Average handle time is the wrong contract

an input that became a verdict

Average handle time started as an arithmetic convenience. If you know volume and you know how long a contact takes, you can staff the queue. Somewhere in the last two decades it stopped being an input to a model and became a target on a scorecard, and the moment a target is attached to a person, the behaviour changes. Representatives learn to close the interaction, not the case. The number improves. The operation does not.

This is not a morality tale about metrics gaming. It is a structural observation: AHT is a cost proxy that sits inside the contact center budget, while the cost it creates — the re-contact, the escalation, the churn — usually lands outside it. A leader who lowers AHT is rewarded locally and taxed globally, and the tax is invisible because nobody joins the two ledgers.

the automation version of the same mistake

Automation inherits whatever metric it was pointed at. Deploy agents against AHT and they will get very good at ending conversations quickly, because that is the objective you specified. Deploy them against resolution and containment of downstream contact, and you get slower interactions that cost less in total. Both programs will report success. Only one of them changed the economics.

The uncomfortable implication is that a well-designed agent program can make AHT worse on purpose. If the agent handles the short, mechanical contacts, the residual human population is by definition harder and longer. AHT rises. That is the correct outcome and it will look like a regression on every dashboard built before the program started.

the contract worth writing

Replace the AHT target with three commitments: resolution rate per customer per issue, cost per resolved issue including downstream contact, and a quality floor that cannot be traded away for speed. Keep AHT as a staffing input, where it belongs, and remove it from every scorecard that touches an individual.

Operations leaders resist this because AHT is the one number the whole organisation understands. That is precisely the argument against it. A metric that everyone understands and nobody can defend is not a control — it is a habit.

how a staffing input became a personal verdict

Trace the history and the drift is easy to see. Workforce management needs an average handle time to forecast staffing, so someone calculates it from historical data and builds a schedule around it. That is a legitimate, narrow use of the number. At some point a director asks why AHT varies by three minutes across two teams doing the same work, and the answer becomes a scorecard line, and the scorecard line becomes a coaching conversation, and the coaching conversation becomes a number every representative on the floor knows is being watched every single day.

Once a number is watched daily by someone whose bonus or continued employment depends on it, it stops measuring the underlying reality and starts measuring how well people can manage the number. This is not cynicism about frontline staff; it is the most predictable finding in the entire measurement literature, and contact centers have run the experiment at enormous scale without ever citing the result.

the ledger split that hides the cost

The reason this persists is that the two costs live in different budgets, reviewed by different people, on different calendars. AHT sits in the contact center's monthly operating review, reported in seconds, trending in the right direction, presented with pride. The re-contact it produces shows up as call volume next month, in a different report, attributed to demand rather than to last month's handling. Churn shows up in retention's quarterly numbers, attributed to competitive pressure or price. Nobody in the room owns the causal chain, so nobody questions the metric that started it.

We have seen operations run the actual join — following individual reps' AHT scores against the ninety-day re-contact and cancellation rate of the customers they handled — exactly once, usually after a new analytics leader arrives with no attachment to the existing scorecard. The correlation is rarely subtle. The reps with the best AHT scores frequently carry the worst downstream outcomes, because the fastest way to end a call is to tell the customer what they want to hear and let someone else deal with the consequence.

what changes when an agent is in the loop

This gets sharper, not softer, once automation enters the picture, because an agent will optimise ruthlessly against whatever objective it is given, without the human hesitation that sometimes limits the damage of a bad incentive. Point an agent at call duration and it will learn to truncate ambiguous requests, offer the first plausible answer, and end the interaction — all while producing a beautiful trend line. Point the same agent at resolution confirmed fourteen days later and it will take longer per interaction and produce a lower total cost of ownership for the same volume of problems.

The operational implication is that AHT should genuinely go up when an automation program is working correctly. The agent absorbs the short, low-complexity contacts that used to pull the average down, leaving a human queue that is, by construction, harder and longer than it used to be. Leaders who have not prepared their board for this will get asked to explain a regression that is in fact the entire point of the investment.

the contract to put in its place

Replace the individual AHT target with three commitments measured at the team and program level rather than the person level: resolution rate per customer-issue, fully loaded cost per resolved issue including everything the first contact generated downstream, and a quality floor — measured on the full population, not a sample — that cannot be traded for speed. Keep AHT alive exactly where it belongs, as a forecasting input for workforce management, and strip it from every individual scorecard within ninety days.

Expect resistance, and expect it to come from the people who currently score well. That resistance is informative. A metric that rewards ending conversations quickly will always have defenders among the people best at ending conversations quickly, and their fluency with the old number is not evidence it was measuring the right thing.