Contact Centre Metrics That Actually Matter

A contact centre report that says "average wait 42 seconds, 91% of calls answered" sounds healthy and can describe a queue where a hundred people a week hang up after four minutes. Averages are the problem: they are dominated by the many easy calls and say nothing about the tail, and the tail is where your complaints, your churn and your reviews come from. This is a practical guide to the six measurements that carry real information — service level, average speed of answer, abandonment, first contact resolution, handle time and occupancy — what each one actually measures, what Australian benchmark data says good looks like in 2026, the four ways these numbers routinely mislead, and how to turn each symptom into a specific change rather than a target on a wall.

Contact Centre · Measurement · 2026

Contact Centre Metrics That Actually Tell You Something

Most contact centre reporting is built from averages, and averages conceal exactly the customers you are failing. Six metrics carry real information, four of them mislead if you read them the usual way, and one number almost nobody looks at explains more than the rest combined.

📅 ⏱ 16 min read 🇦🇺 Australian owned, Australian hosted, Australian supported
TL;DR

Six metrics matter, four of them lie if read the usual way, and the one nobody looks at is the tail. Service level (the share of calls answered within X seconds) is the only headline number that describes an experience rather than an average — 80/20 became the default target through history rather than analysis, and should be chosen deliberately. Average speed of answer is useful for capacity and misleading for experience; read the 90th percentile instead. Abandonment is the closest thing to a direct revenue measure: Australian self-reported figures for 2026 sit around a 9% average and 5% median, and the gap between those two tells you the distribution is skewed. First contact resolution is the hardest to measure and the most valuable — ACXPA rankings put banking near 32% in early 2026. Average handle time should be a diagnostic, never a target. Occupancy is a ceiling, not a goal: sustained operation above roughly 85% buys short-term throughput and pays for it in attrition. And measure by interval, not by day — a good daily number routinely hides a broken hour.

Why Your Current Report Looks Fine

Consider a queue that took 500 calls last week. 450 were answered in under 15 seconds. 50 waited more than four minutes and 30 of those hung up.

How it reportsWhat it saysWhat it hides
Average speed of answerAbout 40 seconds. Sounds fineThat 450 people waited almost nothing and 50 waited a very long time. The average describes nobody's actual experience
Calls answered94%. Sounds excellent30 people gave up. In most businesses that is the most commercially significant fact of the week
Average handle timeSteady. No signalNothing at all about whether the problem was solved

Nothing in that report is wrong. It is simply built from averages, and an average is dominated by the majority of easy calls. The customers you are failing are, by definition, a minority — so they are invisible in every mean you calculate.

The reframe that fixes most reporting

Stop asking “how did the queue perform?” and start asking “how many people had a bad experience, and when?” Every recommendation below follows from that change of question. It is also the question a manager can act on, whereas “average wait rose four seconds” is not actionable by anyone.

The Six That Matter

Modern platforms will report dozens of things. Six carry information you can act on; the rest are mostly decoration or inputs to these six.

MetricWhat it measuresRead it as
Service level% of calls answered within X secondsYour headline. The only common metric that describes an experience rather than an average
Average speed of answerMean wait before answerA capacity input. Poor for judging experience — use percentiles for that
Abandonment rate% of callers who hang up while waitingThe closest thing to a direct revenue measure you have
First contact resolution% resolved without the customer coming backThe most valuable and the hardest to measure honestly
Average handle timeTalk time plus after-call workA diagnostic and a capacity input. Never a target
Occupancy% of logged-in time spent on contactsA ceiling to stay under, not a number to maximise

Two things deliberately absent. Calls answered per agent per day is a productivity number that rewards rushing and punishes the person who handles the hard calls well. And average call duration in isolation tells you nothing without knowing whether the issue was resolved — a shorter call that produces a callback tomorrow is more expensive, not less.

Service Level, and Where 80/20 Came From

Service level is expressed as two numbers: the percentage of calls answered, and the threshold in seconds. 80/20 means 80% of calls answered within 20 seconds. In Australia both 80/20 and 80/30 are common defaults.

