Answering 100 calls an hour at an 80/20 service level (80% of calls answered within 20 seconds) takes 10 agents on the phones at a 4-minute average handle time (AHT), and 15 on the schedule once you allow 30% shrinkage. Those 10 agents answer 86.8% of calls within 20 seconds, with an average wait of 12.6 seconds.

That can look padded: the calls add up to only 6.7 hours of work an hour, and each agent is idle for a third of it. The spare capacity is what makes 80/20 possible in a queue this small. Nine agents fall to 74.2%, and an AI agent resolving half the calls removes 4 of the 10 seats, not 5. You can change any input in our free Erlang C calculator, which opens prefilled with this case.

Lorikeet publishes this guide and is one of the platforms compared: it makes the free calculator at lorikeet.tools and Workforce Manager, both mentioned here. We checked competitor details on their own websites in November 2026; if something has changed, tell us and we will update the page.

Key takeaways

• 10 on the phones, 15 scheduled. The workload is 6.7 erlangs, and 10 agents run at 66.7% occupancy.

• Small queues pay a premium. You need 10 agents per 100 calls here, against 8.5 at 200 calls an hour and 7.8 at 500.

• One absence breaks the target. Nine agents answer 74.2% within 20 seconds, and eight answer 52.3%.

• AI savings arrive in steps. Resolving 20% or 30% of calls both leave 8 agents on the phones; resolving 50% leaves 6.

Why does 100 calls an hour need 10 agents when the work only fills 6.7?

Because calls arrive at random, not on a timetable. The workload is 100 calls x 240 seconds / 3,600 seconds = 6.7 erlangs, so 6.7 people could cope only if every call landed the moment someone came free. In practice a short burst can occupy a small team completely, and Erlang C prices that risk. Counting up from 7 agents gives this:

Ten is the first count above 80%. Then add shrinkage, the paid time people spend off the phones: 10 / 0.7 = 14.3, which rounds up to 15 scheduled. Our Erlang C staffing guide explains the formula itself, and the 200 calls an hour page walks through it step by step.

What is this answer assuming?

How big is the small-queue penalty?

The same target at the same handle time gets cheaper per call as volume grows, because a larger pool absorbs bursts that would swamp a smaller one:

Doubling from 100 to 200 calls adds 7 agents, not 10. Two practical points follow. First, avoid splitting a queue this size: two specialist lines of 50 calls an hour need 6 agents each, so 12 people do the work of 10. Second, the idle time is not waste you can trim. Each agent has about 20 free minutes an hour, or two minutes between calls on average, and removing one person fails the target. If you want that time back, give agents work they can drop when the queue builds. At 500 calls an hour the problem reverses: occupancy, not service level, sets the limit.

What does one absence do to a queue this small?

One person is 10% of your phone capacity here, against 2.5% in a 40-agent queue. The service level falls fast:

The schedule of 15 gives less slack than it seems. At 30% shrinkage, 15 scheduled yields 10.5 on the phones on average. One unplanned absence on top of that leaves 14 x 0.7 = 9.8, already under the 10 you need. Shrinkage covers the average rate of absence, not two people out on the same morning. A 16th person on the schedule absorbs one absence and still leaves 10.5. To size that allowance from your own data, see our guide to shrinkage, adherence and occupancy.

Volume surprises hurt less here than absences do. With 10 agents, 10% more calls (110 an hour) gives 77.9% and 20% more (120 an hour) gives 65.4%. Bunching works the same way: if 60 of the 100 calls arrive in the first half hour, that half needs 11 agents and the second needs 8, and a flat 10 answers 65.4% within 20 seconds while the rush lasts. Our guide to forecasting support volume covers interval forecasts.

How much does AI resolution save at 100 calls an hour?

