The hidden cost of the human handoff

An automation that handles eighty per cent of cases is usually sold as an eighty per cent saving. It rarely is, and the reason is the other twenty.

Relocated work is not removed work

When a case is routed to a human, three costs appear that did not exist before. The person has to rebuild context the automation already had. They have to decide whether the automation was wrong or merely uncertain. And they are now doing only the hard cases, back to back, with none of the easy ones that used to provide recovery time.

That last effect is the one nobody budgets for. A queue of pure exceptions is slower per case than the same cases mixed into normal work, sometimes by half.

What a good handoff carries

A case arriving at a person should bring everything the automation saw, what it concluded, how confident it was, and what specifically it could not resolve. "Failed validation" is not a handoff. "Supplier VAT number does not match the master record; extracted VAT is BG123456789, master is BG123456780; confidence 0.94" is.

The difference is measured in minutes per case, and at volume it is the difference between a saving and a transfer.

Design the queue, not just the route

Three questions decide whether the exception queue works: who owns it, what the target turnaround is, and what happens when it grows. A queue with no owner becomes a backlog nobody has agreed to carry, and it is usually discovered at quarter-end.

Measure both sides

Report the automated cases and the exception cases as one number: total cost per case across the whole process. Teams that report only the automated share get a flattering figure and an unexplained gap in their actual capacity.

The rule we use

If you cannot describe the exception path to the person who will staff it, in one paragraph, without using the word "manual", the automation is not ready to go live.