Worth knowing where that came from, because it changes how much authority you give it. 80/20 is not the output of research into customer tolerance. It emerged decades ago as a workable planning convention and hardened into a default through repetition. Treating it as a universal standard is a mistake — the right threshold depends entirely on what your callers are ringing about.

QueueReasonable targetReasoning
Sales and new enquiriesAggressive — 80/10 or betterThe caller is comparing you with someone else right now. This is the queue where waiting costs revenue directly
Emergency or fault reportingAggressive, and measure the tail hardThe consequence of a long wait is not annoyance
General service80/20 to 80/30 is defensibleThe conventional default, and reasonable when calls are routine
Complex case work, applications, claimsRelaxed threshold, high percentageCallers will wait for something that matters to them. What they will not tolerate is waiting and then not being helped
Back office and internalDo not set oneMeasuring it produces the appearance of rigour and changes nothing
The most common way service level gets gamed

Answering quickly and then parking the caller. The service level target is met at the moment of answer, so a queue can report 85/20 while callers routinely spend six minutes on hold after being greeted. If your service level is excellent and your abandonment is also high, this is the first thing to check — measure hold time after answer as a separate number, and look at where those abandons occur in the call.

Average Speed of Answer vs the 90th Percentile

ASA is the mean wait before a call is answered. It is genuinely useful for capacity planning and genuinely misleading for judging customer experience, and most reporting uses it for the second purpose.

Australian benchmark data illustrates the spread. ACXPA's 2026 Australian Contact Centre Best Practice Report puts self-reported speed of answer across a very wide range by sector — utilities reported the slowest at a 227-second average, while banking and finance carried the highest median at 79 seconds. The distance between sectors is large, and the distance between a sector's average and its median is itself informative.

Ask your platform for one number instead

The 90th percentile wait. “90% of callers waited less than X.” That single figure tells you what your worst-served tenth actually experienced, and it is the population that generates complaints, escalations, negative reviews and churn. A queue with a 40-second ASA and a 6-minute 90th percentile is a materially different business problem to one with a 40-second ASA and a 90-second 90th percentile — and the two are indistinguishable on an average.

If your reporting cannot produce percentiles, that is a real limitation worth raising with your provider. It is standard capability in a current platform, and the absence of it forces you to manage a distribution using only its mean.

Abandonment: The Closest Thing to Revenue

An abandoned call is a person who wanted to speak to you badly enough to ring, waited, and gave up. In a sales queue that is lost revenue outright. In a service queue it is a customer who is now more likely to leave, more likely to complain publicly, and quite likely to ring again — meaning you carry the cost twice.

ACXPA's 2026 self-reported Australian figures put voice abandonment at an average of 9% and a median of 5%. That gap is the interesting part: it means a minority of operations with very high abandonment are dragging the average well above what a typical centre experiences. If you are above 9% you are not near the middle — you are in the tail.

Before comparing yourself to anythingWhy
Check how short abandons are countedMany platforms exclude calls abandoned within the first 5–10 seconds, on the reasoning that those are misdials. That is defensible, but it changes the number materially and makes cross-organisation comparison unreliable unless both count the same way
Check whether callbacks count as abandonsIf a caller accepts a callback offer and hangs up, some systems record an abandon. That is a successful interaction being recorded as a failure
Look at when they abandon, not just how manyAbandons clustered at 30 seconds mean your greeting or menu is losing people. Abandons clustered at 5 minutes mean your capacity is wrong. Completely different fixes

That third row is the most actionable measurement on this page, and almost nobody produces it. A histogram of abandonment by wait time tells you which problem you have. Early abandons are a design problem — too many menu options, an over-long greeting, no indication that a human exists. Late abandons are a resourcing problem, and no amount of menu redesign will touch them.

First Contact Resolution: Hard, and Worth It

FCR is the share of contacts resolved without the customer having to come back. It correlates more strongly with satisfaction and with cost than anything else here, and it is the least well measured because measuring it honestly is genuinely difficult.

For scale: ACXPA's Australian Call Centre Rankings put the banking sector at roughly 32% first contact resolution in the first quarter of 2026. Two-thirds of contacts in that sample were not resolved first time. Whatever your own number is, if you have never measured it, it is likely lower than you assume.

