Over two years we automated roughly seventy percent of one team's routine volume: account provisioning, standard access grants, environment refreshes, a long tail of repeatable requests. The business case held up. The time saved was real and nobody wants those tasks back.
What nobody modelled was what the remaining thirty percent would be made of.
It is not a smaller version of the old work. It is the residue, and the residue is the part that resisted automation because it contains judgement, ambiguity or an exception. Every item now needs someone to read it properly, understand a context, make a decision they could be asked to defend, and often talk to a person in another department. The queue used to be a mixture of demanding work and undemanding work, and the undemanding work was doing something we never credited it with: it gave people somewhere to put the tail end of a difficult afternoon.
The measurements did not move with the mix. Average handling time rose sharply, because the easy items that used to pull the average down are now handled by a pipeline in nine seconds and no longer appear in the figures at all. Satisfaction scores dipped, because the requests still reaching humans are the ones most likely to end in a no. For two quarters our reporting said the team had got worse at its job while it was demonstrably doing harder work than before, and I had to spend real credibility explaining that to people who only saw the trend line.
There was a second effect I did not anticipate. Handling an exception well requires knowing what the normal case looks like, and the team no longer performs the normal case. The knowledge that used to come from repetition now has to come from documentation, which means the automated path needs to be written down in a form a human can read, not just executed.
We report by category rather than by average, so the difficult work is visible as difficult. Staffing is planned on complexity rather than volume. And when we now propose automating something, the paper says what will be left behind, because that is the job we are actually designing.
Automation does not make a job smaller. It makes it denser.
– Serguey Shinder
