When not to automate
Automation applied to the wrong process makes it fail faster. Four situations where the right answer is to fix something else first.
Every automation project has a moment where the honest answer is "this should not be automated". Recognising it early saves a great deal of money.
1. The process is broken, not slow
Automating a broken process does not fix it. It produces the same wrong outcome, faster, and at greater volume, while making the fault harder to see, because nobody is manually touching it any more.
If the current process produces inconsistent results when a careful person runs it, automation is premature. Fix the logic, then automate it.
2. The step should not exist
A surprising number of repeated tasks exist because of a decision nobody remembers making. A report nobody reads. A form field nobody uses. An approval step added after an incident five years ago.
Before automating a task, ask what would happen if it simply stopped. Occasionally the answer is "nothing", and the cheapest automation is deletion.
3. Judgement is the actual work
Some tasks look repetitive but are not. Pricing an unusual job, handling an unhappy customer, deciding whether to take on a client, these have a repeatable shape and a non-repeatable substance.
The useful move here is to automate the preparation, not the decision. Gather the information, assemble the context, draft the options, and leave the call to a person.
4. The volume does not justify it
A task performed twice a month, taking ten minutes, costs about four hours a year. If automating it reliably takes two days and needs maintaining, the arithmetic does not work.
This one is worth being disciplined about, because low-volume tasks are often the most irritating, and irritation is a poor guide to return.
What this leaves
What remains after those four filters is usually a short, unglamorous list: high frequency, low judgement, well understood, and currently dependent on someone remembering. That list is where automation earns its money.
- Acknowledging and routing enquiries
- Collecting information you always ask for anyway
- Scheduling and chasing follow-ups
- Moving data between systems that should already be connected
- Requesting reviews after completed work
- Assembling the same report every month
It is a less exciting list than most automation pitches. It is also the one that tends to pay for itself.