Less than the percentage suggests, and in whole seats rather than smoothly. These figures are illustrative and keep the 4-minute AHT on the calls people still take:

Moving from 20% to 30% resolved saves no seats at all. Seven agents on 70 calls reach 79.4%, just short of the target, so you keep 8 and the gain shows up as faster answers. You can open the 70-call case to see it. If the simpler calls went to the AI agent and the rest take 4.5 minutes, 20% resolved needs 9 on the phones and 13 scheduled, which saves one seat, while 30% still needs 8 and 12.

A smaller human queue is also more exposed. At 50 calls an hour with 6 agents, one absence is 17% of your capacity and takes the service level to 71.6%; two absences take it to 37.8%. So round to whole seats before you build a business case, and keep an absence allowance when you remove people. If the case depends on cutting seats at a 20% or 30% resolution rate, it is weak at this volume; shorter waits may be the bigger gain. Our guide to AI support staffing and volume planning works through this for larger teams.

How do handle time and target change the count?

Each extra minute of handle time adds two agents on the phones at this volume, so a coaching change that trims 30 seconds can matter as much as a hire: 4.5 minutes needs 11 on the phones and 16 scheduled.

Does 100 chats an hour work the same way?

Approximately. Divide handle time by the number of chats each agent runs at once. At 8 minutes and two chats at a time, the effective handle time is 240 seconds, so 100 chats also needs 10 agents. Pushing to three at a time usually lengthens each chat: at 10 minutes and three chats the effective time is 200 seconds and you need 9, so the extra load saves only one agent. Call Centre Helper's Erlang C for chat guide explains why handle time rises with concurrency.

Do you need WFM software at this size?

Often not. Fifteen people in a peak hour can live in a spreadsheet if one person owns it. The test is how many people the hour turns into. If you held 100 calls an hour across 60 open hours a week, that is 15 x 60 / 40 = 23 people on 40-hour contracts, as an upper limit because demand usually falls away from the peak. At that point rest, leave and skills checks start to take real time, and our guide to replacing a scheduling spreadsheet covers when to switch.

What this answer leaves out

• Quiet hours. Erlang C gives very low counts at very low volume. You still need enough people to cover a break or a long call.

• Skills. It assumes any agent can take any call. If only some can handle billing, you effectively have the split queues described above.

• Hang-ups. Erlang C treats every caller as patient. If many give up, Call Centre Tools offers Erlang A, which models it.

How we chose and checked these tools

We computed every figure with a short Erlang C script (no abandonment, smallest whole number of agents that meets the target, shrinkage applied by dividing and rounding up) and matched the main, absence, surge and AI cases against the live lorikeet.tools calculator API on 6 November 2026. The table lists free tools that answer this question, described from each tool's own page on the same date. Soon publishes the same 10 and 15 for these inputs, as you would expect from the same formula.

Is Lorikeet the right fit?

At this volume you may not need Lorikeet at all: the calculator is free. Lorikeet becomes relevant if you want an AI concierge taking part of these calls, or once the team has passed 20 agents and the roster has become a job of its own. Workforce Manager builds the roster from ticket and call volume, checks each draft against contracted hours, rest, skills and leave, according to the launch post, and now "manages the day as it happens", per Lorikeet. It is not the right choice for everyone:

• If 15 people on one phone queue is the whole team, Workforce Manager is built for 20 or more agents, and a spreadsheet or lighter tool will do.

• If all you wanted was the count, you have it, and your current roster tool can stay.

• If you must see a Workforce Manager price or integrations list before trying it, Lorikeet publishes neither.

• If an AI agent would resolve only 20% to 30% of your calls, the seat math above may not justify it on its own.

Workforce Manager pricing is not published. For the AI concierge that resolves calls, Lorikeet publishes its pricing: the Start plan is $2,100 a month and Scale is $5,100 a month (both billed annually), with resolved chat, email and SMS conversations at $0.99 (Start) or $0.90 (Scale) and voice at $1.50 or $1.20. Unresolved tickets and handoffs to a person are not charged. Signature is custom, with a $75K platform fee. Full details are on the pricing page. To test either on your own calls, start a 30-day free trial.