Call center service level is the share of calls answered within a set number of seconds, written as two numbers: 80/20 means 80 percent of calls answered within 20 seconds. The basic formula is calls answered within the threshold divided by calls offered, measured for each half hour. Platforms and contracts disagree about what counts as offered and what happens to callers who hang up, so the definition has to be written down before anyone reports the number.
This guide covers four common ways the formula is set up, where the 80/20 convention comes from, what one agent and 30 seconds of handle time do to service level, why a monthly figure hides the half hours that failed, and the terms a call center SLA needs so both sides read the same number.
What is service level in a call center?
Service level measures how quickly callers reach a person: the fraction of calls whose wait stayed under a threshold you choose. The research literature also calls it the telephone service factor (TSF), and notes that "typically the target is 20 or 30 seconds" (Gans, Koole and Mandelbaum, 2003). Platforms report it per interval, commonly each half hour.
A service level agreement (SLA) is a different thing. It is the commitment that contains a service level target, along with how it is measured, which hours and queues it covers and what a miss triggers. When a BPO runs calls for a client, the SLA is where that target lives; an in-house center may agree an internal one with the business. Service level is also a different number from average speed of answer and from the share of callers who waited at all, and the call center metrics guide shows how those three read together.
How do you calculate call center service level?
Divide the calls answered within the threshold by the calls offered in the same interval, then multiply by 100. With a 20 second threshold, 300 calls answered inside 20 seconds out of 400 offered is a 75 percent service level. The disagreements start with the denominator and with callers who hang up.
Take one half hour on a queue with a 20 second threshold. 400 calls entered the queue. 300 were answered within 20 seconds and 60 after it, 25 callers hung up inside 20 seconds, and 15 hung up later.
| Definition | Calculation | Service level |
|---|---|---|
| Answered within 20 seconds, out of calls offered | 300 ÷ 400 | 75.0 percent |
| Answered within 20 seconds, out of calls answered | 300 ÷ 360 | 83.3 percent |
| Answered within 20 seconds, out of calls offered minus hang-ups inside 20 seconds | 300 ÷ 375 | 80.0 percent |
| Answered or hung up within 20 seconds, out of calls offered | 325 ÷ 400 | 81.3 percent |
Against an 80 percent target, this half hour fails under the first definition and passes under the other three. The second leaves out every caller who gave up, so it reads highest exactly when abandonment is worst; the metrics guide walks through that gap on a full interval of figures.
The fourth is how Amazon Connect computes its Service level X metric: contacts removed from the queue within X seconds, divided by contacts queued. A contact leaves the queue when an agent answers it, when the customer abandons, or when the customer requests a callback (Amazon Connect metric definitions). X is picked from preset values between 15 and 600 seconds, and reports can add custom ones. With a 20 second threshold, a caller who hangs up after 10 seconds counts toward that service level. Find out which definition your platform uses before you compare its figure with anyone else's.
What is the 80/20 rule in a call center, and is it the standard?
The 80/20 rule is a service level target of 80 percent of calls answered within 20 seconds. It is the most quoted target, and the research literature uses it as its worked example, but it is a convention rather than a figure anyone calculated for your center. The same literature describes common practice as one where "upper management decides on the desired service level," and call center managers then defend the budget it takes (Gans, Koole and Mandelbaum, 2003).
COPC, which publishes performance standards for customer experience operations, suggests working backward from abandonment. Decide the abandonment rate you can live with, which it puts "in the range of three to five percent for most customer service environments," and then read the service level that goes with it from your own queue's history (COPC, 2024). That relationship differs by line. COPC's example is an airline: a caller rebooking a flight canceled by weather is more willing to wait than one asking about seats or meals.
COPC also argues for a band instead of a floor, for example 78 to 85 percent instead of "above 80." Its reason is cost: over-servicing "drives higher costs to the business and will lead to lower occupancy levels," and callers do not necessarily value the faster answer. It suggests a wider band for a low-volume queue, since consistent performance is harder to hold on a small one.
How many agents does a service level need?
More than a simple ratio suggests, and the relationship is steep near the target, where one agent moves service level by several points. In a half hour with 180 calls, a 5 minute average handle time and an 80/20 target, Erlang C puts the requirement at 36 agents on the phones. With 35, the queue misses by less than half a point.
| Agents on the phones | Service level (answered within 20 seconds) | Average speed of answer | Occupancy |
|---|---|---|---|
| 33 | 59.8 percent | 49 seconds | 90.9 percent |
| 34 | 71.2 percent | 28 seconds | 88.2 percent |
| 35 | 79.6 percent | 17 seconds | 85.7 percent |
| 36 | 85.8 percent | 11 seconds | 83.3 percent |
| 37 | 90.3 percent | 7 seconds | 81.1 percent |
| 38 | 93.4 percent | 4 seconds | 78.9 percent |
The 34th agent adds 11.4 points of service level. The 38th adds 3.1. Across the whole table occupancy moves 12 points while service level moves 34, so one short-staffed half hour does a lot of damage, and the last few points above target are the most expensive ones you can buy, which is the cost COPC's band is meant to cap.
