Call centre metrics that actually change decisions

Contact-centre reporting has a specific failure mode: a dashboard with thirty numbers, none of which anyone acts on. The fix is not fewer metrics but understanding which ones are levers, which are outcomes, and which are traps when you optimise them directly.

Service level: the outcome everything else serves

The percentage of inbound calls answered within a target time, conventionally written 80/20: eighty per cent answered within twenty seconds.

service level = calls answered within target ÷ total calls answered × 100

80/20 is a convention, not a law. The right target depends entirely on what the caller is calling about. An emergency line and a billing enquiry line should not share one, and forcing them to is how you end up overstaffing the queue nobody minds waiting in.

Abandonment: the number service level hides

Inbound abandonment is the share of callers who hang up while waiting.

abandonment = calls abandoned in queue ÷ calls offered × 100

Most centres treat 2–5% as acceptable, but the useful cut is by wait time. Callers who abandon in the first five seconds mostly misdialled; those abandoning after ninety are the ones you are actually losing. Reporting a single blended figure buries that distinction.

Note that the same word means something completely different in outbound: a call the dialer dropped because no agent was free. That one is legally capped at 3% of live answers in the US; see abandonment rate and the TCPA guide. Be explicit about which you mean in any report that touches both.

Occupancy: the trap metric

The share of an agent's logged-in time spent handling contacts, including after-call work.

occupancy = (talk time + after-call work) ÷ logged-in time × 100

Most centres target 80–85%. It is the metric most likely to cause harm when optimised, because on a dashboard it looks like pure efficiency: 95% occupancy means agents are almost never idle.

In practice sustained occupancy above about 90% produces longer handle times, more errors, more sick days, and attrition, which is by far the most expensive line item a contact centre has. There is also a structural reason it cannot go to 100%: keeping queue waits short requires idle capacity by definition. Occupancy and service level trade directly against each other.

Adherence: did people work the schedule?

The share of scheduled time an agent was in the state the schedule called for: available when rostered available, on break when rostered on break.

adherence = time in adherence ÷ total scheduled time × 100

Worked example: an agent scheduled for eight hours (480 minutes) who was in adherence for 432 of them is at 90%.

Targets of 85–95% are normal. Chasing 100% is actively counterproductive: it punishes an agent for finishing a call that ran past the start of their break, which is exactly the behaviour you want. Most sensible schemes allow a few minutes of grace precisely so the metric does not fight the job.

Shrinkage: the number that breaks forecasts

The proportion of paid time not available for handling contacts: breaks, training, meetings, coaching, sickness, and holiday.

shrinkage = (scheduled hours − productive hours) ÷ scheduled hours × 100

Typical shrinkage runs 30–35%, and forecasts that ignore it are always short-staffed. The arithmetic compounds: needing ten agents on the phones at 35% shrinkage means scheduling about fifteen, not eleven.

agents to schedule = agents needed ÷ (1 − shrinkage)
15.4 = 10 ÷ (1 − 0.35)

The outbound equivalents

Dial count is the metric most often reported and least worth reporting. It rewards activity that produces nothing, and it goes up precisely when answer rates collapse.

Quality: stop sampling

The traditional approach (a supervisor scoring three calls per agent per month) has a statistical problem nobody enjoys naming: three calls cannot distinguish a good agent from a lucky one, and everybody involved knows the sample is not representative.

Transcript-based scoring changes the economics. Every call can be scored against the same checklist, which turns quality from an opinion into a distribution and makes coaching specific: not “be more empathetic” but “you asked for the appointment on 40% of qualified calls”.

Keep the scorecard under ten criteria, make every one observable from the recording, and agree the weighting before scoring starts rather than arguing about it afterwards.

A dashboard worth having

1. Service level and abandonment, side by side, by queue and by hour.
2. Occupancy, watched as a health indicator rather than a target to maximise.
3. Adherence, with grace built in and reviewed weekly rather than policed hourly.
4. Answer rate per outbound number, alerting on divergence.
5. Conversations per rep hour, not dials.
6. One quality score, from every call rather than a sample.

Six numbers. If a metric on your dashboard has never once caused someone to do something differently, it is decoration, and it is making the five that matter harder to see.

First contact resolution and average handle time

Two metrics that are almost always reported together and almost always misused together, because they pull in opposite directions.

Average handle time is talk time plus after-call work, averaged across contacts. It is a capacity input (you need it to forecast staffing) and it is a terrible performance target. Push AHT down and agents rush, resolve less, and generate repeat contacts, which raises total workload while the dashboard shows improvement.

AHT = (total talk time + total after-call work) ÷ contacts handled

First contact resolution is the share of contacts fully resolved without the customer having to come back. It is the metric AHT should be judged against: a longer call that resolves the issue is cheaper than two short ones that do not.

The measurement problem with FCR is defining the window. A common approach is to count any repeat contact from the same customer about the same issue within a set period (often a few days) as a resolution failure. The exact window matters less than applying it consistently, because the value of FCR is in the trend rather than the absolute number.

Why after-call work is the cheapest thing to fix

Wrap-up time is the most compressible number in the whole set, and it is usually the least examined. It sits inside AHT, it directly constrains dialer pacing, and a meaningful share of it is a person retyping something the system already knows.

Automatic transcription and summarisation routinely halve it, because the agent is editing a draft rather than composing a note from memory. One-key dispositions that trigger the follow-up automatically remove most of the rest.

The arithmetic is worth doing. Thirty agents, two hundred contacts each per week, ninety seconds of wrap-up saved per contact is roughly 150 hours a week returned, without changing headcount, targets, or anything a customer experiences. Very few operational changes have that ratio of effort to effect.