Why does it rise for our people after automation?

Because the easy cases stopped reaching them. What is left is the awkward and unusual work, so a rising average for human-handled cases is the expected consequence of the change rather than evidence that anybody has become slower at their job.

Is it safe to set a target on it?

Rarely, and never on its own. As a target it is satisfied fastest by closing early or answering partially, both of which produce a second contact later that nothing connects back. As a constraint, watching for unusually long cases, it behaves well.

What should it be paired with?

Something about the outcome: whether the case came back, whether the customer did what they were trying to do, or a quality review on a sample. Any of those turns a duration into information about the work rather than about the clock.

Glossary

Handle time

How long a case takes from start to finish, a measure that behaves well as a constraint and badly as a target.

In plain terms

How long a case takes. It is easy to measure, which is why it gets watched, and easy to improve by doing worse work, which is why watching it on its own is dangerous.

Why it matters

Because of everything in support measurement it is the cheapest to collect and the quickest to distort behaviour. Nobody has to be told to hurry: a number displayed on a wall does the work, and the cases that suffer are the ones that needed more time rather than less.

How it works

It is a duration, so it says nothing about what happened during it. Two cases of identical length can differ completely in whether anybody was helped, which is why it only means something in the presence of a quality measure.

Shortening it is trivially achievable by doing worse. Closing early, giving a partial answer or moving somebody on all reduce it immediately, and the cost lands later as a second contact that nobody attributes to the first.

Automation moves time rather than removing it. Work leaves the handling column and reappears as reviewing, correcting and dealing with the cases the system passed on, and a measurement covering only the first column reports a saving that the team does not feel.

It gets worse for people even when the average improves, because automation takes the short cases. What reaches a person is the residue: the awkward, the unusual and the already-annoyed, so average handle time for human-handled cases should be expected to rise, and a target that ignores this punishes the team for a change nobody asked them for.

As a constraint it behaves well. Watching for cases that take far longer than usual finds genuine problems, because an outlier is a signal about a case rather than a judgement about a person, and nobody games their way to fewer outliers.

Where the clock starts and stops is a decision, and it moves the number as much as any change to the work. Counting only active handling produces one figure; counting from the customer's first message until their problem was settled produces another, and the second is the one they experienced.

Its honest use is capacity, not performance. Knowing how long work takes is how staffing is planned, and that question does not require anybody to make the number smaller.

Two reasons the number falls

Separating them is easier than it sounds and almost nobody does it, because it requires comparing like with like rather than watching a single line. Take one category of case that a person still handles, look at how long it took before and after, and the mix effect disappears because the category has not changed. That is a few minutes of work and it converts an ambiguous chart into an answer. The reason it rarely happens is that the aggregate line is already on a dashboard and already pointing downwards, and nobody is rewarded for asking whether it means anything. The deeper habit worth building is to treat every average as a question about what changed in the population being averaged. A support queue after automation is a different population from the one before, and comparing the two is comparing different work done by different people under different conditions, which is not a comparison at all.

Seen in the wild

Average time falling because the short cases stopped reaching people at all.
Sierra
Drafting time saved and review time gained, with only the first half measured.
ChatGPT
A documented process cutting time honestly, by removing steps rather than care.
Scribe

Common misconceptions

People assume

Falling handle time means the tools are working.

In fact

It falls when automation takes the short cases, which changes the mix without improving anything for anybody. It also falls when people rush, and neither cause is distinguishable from the number alone.

People assume

Time saved on a task is time saved overall.

In fact

Work moves rather than disappearing. Drafting gets shorter and reviewing appears, and a measurement covering only the part that shrank will report a saving nobody in the team can feel.

Questions

Why does it rise for our people after automation?: Because the easy cases stopped reaching them. What is left is the awkward and unusual work, so a rising average for human-handled cases is the expected consequence of the change rather than evidence that anybody has become slower at their job.
Is it safe to set a target on it?: Rarely, and never on its own. As a target it is satisfied fastest by closing early or answering partially, both of which produce a second contact later that nothing connects back. As a constraint, watching for unusually long cases, it behaves well.
What should it be paired with?: Something about the outcome: whether the case came back, whether the customer did what they were trying to do, or a quality review on a sample. Any of those turns a duration into information about the work rather than about the clock.