Automate the exception, not just the happy path

Where the hours actually go

Ask a team how long the standard case takes and they will tell you three minutes. Ask what ruins a Thursday and you will hear about the invoice with no purchase order, the customer whose name is spelled two ways, the request that arrived by WhatsApp instead of email.

Automating only the standard case therefore removes the cheap part of the work and leaves the expensive part untouched. The team reports that nothing much changed, and they are right.

Exceptions are not chaos

The word makes them sound unpredictable, but most exceptions repeat. There are usually five or six recurring varieties, each with an informal rule that someone experienced applies without thinking about it.

Getting those rules written down is the actual project. It is also the part that resists automation, because the person holding the knowledge has never had to articulate it, and often does not agree with the colleague who handles it differently.

Resolving that disagreement is worth more than the software. It is the reason discovery phases pay for themselves even when nothing gets built.

Design for the ones that stay manual

Some exceptions genuinely need judgement, and the system should route them cleanly rather than attempt them. That means a queue a person can work through, with the context already assembled and the reason it was flagged stated plainly.

Done well, a person handling twelve exceptions a day with everything in front of them is faster than the same person handling three hundred cases of which twelve were unusual.

What to measure

Not the volume automated. Measure how long the exceptions take now, how many need a second pass, and how often something reaches a customer that should have been caught. Those are the numbers that tell you whether the process improved or simply moved.