Short answer
Automate work that is frequent, recognisable and verifiable. Keep work that requires judgement, commitment, negotiation or accountability with people. The boundary is not human versus machine; it is predictable work versus contextual decisions.
What to observe
Break the process into collection, rules and decisions. The first two are natural candidates; the third needs explicit boundaries. A sound solution prepares the decision, shows its sources and flags exceptions instead of hiding what it did.
The same information is copied between tools.
Rules are stable and the outcome can be checked.
The team can state exactly when an exception needs a person.
A concrete example
For a sales enquiry, a solution can collect product, volume, deadline and history, flag missing data and prepare the next action. A person still decides terms, priority and the final response.
How to move safely
- Map inputs, rules, decisions and outputs.
- Define where a person must validate.
- Test with a small volume and keep an action record.
Test your process
Tick what happens today. The result changes as you answer.
There is not enough evidence to automate yet. Observe the process for a few days and record volume, exceptions and time spent.
Frequently asked questions
Does automation reduce visibility?
Not when it is designed well. It records actions, exposes sources and explains why a case was routed.
Must the whole process be automated?
Almost never. Start with the most repetitive, lowest-risk segment, measure it and expand only when the evidence supports it.