Ability · Calculation with example

Estimate call-handling staff by time slot

Content updated:

Short answer

Estimate phone staffing from calls per time slot, talk and after-call time, and the time each person has available. Dividing the workload gives an initial reference; if you must promise a waiting time, calculate each interval with Erlang C, which often needs more agents when volumes are low. Validate peaks and concurrent calls before turning the result into shifts.

Key verification: Compare expected workload, answered calls and observed waiting during a trial shift.

Sources and limitations

To estimate how many people should answer calls, calculate the workload of each time slot and divide it by the minutes each person can realistically give to the phone. Include talk time and after-call work. The result is a first estimate of capacity; it does not guarantee a waiting time or absorb arrival peaks. If you need to promise a response time, repeat the calculation with Erlang C, which this guide also works through with the same figures.

A daily average can hide a saturated morning and a quiet afternoon. Use time slots that match your shifts and make them shorter when calls are concentrated in brief periods.

The data you need

  • Calls offered to the team in each slot, including those it could not answer.
  • Average talk time for comparable enquiries.
  • After-call work: notes, preparing a reply or closing a task.
  • Minutes reserved for other duties, breaks and meetings.
  • People who are actually present and able to handle that enquiry.

Call history helps you check timings and records; Aircall explains how to query and export it in its Call History guide. Do not confuse the time since a call entered the system with an agent's working minutes. Waits and transfers require you to check how each field is defined.

If you only have data on answered calls, the estimate may understate demand: some potential work was lost before anyone spoke. Record that limitation on the calculation sheet.

A workload formula with explicit assumptions

Slot workload = expected calls × (talk minutes + after-call minutes).

Capacity per person = slot length × share of time dedicated to the phone × planned occupancy within that time.

Estimated people = workload ÷ capacity per person, rounded up.

The dedicated share removes other tasks. Occupancy leaves headroom within phone time for variation and gaps between calls; do not deduct the same meetings there again. None of the percentages in this example is a standard you should copy.

Example with three time slots

Assume five-minute enquiries plus one minute of after-call work. Each person spends 75% of the hour on the phone; within those 45 minutes, 80% occupancy is planned. Usable capacity is 60 × 0.75 × 0.80 = 36 minutes per person.

Estimate call-handling staff by time slot: table 1
Time slotExpected callsWorkloadDivisionEstimated people
9:00–10:00848 minutes48/36 = 1.332
10:00–11:001272 minutes72/36 = 22
11:00–12:0018108 minutes108/36 = 33

The data is invented to make the arithmetic reproducible. The table estimates planned capacity; it does not show that two people will answer every call in an hour, nor does it recommend hiring three permanent staff.

What changes if conversations run longer

In the twelve-call slot, if talk time and after-call work add up to eight minutes, the workload rises to 96. The division 96/36 = 2.67 requires three people under the same assumptions. Reviewing handling time can matter as much as looking at volume.

Separate enquiry types if their workloads differ. A new quote and a short confirmation should not share an average without checking the mix. You can calculate each type separately and add their minutes within the slot.

The main limit: simultaneity

Twelve calls spread across an hour behave differently from twelve calls that arrive together. The formula above adds up work, but it does not model the order of arrival, caller patience, the number of lines or the spread of call durations. That is why the workload division is not an Erlang C calculation and does not promise any waiting time.

Choose the method according to the decision you need to make, not according to which one gives fewer people:

Which calculation method to useDecision diagram. If you do not need to promise a waiting time, divide the workload by the usable minutes per person to get the people needed in the slot. If you must promise one, use Erlang C per interval to get the agents logged in, then divide by one minus absence to get the headcount to schedule. Both branches end in a trial shift that compares the forecast with real records. Must you promise a maximum wait? No Yes Workload ÷ usable minutes per person Erlang C per interval and target People in the slot, no waiting promise Agents logged in ÷ (1 − absence) Trial shift and real-world check
The two branches answer different questions: the left one distributes minutes of work; the right one estimates how many people must be logged in to meet a waiting target. Both are validated with a real shift.