The numbers come from the Erlang C formula as the research literature writes it: service level = 1 − C(N, R) × e^(−(N − R) × T ÷ AHT). N is the number of agents, R is the offered load in Erlangs (calls per second times handle time in seconds), T is the threshold in seconds, and C(N, R) is the probability that a caller waits at all (Gans, Koole and Mandelbaum, 2003).
Our Erlang C calculator runs the same math on your own volume; 360 calls an hour at 5 minutes reproduces the 36 agents above. It also applies shrinkage to get from agents on the phones to agents on the schedule, which the workforce management guide works through. Erlang C assumes every caller waits until answered. Real callers hang up, so read each figure as an approximation.
What moves service level besides headcount?
Handle time and volume move service level as much as staffing does, and unevenly: missing on the high side costs far more than the same miss on the low side gains. With the 36 agents from the table above, 30 extra seconds of handle time or 10 percent more calls than forecast takes service level below 60 percent. These are computed examples, not benchmarks or results from any deployment.
| What changes in the half hour | Service level with 36 agents | Agents needed to hold 80/20 |
|---|---|---|
| Nothing: 180 calls, 5:00 handle time | 85.8 percent | 36 |
| Handle time 30 seconds shorter | 96.5 percent | 32 |
| Handle time 30 seconds longer | 57.7 percent | 39 |
| 18 more calls than forecast (10 percent) | 58.5 percent | 39 |
| 18 fewer calls reach this queue (10 percent) | 96.3 percent | 32 |
| 36 fewer calls reach this queue (20 percent) | 99.3 percent | 29 |
Handle time is a staffing input. A change that adds 30 seconds to every call, such as a new verification step or a slower screen, costs three more agents in this half hour, or 28 points of service level if staffing stays put. The average handle time guide covers what moves it, after call work included.
A forecast miss is lopsided in the same way. Coming in 10 percent under forecast lifts service level by 10.5 points, while 10 percent over cuts it by 27.3. An SLA that holds a provider to 80/20 in every interval, with no allowance for volume above the client's forecast, is really asking for staffing well above that forecast. The research literature says as much: when volume is random, a fixed staffing level "will only achieve the desired TSF a fraction of the time" (Gans, Koole and Mandelbaum, 2003).
Calls that leave the queue have the same effect as added staff: anything that goes to self-service, a callback, an overflow destination or an option in the menu is a call this queue's agents do not take. An overflow rule tied to the wait sends calls elsewhere only while the queue is behind, so it can protect the peak half hours without staffing every interval for the worst case. Calls that come back, such as a transfer from the overflow destination or a callback that turns into a live call, land on the same agents, so keep them in the volume. The call center call flow guide shows where the wait limit and the overflow branch sit.
Why does a monthly service level hide bad half hours?
A monthly figure is an average weighted by calls, so a few heavy half hours that miss can sit under a total that looks fine. The same holds for a week or a single morning. COPC makes the point about month-end numbers, where poor performance early in the month "can be outweighed by over-delivery at the end of the month to 'make the number'" (COPC, 2024).
Here is a morning on one queue, with each half hour staffed to the 80/20 requirement at its forecast volume and the same 5 minute handle time. The morning's total came in 7 percent under forecast, but the two peak half hours came in over it.
| Half hour | Forecast calls | Agents on the phones | Actual calls | Service level |
|---|---|---|---|---|
| 9:00 | 120 | 25 | 100 | 97.8 percent |
| 9:30 | 150 | 30 | 125 | 97.8 percent |
| 10:00 | 180 | 36 | 195 | 64.7 percent |
| 10:30 | 180 | 36 | 200 | 53.8 percent |
| 11:00 | 160 | 32 | 140 | 96.6 percent |
| 11:30 | 150 | 30 | 128 | 97.1 percent |
| 12:00 | 140 | 28 | 120 | 96.3 percent |
| 12:30 | 120 | 25 | 104 | 96.5 percent |
| Morning | 1,200 | 1,112 | 83.6 percent, weighted by calls |
The morning report shows 83.6 percent, over target. Yet 395 callers, more than a third of the morning, arrived in the two half hours that ran at 64.7 and 53.8 percent. Not one half hour landed inside a 78 to 85 percent band: six were over-served and two were under-served.
Report service level per interval, and add a second number that counts intervals: the share of days, and later half hours, that landed in your band. COPC recommends that progression, from an end-of-month figure to daily targets and then to intervals, with the closest attention on the "prime intervals" that carry most of the volume. The schedule adherence guide shows how to check whether the agents the plan called for were actually on the phones in those half hours.
What should a call center SLA say about service level?
A service level clause needs more than "80/20." It should name the threshold and target, the denominator, how hang-ups and callbacks count, the measurement interval and how intervals roll up, the hours and queues covered, what happens when volume runs over forecast, and which report is the record. A term left out tends to come back in the first bad month as an argument about the definition.