📊

The repeat-contact method

Count contacts from the same customer within 7 days about the same issue. Objective and automatable, but it needs the phone system and the CRM to be linked, and it misses the customer who gave up rather than ringing back.

🙋

The agent-declared method

The agent marks whether it was resolved. Cheap and immediate, and systematically optimistic — people are poor judges of whether they solved someone else's problem.

💬

The ask-the-customer method

A one-question survey after the call. The most accurate and the lowest response rate, and the responses skew toward the very satisfied and the very unhappy.

Use the repeat-contact method as your tracked number, because it is the only one that is consistent over time, and sample the other two occasionally to check it is not drifting. What matters far more than the absolute figure is the trend and the breakdown by contact reason — a low overall FCR is usually two or three specific issue types dragging everything down, and those are fixable once identified. AI transcription and call summarisation make the reason breakdown considerably cheaper to produce than it used to be; AI transcription and CRM notes covers how.

Can your current system answer these questions?

90th percentile wait. Abandons by wait-time bucket. Service level by half-hour interval. Repeat contacts within seven days. If the answer is no, the problem is the reporting, not your team. We will show you what a current platform reports as standard — and what it changes when a manager can see it.

See the Reporting Or call 1300 881 662

Handle Time and Occupancy: Diagnostics, Not Targets

Two metrics that are useful when watched and destructive when targeted.

Average handle time is talk time plus after-call work. It belongs in capacity planning, where you genuinely need it. It does not belong on a wall, because the moment it becomes a target the behaviour it produces is rushing — and rushing reduces first contact resolution, which increases repeat contacts, which increases total handle time across the operation. You end up paying more for a worse experience while the metric improves.

The rule

Watch AHT for changes and for outliers. A sudden rise means a new issue type, a system problem, or a process change. A persistent gap between two agents doing the same work is a coaching conversation. An organisation-wide reduction target is a request for worse service, and it is almost always answered.

Occupancy is the share of logged-in time spent handling contacts. Intuitively, higher looks like better utilisation. In practice it behaves like a physical limit: sustained operation above roughly 85% produces measurable degradation — errors rise, courtesy drops, sick leave rises and attrition follows. Because queues are stochastic, a team running at very high occupancy also has no absorption capacity, so a small spike in volume produces a disproportionate collapse in service level.

The related figure worth knowing is shrinkage: the proportion of paid time not available for contacts — breaks, training, meetings, leave, coaching. It is commonly 30–35% and is routinely underestimated when rosters are planned, which is the single most frequent cause of a queue that is inexplicably short-staffed every afternoon.

Read the Interval, Not the Day

The most common reporting error, and the easiest to correct.

A queue reporting 82% service level for the day looks like it hit an 80% target. Break the same day into half-hour intervals and the usual pattern appears: 95% for most of the day, and 45% between 9:00 and 10:00, and again between 1:00 and 2:00. Every customer who rang in those two windows had a poor experience, and the daily number concealed all of them.

Look atWhat it reveals
Service level by half-hourThe two or three windows where your queue actually fails. Almost always the morning open and immediately after lunch
Volume by half-hour, by weekdayMonday is not Wednesday. Rosters built on a daily average under-staff Monday morning every week without fail
Abandons by intervalConfirms which windows are hurting people rather than merely being busy
Occupancy by intervalShows where the team is being asked to run at a rate no team sustains
The cheapest fix in contact centre management

Move breaks and meetings out of your two worst intervals. It costs nothing, requires no headcount, and frequently produces a larger improvement in service level than anything else available. Most teams schedule the morning meeting at 9am and the team huddle right after lunch — precisely the two windows that are already failing.

From Symptom to Change

A number is only useful if it points to an action. This table maps what you see to what is usually causing it.