Second calculation: the same slots with Erlang C

If you commit to a target, for example answering 80% of calls within 20 seconds, enter each slot in the Erlang C staffing calculator. With the same sample data: a 60-minute interval, an average handling time of 360 seconds (five minutes of talk and one of after-call work), no occupancy cap and 25% unavailable time, equivalent to the 75% dedicated share used above.

Estimate call-handling staff by time slot: table 2
Time slotTraffic (erlangs)Agents logged inService level (20 s)ASAOccupancyHeadcount with 25% absenceWorkload method
9:00–10:000.8395.4%8.5 s26.7%42
10:00–11:001.2387.2%28.2 s40%42
11:00–12:001.8488.6%21 s45%63

Traffic is calls × handling time ÷ interval length: 12 × 360 ÷ 3,600 = 1.2 erlangs. In the 10:00 to 11:00 slot, two agents give a forecast service level of 57% and an ASA of 202.5 seconds, which is why the model asks for three. With few calls, the resulting occupancy is low; that is not inefficiency but the cost of answering quickly when calls arrive at random.

Edge case: if, in the 11:00 to 12:00 slot, ten of the eighteen calls arrive between 11:00 and 11:15, that quarter-hour carries 10 × 360 ÷ 900 = 4 erlangs and needs 7 agents logged in for the same target. The hourly average hides this, so calculate in 15- or 30-minute intervals when calls bunch together. Erlang C also assumes that nobody hangs up while waiting and that any agent can take any call.

If your company is covered by Spain's Law 10/2025

Article 10.3 of Law 10/2025 on customer service requires 95% of calls to be answered, on average, in under three minutes. It applies to utilities, transport, postal services, electronic communications and financial services, and to other businesses when the group has at least 250 employees or exceeds certain turnover or balance-sheet thresholds; the law allows a twelve-month adaptation period. Check whether you are in scope with your adviser.

With the example and a target of 95% within 180 seconds, the calculator asks for 3, 4 and 4 agents logged in across the three slots. Planning against that target helps, but compliance is demonstrated with the phone system's real records and a definition of "answered" consistent with the law.

Service level: write the measurement contract

The proposal here is to count from queue entry to human answer, with an inclusive 20-second threshold: a call answered at exactly 20 seconds counts as inside. Count one entry per call rather than each ringing attempt. Exclude outbound calls and identified tests before calculating. These are the rules of this proposal, not a guaranteed phone-system formula.

A recorded greeting picking up is not a human answer. Spain’s Law 10/2025 on customer service says so expressly for businesses within its scope: a call is not considered answered unless the customer can state the reason and ask for personal attention from an operator (article 10.3, consolidated text in the BOE, accessed 27/10/2026). If your timer starts when the call reaches the phone system rather than the queue, time spent in the menu also uses up the threshold; state which you use.

CloudTalk defines the Service Level in its Wallboards as the percentage of incoming calls answered within the target response time, and shows abandonment rate and average waiting time separately. The page does not say when the timer starts or how abandonments are treated (accessed 27/10/2026). Before comparing the dashboard with your spreadsheet, confirm this with support: two fields sharing a name do not guarantee the same denominator.

From measured service level to number of agents

Service level describes what happened. To plan how many people you need, the Erlang C calculator for agents and service level uses the same definition, the share of calls answered at or within the threshold, but with a different assumption: nobody hangs up while waiting. With its example values, 120 calls in 30 minutes and an average handling time of 180 seconds, 16 agents give 86.9% in 20 seconds and 15 agents give 77.1%.

That forecast is not directly comparable with the percentage you measure in your reports. In a real queue, callers who abandon free up capacity, and your report includes abandonments in the denominator; the model does not. Use it to size each interval and then check against real measurement. Interval-based staffing is covered in the opening sections of this guide, and abandonments are counted separately with the abandonment rate, discussed further down.

Occupancy: why the mean of percentages can mislead

Take two sample agents, one on a full shift and one on a shorter shift. Both use mutually exclusive states and the same handling definition. The purpose is to calculate a team figure, not compare personal effort: shift allocation and demand may differ.

