What processes should not be automated?

Automation is an amplifier. Point it at a good process and the process gets faster and more reliable. Point it at a bad one and you get the same problems, faster, with less chance of a person noticing. Knowing what to leave alone is as important as knowing what to automate.

1. Processes that aren't settled yet

If the way a task is done changes every few weeks, automating it locks in a version that will soon be wrong. Let the process settle, write down how it really works, then automate. Until then, a checklist does the job.

2. Work that depends on relationships

A difficult conversation with a key client, an apology, a negotiation: these carry weight precisely because a person does them. Automating the message around them (the reminder to call, the summary of the account history) helps. Automating the relationship itself doesn't.

3. High-stakes judgement calls

Legal advice, pricing exceptions, safety decisions, anything where a wrong answer is expensive or harmful. Here the right approach is to automate the preparation, not the decision: gather the information, flag what's unusual, route it to the right person, and let them decide.

4. Tasks that rarely happen

Something done once a quarter usually isn't worth the build, however tedious it is. Run the numbers: time per occurrence × how often it happens × what that time costs. If the yearly total is small, a well-written instruction is the better investment. There's a fuller version of that calculation in what a business should automate first.

5. Processes that are broken

If a process has steps that exist only to work around other problems, such as double-checking a sheet because the sheet is often wrong, automating it preserves the workaround. Fix the underlying problem and the step disappears. That's why we map and simplify every process before anything is built: we don't automate bad processes. We improve the process first.

The middle ground: automate around the person

Most of the time the answer isn't “automate” or “don't”, but where the line sits. A good system does the preparing, routing, reminding and summarising, and hands a person exactly what they need at the moment their judgement matters. The person keeps the part of the job that needs a person, and loses the part that never did.

A simple way to find the line for any process is to ask, step by step:

• Does this step need a human, or just the information a human would gather?
• Would a mistake here be cheap to correct, or expensive?
• Is this step the same every time?

Steps that are repeatable, low-risk and information-only are candidates. The rest stay with people, supported by the system rather than replaced by it.

If you'd like a second opinion on where that line sits in your business, walk us through the process. Sometimes our recommendation is to leave it exactly as it is. See how we decide on our workflow automation page.