What to automate first

The selection criteria

Four properties make a process a good first candidate, and they compound.

1. Frequency. It happens often enough that consistency matters.
2. Rules. The correct behaviour can be stated without hedging.
3. Documentation. Someone can already describe it end to end.
4. Low reversal cost. A wrong result is inconvenient, not damaging.

A process with all four is a straightforward build. A process missing only documentation is a documentation task first. A process missing rules is usually not an automation candidate at all — it is judgment, and it should be recognised as such.

Where the first wins usually are

In most businesses the same handful of processes qualify.

None of these are glamorous. All of them are places where work is routinely lost because a person was busy, and where consistency produces a visible difference within weeks rather than quarters.

What should stay with people

Some work should be protected from automation deliberately.

Judgment that depends on relationship or context — how to handle a long-standing customer's unusual request — does not have stated rules, and encoding a guess makes the business worse at the thing it is best at. Decisions with legal, financial or safety consequence need a person accountable for them. And rules that change monthly produce automations that are permanently out of date.

How to sequence

Order the work so that each step reduces the cost of the next. That usually means resolving record definitions between systems before building processes that cross them, because every later automation depends on that agreement. It also means building the capture step before the follow-up step, since follow-up on incomplete records produces noise.

A workable sequence for most businesses:

1. Agree what the core records mean and which system owns each.
2. Automate capture and immediate response.
3. Automate routing and record creation.
4. Automate structured follow-up.
5. Add measurement, so the next decision has evidence behind it.
6. Revisit the review boundaries using what the logs show.

How to decide when the list is long

When several processes qualify, choose the one where the gap between current and consistent behaviour is largest. That is rarely the process people complain about most; it is usually the one nobody mentions because it fails silently.

If you want a structured version of this exercise for your own business, the Growth Intelligence Assessment walks through the same questions and returns a prioritised view. The failure modes to avoid while sequencing are covered in where automation fails.