What you seeUsually meansDo this
Good service level, high abandonment Callers are answered quickly then parked, or the menu is losing them before the queue Measure hold time after answer separately. Plot abandons by wait time. Shorten the greeting
Abandons cluster under 30 seconds Design problem — menu too deep, greeting too long, no sign a human exists Cut menu options to three or four. Put the option to reach a person early
Abandons cluster past 3 minutes Capacity problem, in specific intervals Re-roster to the interval curve. Offer callback rather than making people wait
Daily service level fine, customers complaining Interval failure hidden by a daily average Report by half-hour. Move breaks and meetings out of the failing windows
Low FCR concentrated in a few reasons Knowledge, access or authority gap on specific issue types Fix those two or three flows. Give front line authority to resolve them without escalation
AHT falling, repeat contacts rising The team is being measured on speed and has responded rationally Remove AHT as a target. Measure resolution instead
Occupancy above 85% routinely Understaffed, or shrinkage was underestimated in planning Recalculate shrinkage honestly — it is usually 30–35%, not the 20% in the plan
Everything looks good, revenue queue underperforming Sales and service share a target that suits neither Split the queues and set an aggressive threshold on the revenue one
9% / 5%
AU abandonment: average / median, 2026
~32%
Banking FCR, Q1 2026
85%
Occupancy ceiling, not target
30–35%
Typical shrinkage

Benchmarks are context, not goals. A queue handling complex claims and one handling order status are not comparable, and a business that adopts a sector average as a target has borrowed someone else's problem. Measure your own distribution, watch the trend, and fix the two intervals where the failures actually live. That approach beats benchmark-chasing in every operation we have seen it applied to.

If you are building or rebuilding the queue itself rather than measuring an existing one, the contact centre software buyer's guide covers the capability set, one queue for voice, SMS and chat covers what changes when the same team handles more than calls, and AI call scoring covers the quality side that none of the metrics here measure.

