The distinction that changes the decision
Automation repeats a defined rule. A process establishes when the rule applies, who owns the outcome and what happens when something goes wrong. Frequent exceptions may indicate that the decision still needs human judgement.
Check your actual workflow
Write the trigger, inputs, success condition, exception and recovery owner on one page. Try the flow manually on five varied items. If operators disagree about the right action, resolve that disagreement before selecting a connector.
Run this boundary test
Test a missing field, a duplicate and an unavailable destination. A recoverable failure is a stronger readiness signal than a perfect demo using ideal data.
When to pause
A low-volume workflow with unclear rules can be simpler and cheaper to maintain manually. Review measured time and errors before changing that decision.
Compare the actual requirements
Scroll horizontally to read all columns.
| Decision | Keep or repair the process | Investigate automation |
|---|---|---|
| Rule | Operators cannot yet agree on the correct next action | A repeatable rule and its exceptions are documented |
| Volume | A small manual workload remains manageable | Measured repeated work may justify setup and maintenance |
| Owner | Reply or recovery responsibility is unresolved | A named person can monitor, recover and stop the flow |
| Acceptance case | Five items are handled consistently by people | The same items and exceptions remain correct after automation |
A worked boundary decision
Two people classify the same enquiry differently. Automating one person’s judgement would make the disagreement harder to notice. First agree which information changes the route and how an uncertain case reaches a person. The manual exercise produces the rule that automation would later repeat.
If the work crosses the boundary
Record the trigger, authoritative input, branch decision, destination and recovery owner. Then compare a manual run with a controlled automated run. Count the time needed to detect and repair failures as well as the time saved on normal items.
Use this review sequence
- Ask two operators to handle the same varied sample.
- Resolve disagreements and write the exception path.
- Test a duplicate, unavailable destination and stop condition before increasing the sample.