 | +4 (93% of intervals) | about 3% |
Smaller queues need a bigger percentage buffer, because one absence is a larger share of the team. Larger queues need more agents in total, but a smaller percentage. In every case, a buffer of about 3% to 6% lifts the share of intervals on target from a range of 62% to 87% up to above 90%.
When absences cluster, it gets worse
The model above assumes each agent decides independently. Real absence is not independent. Mondays, the day after a payday, school holidays, flu waves, public transport problems and a bad shift culture all push many people out on the same day.
We tested this with the large team (57 needed, 62 scheduled) by letting absences cluster, while keeping the average at 8%.
| How absences behave | Average service level | Intervals meeting 80/20 |
|---|---|---|
| Independent | 80.5% | 62% |
| Mild clustering | 77.3% | 63% |
| Moderate clustering | 75.1% | 64% |
| Strong clustering | 74.2% | 67% |
The average absence rate never changed. Yet the average service level fell by about 6 points. Clustering makes the bad days much worse, and the roster has no way to recover. This is why the day of the week and the season matter more than the yearly absence average. If your Monday absence is double your Wednesday absence, you should not schedule the same buffer on both days. Read more about the shrinkage side of this in the shrinkage formula mistake that breaks service level.
What to do about it
- Plan absence by day of week and month, not as one yearly number. Use at least 12 months of history.
- Add a variability buffer on top of the average. A starting point is 3% to 6% extra scheduled agents for teams of 15 to 100 on the phones, higher for small teams.
- Measure how often intervals hit the target, not only the average service level. A team can have 81% average and 62% hit rate.
- Pre-agree a cover plan for a bad day: voluntary overtime, shift swaps, split shifts and a short list of people who can start early or stay late. The earlier you act, the cheaper it is. See the intraday reforecast trigger is too late.
- Use flexible shifts. Teams with flexible start times and self-scheduling cover gaps faster. See shift flexibility and self-scheduling.
- Pool small queues. A small team needs a larger percentage buffer. Combining skills reduces the variability.
- Do not punish all absence the same way. Genuine illness needs a different response than a pattern of last minute absence. Fair, simple rules protect both service level and morale.
Quick checklist for your own data
| Question | Why it matters |
|---|---|
| What is absence by day of week? | Mondays and Fridays are often far above average |
| What is the 90th percentile absence day? | Plan the buffer against this, not the average |
| How many intervals hit target each week? | Shows the true risk |
| How many agents can be recalled or extended within an hour? | Sets the size of the buffer you need |
| How is absence split between sick leave, no-show and late arrival? | Different causes need different controls |
Use the shrinkage calculator to combine planned and unplanned time, the Erlang C calculator to find agents needed on the phones, and the schedule adherence calculator to see how much service level is lost during the day.
Key takeaways
- Planning for the average absence rate leaves roughly 25% to 38% of intervals short for teams needing 30 or more agents on the phones in this model.
- An average service level of 81% can hide only 62% of intervals meeting the target.
- A buffer of about 3% to 6% extra scheduled agents lifts the hit rate to above 90%.
- Clustered absence (Mondays, holidays, illness waves) can cut average service level by about 6 points without any change in the yearly absence rate.
- Report the hit rate and the 90th percentile absence day, not only the average.
The figures come from a simplified model with the stated assumptions (one 30 minute interval, 360 second AHT, 80/20, 8% average absence). Test the buffer against your own absence history before changing the roster.