Estimate call-handling staff by time slot: table 3
AgentHandling, minAvailable without handling, minOther activities, minShift, min
A30010080480
B6014040240
Team360240120720

A has 75% occupancy and B has 30%. Their simple average is 52.5%, but team occupancy is 360 / (360 + 240) = 60%. Agent A's denominator has twice the weight: 400 minutes versus 200. Team shift utilisation for handling is 360 / 720 = 50%. That percentage uses the whole shift, including time classified as other activities.

Occupancy: preparing intervals that can be added up

  1. Export start, end, agent and state. Declare the time zone and split intervals at reporting boundaries: a state running from 23:50 to 00:10 contributes ten minutes to each day.
  2. Sort by agent and check overlaps. Two rows covering 09:00–09:10 do not equal twenty available minutes. Retain the conflict for investigation against the source.
  3. Separate unrecorded gaps. Fifteen missing minutes must not automatically become waiting time or disappear from the total to improve occupancy.
  4. Apply a stable state catalogue: talk, hold and after-call work can count as handling in this exercise, without duplication. Document differences from the vendor's report.

The calculation does not transfer directly to concurrent chats: two conversations open for ten minutes do not create twenty minutes of shift time. Combining channels requires a definition of elapsed person time versus interaction workload.

Abandonment rate: agree the unit and the exclusions

Use queue entries, rather than every attempt to the number or unique people. A call that hangs up in the menu before entering the queue belongs to another stage of the journey. Write down how you treat transfers, repeat calls from the same person and exits to voicemail or a scheduled callback: if the queue offers that exit and the caller accepts it, it is not abandonment but an intended outcome of the queue configuration (in Spanish).

If the system cannot distinguish a voluntary hang-up from a fault, keep an “unclassified” category instead of inventing a cause. And if the phone system itself disconnects callers because of waiting time, do not count it as customer abandonment: it is a configuration decision that needs a separate review.

Classifying a queue entryDecision diagram for classifying each queue entry. First question: if a person answered it, it is an answered call. If not, second question: if the queue offered voicemail or a callback and the caller used it, it is an offered exit and does not count as abandonment. If not, third question: if there is an identified technical fault, it is recorded as a fault, outside the numerator. If not, fourth question: if the caller hung up before the short-call threshold, it is a short hang-up excluded in the adjusted view. In every other case it is abandonment and goes into the numerator. Queue entry Answered by a person? Yes Answered No Voicemail or callback offered by the queue? Yes Offered exit not abandonment No Technical fault identified? Yes Technical fault outside numerator No Hung up before the short-call threshold? Yes Short hang-up excluded if adjusted No Abandonment: numerator
Each queue entry ends up in exactly one category. The order of the questions matters: offered exits and identified faults are ruled out first, and only then do you decide whether a short hang-up counts. The categories and threshold are a CallsIQ proposal, not a standard.

Abandonment and Law 10/2025: if your company is bound by it

Spain’s Law 10/2025 on customer service, in addition to the Article 10.3 target of 95% in under three minutes already discussed, forbids in Article 8.4 cutting off a call because of a long wait. The law does not define abandonment or say how to count callers who hang up before being answered. Keep the full journey and the written definition, and ask your adviser how they fit the calculation you present.

AHT: define the components before adding them

In this model, talk excludes hold and after-call work starts at hang-up. Earlier waiting in the queue is not agent handling time: it is the customer's time and is measured with other metrics, such as service level or average speed of answer. Check your provider's definition, because its talk column may already include hold. Also state what happens to transferred calls, internal calls and follow-up actions completed outside the phone system.

Which parts of a call make up AHTTimeline of a call with five segments: queue wait, talk, hold, talk and after-call work. A bracket over the last four segments shows that they make up the AHT of the call. Queue wait is outside AHT. A dashed line marks the moment of hang-up, between the second talk segment and after-call work. A note warns that if the talk column already includes hold, it must not be added again. AHT of this call Queue wait Talk Hold Talk After-call work Hang-up Outside AHT Time If “talk” already includes hold, do not add it again.
Queue wait sits outside AHT; hold and the work after hang-up are part of it. Compare this outline with the definition of each column in your report before adding them up.