Frequently Asked Questions

What contact centre metrics actually matter?
Six carry information you can act on. Service level, the percentage of calls answered within a set number of seconds, is the only common metric that describes an experience rather than an average, so it belongs as your headline. Average speed of answer is a useful capacity input but a poor guide to experience. Abandonment rate — the share of callers who hang up while waiting — is the closest thing to a direct revenue measure you have. First contact resolution is the most valuable and the hardest to measure honestly. Average handle time, meaning talk time plus after-call work, is a diagnostic and a capacity input but should never be a target. And occupancy, the share of logged-in time spent handling contacts, is a ceiling to stay under rather than a number to maximise. Two commonly reported figures are deliberately absent from that list. Calls answered per agent per day rewards rushing and punishes whoever handles the difficult calls properly. Average call duration on its own tells you nothing without knowing whether the issue was resolved — a shorter call that produces a callback tomorrow is more expensive, not less.
Is 80/20 the right service level target?
Not automatically, and it is worth knowing where it came from before granting it authority. Eighty per cent of calls answered within twenty seconds is not the output of research into customer tolerance; it emerged decades ago as a workable planning convention and hardened into a default through repetition. In Australia both 80/20 and 80/30 are common. The right threshold depends entirely on what people are ringing about. For sales and new enquiries, be aggressive — 80/10 or better — because the caller is comparing you with someone else at that moment and waiting costs revenue directly. For emergency or fault reporting, be aggressive and watch the tail closely, since the consequence of a long wait is not merely annoyance. For general service with routine calls, 80/20 to 80/30 is defensible. For complex case work, applications or claims, a relaxed threshold with a high percentage is more honest, because callers will wait for something that matters to them but will not tolerate waiting and then not being helped. For back office and internal queues, do not set one at all — it creates the appearance of rigour and changes nothing.
Why is average speed of answer misleading?
Because an average is dominated by the many easy calls and says nothing about the tail, and the tail is where complaints, escalations, negative reviews and churn come from. Take a queue that handled 500 calls where 450 were answered in under fifteen seconds and 50 waited more than four minutes. The average speed of answer lands around forty seconds, which sounds perfectly healthy and describes nobody's actual experience. Australian benchmark data shows how wide the spread runs: ACXPA's 2026 Australian Contact Centre Best Practice Report put self-reported speed of answer across a very broad range by sector, with utilities slowest at a 227-second average and banking and finance carrying the highest median at 79 seconds. Ask your platform for the 90th percentile wait instead — the figure that completes the sentence 'ninety per cent of callers waited less than X'. A queue with a forty-second average and a six-minute ninetieth percentile is a completely different business problem from one with a forty-second average and a ninety-second ninetieth percentile, and the two are indistinguishable on the mean. If your reporting cannot produce percentiles, that is a genuine limitation worth raising with your provider.
What is a good call abandonment rate in Australia?
ACXPA's self-reported Australian figures for 2026 put voice abandonment at an average of nine per cent with a median of five per cent. The gap between those two numbers is the informative part: it means a minority of operations with very high abandonment are pulling the average well above what a typical centre experiences, so if you are above nine per cent you are not near the middle, you are in the tail. Before comparing yourself to any benchmark, check three things. First, how short abandons are counted — many platforms exclude calls abandoned within the first five to ten seconds on the basis that they are misdials, which is defensible but materially changes the number and makes cross-organisation comparison unreliable unless both count the same way. Second, whether accepted callbacks are being recorded as abandons, because that records a successful interaction as a failure. Third, and most usefully, look at when people abandon rather than only how many. A histogram of abandonment by wait time tells you which problem you have: abandons clustered around thirty seconds are a design problem such as too many menu options or an over-long greeting, while abandons clustered past three or four minutes are a resourcing problem that no menu redesign will touch.
How do you measure first contact resolution properly?
There are three methods and each has a different weakness. The repeat-contact method counts contacts from the same customer within about seven days on the same issue — objective and automatable, but it requires the phone system and CRM to be linked and it misses the customer who gave up rather than ringing back. The agent-declared method has the agent mark whether the issue was resolved, which is cheap and immediate but systematically optimistic, since people are poor judges of whether they solved someone else's problem. The ask-the-customer method uses a one-question survey after the call, which is the most accurate but has the lowest response rate and skews toward the very satisfied and the very unhappy. Use repeat-contact as your tracked number because it is consistent over time, and sample the other two occasionally to check it has not drifted. For scale, ACXPA's Australian Call Centre Rankings put banking at roughly thirty-two per cent first contact resolution in the first quarter of 2026, meaning two-thirds of contacts were not resolved first time. What matters more than the absolute figure is the trend and the breakdown by contact reason, since a low overall number is usually two or three specific issue types dragging everything down.
Should we set a target for average handle time?
No. Average handle time belongs in capacity planning, where you genuinely need it, and it should be watched for changes and outliers — a sudden rise signals a new issue type, a system problem or a process change, and a persistent gap between two agents doing the same work is a coaching conversation. But the moment it becomes a target the behaviour it produces is rushing, and rushing reduces first contact resolution, which increases repeat contacts, which increases total handle time across the whole operation. You end up paying more for a worse customer experience while the metric on the wall improves. An organisation-wide handle time reduction target is effectively a request for worse service, and it is almost always granted. Occupancy deserves the same caution in the other direction. It measures the share of logged-in time spent handling contacts, and higher intuitively looks like better utilisation, but in practice it behaves like a physical limit — sustained operation above roughly eighty-five per cent produces rising errors, dropping courtesy, more sick leave and eventual attrition. Because queues are stochastic, a team at very high occupancy also has no absorption capacity, so a modest volume spike collapses service level disproportionately.
What is the single easiest improvement to make?
Report by half-hour interval rather than by day, then move breaks and meetings out of your two worst intervals. This is the cheapest fix available in contact centre management: it costs nothing, requires no additional headcount, and frequently produces a bigger improvement in service level than anything else on the list. The reason it works is that daily averages conceal interval failure. A queue reporting eighty-two per cent service level for the day looks like it met an eighty per cent target, but broken into half-hours the usual pattern emerges — ninety-five per cent for most of the day and forty-five per cent between nine and ten in the morning and again between one and two in the afternoon. Every customer who rang in those two windows had a poor experience and the daily figure hid all of them. Those two windows are also, almost universally, when teams schedule the morning meeting and the post-lunch huddle. Alongside that, look at volume by half-hour broken down by weekday, because Monday is not Wednesday and rosters built on a daily average under-staff Monday morning every single week. Also check your shrinkage assumption: the proportion of paid time unavailable for contacts is commonly thirty to thirty-five per cent and is routinely planned at twenty.

What to Read Next

Your next reads

Uniden Voice Over Cloud logo

Australia’s smartest AI-powered cloud phone system — Australian owned, Australian hosted, Australian supported. unidenvoice.com | 1300 881 662