In this guide 6 sections
Do not automate a process that is still changing, runs too rarely to repay the effort, or depends on judgement a rule cannot capture. An upset customer’s tone is one example. Often the better first move is to cut steps, after which many processes turn out fast enough to stay manual, with no tool to buy or maintain.
Unstable, rare or judgement-heavy work
Three signs say a process is not ready. The first is instability: the steps keep changing while the business finds the right way to work, so you would be automating a moving target. The second is low volume: a task done a few times a year costs more to automate and maintain than the manual work ever will. The third is judgement. Each case turns on context and nuance, not a rule.
Consider a boutique agency’s client onboarding call. An account lead builds rapport and gauges tone in conversation, and no intake form captures that tacit knowledge, so a form would remove the point of the call. It should stay a conversation.
What a wrong result costs
Ask what a single wrong result costs. A rule that miscategorises an internal report is a minor irritation, while one that sends the wrong invoice amount to a customer belongs in a different class.
Where an error is expensive or hard to reverse, put a person’s check before the step that cannot be undone, and pilot on a low-stakes version first.
Simplify before you automate
Many candidates just need fewer steps. Suppose a purchase approval has seven sign-offs, several of them discretionary habits, and automating all seven only speeds up an overcomplicated route without fixing the design.
Ask whether you would design this process from scratch with this many steps, and if the honest answer is no, cut the surplus steps before touching any software. Often what remains is fast enough by hand.
- Steps that exist only because of a past mistake.
- Approvals nobody ever rejects.
- The same data typed into two places.
Keeping a person in the loop
Some steps deserve to stay manual on purpose. Keep a human review before a large payment leaves, before a condolence note or a reply to a formal grievance goes out, and before any decision with legal weight. People catch what rules miss.
Decide the human steps upfront. Build the automation around that decision from the start, because discovering the gap after a wrong payment or an unfit message costs far more than designing the handover.
When to revisit the decision
A process that is unstable today may become a good candidate once it settles, and a low-volume process may earn automation if the business grows into higher volume. Put the review on the calendar every six months or every year, so the decision does not depend on someone remembering to raise it.
Judgement-heavy work is the category least likely to shift with volume, since a process that needs a person reading context stays that way as it grows. Its administrative edges may still qualify, such as scheduling the call, sending the follow-up notes and filing them against the client record, while the core judgement stays human.
Automate, assist or keep manual matrix
A three-column matrix with columns for automate fully, automate with a human checkpoint and keep manual, and rows for stability, volume, judgement needed and cost of an error. Score each process on its own.
Open a blank worksheet to printQuestions owners ask
Can part of a process be automated?
Yes, and that is often the right answer. Scheduling, reminders and record-keeping around a judgement-heavy task can usually be automated while the judgement itself stays with a person, which is more realistic than all or nothing.
What makes an action too risky?
An action is too risky when a mistake is expensive, hard to reverse or lands on a customer or a compliance duty, such as a wrong invoice amount. Put a person’s check before it.
Is low volume always a reason to wait?
Not always. A task done rarely can still justify automation when each manual run is long or error-prone, because volume is only one input among several.