Use the worksheet below whether you are the BPO writing the clause, the client reading it, or an in-house team agreeing a target with the business.
| Term | What to write down | What goes wrong when it is missing |
|---|---|---|
| Threshold and target | The seconds and the percentage, such as 80 percent within 20 seconds, or a band such as 78 to 85 percent | There is nothing to measure against, or a floor rewards over-staffing |
| Denominator | Calls offered, calls answered, or calls offered minus short abandons | The same half hour reads anywhere from 75.0 to 83.3 percent, as in the example above |
| Hang-ups | The short-abandon cutoff in seconds, and whether callers who hang up inside the threshold count as met, missed or excluded | A platform that counts them as met, as Amazon Connect's Service level X does, reports a higher figure than one that does not |
| Callbacks | Whether a caller who accepts a callback counts as answered in time, and the deadline for returning the call | A quick callback offer stops the clock while the caller still waits |
| Interval and roll-up | Per half hour, per day or per month, and whether the result is weighted by calls or counts the share of intervals in the band | A strong afternoon or a quiet week covers the half hours that failed |
| Hours and queues | Which queues and hours are covered, and whether after-hours and overflow calls are included | Calls sent to another destination show up in one side's report and not the other's |
| Calls handled outside the queue | How self-service, a menu option answered by an AI agent and overflow destinations count: as answered, excluded, or reported on their own | One side counts an immediate answer from a menu option as service level and the other leaves it out |
| Forecast tolerance | What happens to the target in an interval where offered calls run more than an agreed margin above the client's forecast | The provider carries forecast error it did not make; 10 percent over forecast took the example half hour from 85.8 to 58.5 percent |
| Companion metric | An abandonment rate target next to the service level | Under a definition that counts hang-ups as met, callers giving up can hold the service level up |
| Exclusions | Carrier and platform outages, and how an excluded interval is reported | Every bad day is argued over |
| System of record | The platform, the report, its threshold and denominator settings, and access to interval data for both sides | Two dashboards produce two numbers |
| What a miss triggers | How many intervals or days below target count as a miss, and what follows: a review, a recovery plan or a fee adjustment | Misses get reported and nothing changes |
If you run calls for several clients, keep a separate clause and a separate report for each client program, because each one has its own queues, hours and volume. The outsourcing comparison covers what a client is buying when it pays a provider to hold a service level.
How Telvana fits into a service level plan
Telvana is an AI agent platform for contact centers, and it touches service level in two of the places above. As an option in your phone menu, such as "Press 1 to talk to our AI assistant now," it answers callers who choose it right away, with no hold, while callers who would rather wait for a person keep their place in line. If your IVR can play a prompt based on queue time, you can offer the option only when the wait climbs (hold-queue relief). As a forwarding destination for overflow and after hours, it answers the calls your queue can't get to (after hours and overflow).
When a caller needs a person, the agent transfers the call to the number you choose. A transfer is an ordinary call, so point it at a priority queue or a direct line if your phone system has one, and count those calls in the queue's volume. Agree how menu-option and overflow calls count in your SLA before the first live call, using the worksheet row above. Our team builds the agent from your scripts, FAQs, policies and transfer rules, then tests it on your call types, and you approve it before it takes a live call. Pricing starts at $399 a month based on call volume, plus a one-time setup fee (pricing).
Frequently asked questions
What is service level in a call center?
Service level is the percentage of calls answered within a set threshold, such as 80 percent within 20 seconds. Call centers calculate it for each half hour and use it to decide how many agents to schedule. It measures how quickly callers reach a person, not whether their issue was resolved.
What is the formula for call center service level?
Service level equals calls answered within the threshold divided by calls offered, times 100. Some platforms and contracts divide by calls answered instead, exclude short abandons, or count callers who hang up inside the threshold as met, so check the definition before comparing figures. To predict service level from staffing, use Erlang C.
What is the difference between service level and an SLA?
Service level is the measurement: the share of calls answered within the threshold. A service level agreement is the commitment that contains the target, along with how it is measured, which queues and hours it covers and what a miss triggers. One SLA can hold other targets too, such as an abandonment rate.
Do abandoned calls count in service level?
It depends on the definition. Measured against calls offered, a caller who hangs up counts as a call not answered in time. Some setups exclude short abandons below a cutoff, and Amazon Connect counts a caller who hangs up within the threshold as removed from the queue in time, which raises the figure.
What is the difference between service level and average speed of answer?
Service level counts how many callers were answered within the threshold. Average speed of answer is the mean wait of the calls that were answered. A good average can hide a tail of very slow answers, and service level ignores how long the slow ones waited, so read both alongside abandonment.
How many agents does it take to raise service level?
It depends on where you start. In this guide's example, 180 calls in a half hour at 5 minutes each, the 34th agent added 11.4 points and the 38th added 3.1. Run your own volume and handle time through an Erlang C calculator to see the curve for your queue.
Should service level be measured daily or monthly?
Measure it per interval and report it at least daily. A monthly figure is weighted by calls, so a strong stretch can cover the half hours that failed. COPC recommends tracking the share of days, and then intervals, that land in a target band instead of a single end-of-month number.
Bring your busiest half hours to a demo
Book a demo and bring an interval report for your busiest day: calls offered, handle time and service level by half hour. We'll show where a Telvana menu option or overflow rule would sit in your call flow, and you can hear the agent take a call.