What changes in staffing if AHT is wrong

AHT is one of the inputs to the Erlang C staffing calculator. Example: 120 calls in an hour and a target of 80% answered within 20 seconds. With a correct AHT of 6 minutes, the load is 120 × 6 / 60 = 12 erlangs and Erlang C calls for 16 agents, giving an 83.6% service level. If the 1-minute average hold has been added twice and AHT shows as 7 minutes, the load rises to 14 erlangs and the calculation calls for 18 agents. Two more people because of a definition error, not because of more work.

That is why it pays to validate AHT before feeding it into the interval-by-interval staffing calculation or into agent occupancy. If after-call work looks excessive, check how the wrap-up time is configured: a fixed timer that runs out unused inflates AHT with no real work behind it.

Common mistakes and how to spot them

Estimate call-handling staff by time slot: table 4
MistakeSymptomCheck
Using only answered callsThe forecast balances, but waiting and abandonment do not fallAdd abandoned and out-of-hours calls to demand
Forgetting after-call work"Free" agents who do not pick up callsMeasure the time between hanging up and becoming available
Deducting breaks twiceHeadcount inflated compared with real attendanceCheck that dedicated share and occupancy do not include the same meeting
Slots that are too longQueues at a particular time even though the slot looks comfortableRepeat the calculation with 15-minute intervals
Mixing enquiries of very different lengthStable average with unpredictable waitsCalculate each type separately and add the workload
Counting people who cannot handle that queueTransfers and repeat calls on the same dayCount only those who can resolve that enquiry

To check whether the result holds, measure the service level with the same definition you used when planning, and separate the abandonment rate from technical failures.

Turn the estimate into a testable shift

  1. Draft a week with the people actually available in each slot.
  2. Identify where another team can provide cover and which enquiries it can handle.
  3. Set aside time to return pending calls and complete the after-call work included.
  4. Compare forecast workload, calls offered, calls answered and observed waiting.
  5. Adjust handling time, dedicated share and cover when the data contradicts the assumption.

A phone system can help you see activity and distribute calls, but it does not create human availability. Define the shift first; then configure a queue with an exit route (in Spanish). If demand comes from campaigns, also check their ad schedule (in Spanish).

From the author's experience

After eleven years supporting and implementing contact centre software, before calculating headcount I would confirm whether the exported time includes waiting, talk time and after-call work, and whether there is one row per call or per agent segment, because transfers can duplicate records and skew any average. I would not quote a fixed percentage for after-call work without measuring the process: when there is no logged ACW state, I would sample a set of completed cases with start and end timestamps, separating conversation, follow-up and note-taking.

Erlang C can point you in the right direction, but with two or three agents, breaks, absences and skill differences carry real weight; I would always check it against real schedules and traffic rather than treating the mathematical result as a final headcount. For the planning interval, I would start with thirty minutes for a low-volume small business and check peaks with fifteen-minute intervals when there is enough data, since a full hour can hide call concentrations that cause waiting.

For Spain's Law 10/2025, from a technical standpoint I would prepare reports on time-to-answer, excluded calls and compliance by time slot, after first confirming the legal scope with proper advice. The first thing I would check for a staffing error is whether the real distribution of arrivals by time slot and the team's availability states have been reviewed, because sizing from daily averages can hide the fact that people are only missing during specific peaks.

Sources and limitations

Documentary review: . Content type: Calculation with example.

How to report a correction

Check current terms

Consider these options if they solve the problem described. Confirm features, limits and availability in your country.

AircallOfficial plans and termsCloudTalkOfficial plans and terms

Estos enlaces llevan a las páginas oficiales de cada proveedor. CallsIQ no usa enlaces de afiliación. Our commercial policy.

How this guide was prepared

Official sources, explained calculations and clearly labelled examples. Read about our methodology and use of AI in writing.