The first automation should not be the most impressive process. It should be the one where a small, controlled change can remove repeated work without hiding an important decision.
Start with a workflow the team performs frequently and can already explain. Stable steps, usable data and visible human exception handling make a change easier to test, reverse and improve.
The six-part automation scorecard
Score each candidate from 1 (poor) to 5 (strong). Use the total to compare candidates, not as an automatic approval.
| Factor | What to ask |
|---|---|
| Frequency | How often does the work happen, and how many people repeat it? |
| Visible delay | Can you observe waiting, copying, checking or chasing? |
| Rule clarity | Can the normal path be explained without relying on instinct? |
| Data quality | Are the required fields available and reasonably trustworthy? |
| Exception load | How often does a person need to make a judgement call? |
| Reversibility | Can the change be tested safely and rolled back if it is wrong? |
Good first candidates
- Move complete information from an approved form into the system that owns it.
- Create a consistent internal task when a known trigger occurs.
- Prepare a draft or summary for human review.
- Check that required fields are present before work moves forward.
- Notify the responsible person when a job has waited beyond an agreed point.
Poor first candidates
- A broken process whose owner and purpose are unclear.
- Work where most cases are exceptions.
- A high-risk decision that nobody can explain or review.
- A process built on conflicting records that have not been reconciled.
- A task selected only because a new AI tool can demonstrate it.
Simplify before automating
Remove unnecessary approvals, duplicate fields and stale reports first. Then decide whether configuration in an existing system, a simple integration or a focused automation is enough.
Safety, privacy, money and customer commitments require stronger controls even when a process scores well. Custom software should follow a clear operational case, not lead it.
Use it on real work
Bring one current process.
A free 20-minute conversation is enough to test whether the bottleneck deserves a closer look. There is no obligation and no automatic push toward software.