How many agents do you actually need?
The classic Erlang C staffing model, with one extra slider: the share of calls an AI voice agent resolves before they ever reach the queue.
Your busiest hour
Staff for the peak interval, not the daily average.
Add an AI voice agent
Handed-off calls still reach your agents, so only calls resolved end to end are removed. AI agents on WhautoVoice run concurrently with no queue of their own.
Down from 0 without AI, at the same service level.
Every seat you do not need to hire is roughly a salary. An AI agent on WhautoVoice starts at US$59 a month.
Start free trialHow the calculator works
Erlang C is the queueing model call centers have used for decades. It takes the traffic in an interval, expressed in Erlangs (calls per hour multiplied by handling time in hours), and for a given number of agents returns the probability that a caller has to wait at all. From that it derives the share answered within your target time and the average speed of answer. The calculator increases the agent count one at a time until your service level target is met, then checks that occupancy stays under your ceiling and adds agents if not. Finally it divides by (1 minus shrinkage) to turn agents on the phones into people on the roster.
The AI column runs exactly the same model on a smaller offered volume: the calls the AI resolves end to end never reach the queue. Everything else stays the same, so the two columns are a fair comparison.
Methodology
- Traffic (A) = calls per hour x AHT in seconds / 3600.
- Probability of waiting = Erlang C(N, A), computed from the Erlang B recursion for numerical stability.
- Service level = 1 - P(wait) x e^(-(N - A) x T / AHT), where T is the target answer time.
- Average speed of answer = P(wait) x AHT / (N - A).
- Occupancy = A / N.
- Roster = agents on the phones / (1 - shrinkage), rounded up.
- AI deflection reduces offered calls by the share you set before the model runs. Calls the AI hands off are assumed to take the same handling time as any other call.
Why the numbers are not linear
Queues behave badly near capacity. When agents are 95% busy, a small bump in volume produces a big jump in waiting, because there is no slack to absorb it. That is why the last few points of service level are so expensive, and why removing a share of calls with an AI agent often removes a larger share of the waiting. It also means AI deflection is most valuable at your peak interval, which is the one you are staffing for anyway.
Typical planning values
| Input | Typical value | Notes |
|---|---|---|
| Service level | 80% in 20 s | Sales lines 90% in 15 s, back office 70% in 60 s |
| Average handling time | 4 to 6 minutes | Include after-call work |
| Shrinkage | 25% to 35% | Higher with generous leave or heavy training |
| Maximum occupancy | 80% to 85% | Sustained occupancy above 90% drives attrition |
| AI resolves end to end | 30% to 60% | Depends on how routine your call mix is |
Frequently asked questions
Erlang C is the standard queueing formula for call centers. Given the volume of calls in an interval, how long each takes, and how many agents are available, it predicts the probability a caller has to wait, the average wait, and the share answered within a target time. Staffing tools run it in reverse: they add agents until the service level target is met.
The most common target is 80% of calls answered within 20 seconds. Sales lines often aim higher, back-office lines lower. Pushing service level from 80% to 95% costs disproportionately more agents, which the calculator will show you.
Shrinkage is the share of paid time agents are not available to take calls: breaks, training, meetings, sickness and holidays. 30% is a typical planning figure. The calculator divides the Erlang C headcount by (1 minus shrinkage) to get the people you need on the roster.
Calls an AI voice agent resolves end to end never reach the queue, so they are removed from the offered volume before Erlang C runs. Because queueing costs grow faster than volume, removing a third of calls often removes more than a third of the wait, and the agents who remain run at healthier occupancy.