Which workflows are actually worth automating

The request usually arrives as "can we automate this?" The better question is whether you should, and it has a fairly short test.

Start with the arithmetic

Volume multiplied by handling time. A task done four hundred times a month at six minutes each is forty hours — worth a serious look. The same task done twice a month is four hours a year, and no amount of elegance makes that pay back.

People consistently overestimate the frequency of tasks that annoy them and underestimate the frequency of tasks they have stopped noticing. Measure before you build.

Ask how stable the input is

Automation likes inputs that arrive in roughly the same shape every time. Invoices from forty suppliers, support tickets, form submissions — all workable.

What resists automation is anything negotiated. If the input's shape is the outcome of a conversation, the pipeline will spend its life handling exceptions, and exception handling is where automation projects go to die.

Ask what a mistake costs

This is the one teams skip.

If a wrong output is cheap to catch and cheap to reverse — a mis-tagged ticket, a draft that gets edited — automate the whole thing and move on.

If a wrong output is expensive or hard to reverse — a payment, a contract clause, a customer-facing commitment — automate the draft and keep a person on the decision. You still capture most of the time saving, and you keep the failure mode boring.

A workable order

1. Instrument the manual process for two weeks. Get real numbers.
2. Automate the highest-volume, most stable, cheapest-to-undo step first.
3. Ship it to one team, not the whole company.
4. Only then decide whether the next step is worth it.

Most of the value in an automation programme comes from a small number of steps. Finding them is worth more than building quickly.