Average handle time: the metric most teams optimize backwards

What AHT actually measures

Average handle time is total talk time plus total hold time plus total after-call work, divided by the number of contacts handled. All three components matter, and teams that measure only talk time systematically understate their staffing requirement.

The pieces behave differently and should be read separately before they're added together. Talk time is a function of contact complexity and agent skill. Hold time is almost always a systems or authority problem — the agent is waiting on something. After-call work is a tooling problem more often than a diligence problem: if wrap-up takes ninety seconds, someone is re-typing information the system already had.

Two definitional traps distort the number badly enough to break staffing models. First, decide explicitly whether transfers count once or twice — counting the same customer as two handled contacts flatters both AHT and volume. Second, be consistent about whether abandoned-after-answer contacts and misrouted calls enter the denominator. Neither choice is wrong; inconsistency between reporting periods is.

Why a target on AHT alone makes service worse

AHT is the most gameable number in the contact center, and every method of gaming it is invisible in the AHT report itself.

An agent under handle-time pressure has a menu of options that all work: transfer the hard contact to someone else, close it before the customer is finished, promise a callback that becomes another team's problem, or simply skip the second question the customer hadn't asked yet. Each of these lowers AHT and raises repeat contact rate — and because the repeat arrives as a fresh contact, total cost goes up while the metric goes down.

This is why handle time should never sit alone on a scorecard. The honest framing is that AHT is only meaningful paired with first-contact resolution and quality. A ten-second AHT reduction alongside a three-point FCR drop is a cost increase wearing a cost reduction's clothes.

There is no universal benchmark, and chasing one is a mistake

Published AHT benchmarks are close to useless because handle time is dominated by contact mix, not by team quality. A password reset queue and a claims intake queue in the same building can differ by a factor of five, and neither number tells you anything about the other.

The comparisons worth making are internal. Compare the same queue to itself over time, compare agents handling the same contact types to each other, and compare handle time by contact reason to find the outliers. A single contact reason consuming disproportionate minutes is a process finding, not a coaching finding.

The one external comparison with some value is your own AHT against what a partner or vendor quotes for the same work, and even that requires nailing down definitions first — whether after-call work is included, and how transfers are counted, will swing the comparison more than any real operational difference.

Diagnose before you cut: where the seconds actually go

Most AHT-reduction programs fail because they start with a target instead of a measurement. Before setting any number, break the average into its parts and find where time is genuinely being lost.

• Authentication — how long from hello to verified? Multi-step verification against systems that don't talk to each other routinely eats 45–90 seconds of every contact, and it's the least visible chunk because everyone assumes it's fixed cost.
• Search and lookup — time the agent spends hunting across systems. If they need three applications to answer one question, the tooling is generating handle time you're paying for on every contact.
• Hold — every hold is a question the agent couldn't answer or an action they weren't authorized to take. Categorize hold reasons for two weeks and the top three usually account for most of it.
• Transfers — each transfer adds its own handshake and often a re-explanation. High transfer rates in a queue mean the routing or the skilling is wrong, not that agents are slow.
• After-call work — watch what agents type in wrap-up. Anything that duplicates data already in the system is an integration task disguised as an agent-behavior problem.
• Explanation loops — listen for the moment a customer says 'sorry, can you say that again' — repeated explanation usually signals a genuinely confusing policy, which is a product fix with much bigger returns than a coaching one.

The reductions that don't cost you resolution

Once you know where the seconds are, the fixes fall into three tiers by durability.

The most durable are the ones that remove work rather than speed it up: single sign-on and a unified agent desktop so lookups stop costing thirty seconds each; screen pops that carry caller identity and recent history so authentication starts half-solved; expanded agent authority so the refund or credit that currently requires a supervisor hold requires nothing; and templated wrap-up that auto-populates from the interaction rather than from memory.

The second tier is knowledge: a searchable, current knowledge base cuts both hold time and the guess-then-correct loop. Agent-assist tooling that surfaces the likely answer during the conversation belongs here too — it reliably reduces search time, though it does very little for genuinely complex contacts, and vendors tend to price it as though it does.

The third tier is deflection, which reduces average handle time by changing the mix rather than the work: moving simple contacts to self-service raises AHT on the remaining queue while lowering total cost. This is the case where a rising AHT is good news, and it's worth saying out loud before the number moves, because otherwise it reads as a regression.

AHT is the input to your staffing math

The operational reason to care about handle time has little to do with agent performance. AHT and contact volume together produce workload, and workload drives every staffing decision you make.

Because staffing curves are non-linear, small handle-time errors compound. A forecast built on an AHT that is thirty seconds optimistic will under-staff, which lengthens queues, which raises handle time as agents deal with customers who waited — the error feeds itself. This is also why AHT belongs in the forecast by interval and by queue, not as a single site-wide figure: an average across a mixed contact base describes no actual half-hour.

Feed the number back honestly. If a process change genuinely removed forty seconds, update the forecast rather than quietly banking the capacity, and if a new product launch added ninety, say so before the service level slips.

Reporting it without creating the wrong behavior

Publish AHT to supervisors and planners, not to wallboards. Coach on the components — this agent's hold time is double the queue median, let's find out why — rather than on the total, because the total invites the shortcuts. Pair it permanently with first-contact resolution, repeat contact rate within seven days, and quality score, and make it a standing rule that no AHT improvement counts unless those three held. And when it drifts up, check the mix before you check the people: a rising AHT after a self-service launch, a product change, or a new contact type is usually the system working exactly as designed.