Start with the bottleneck, not the technology
The best first automation is rarely the most impressive one. It is usually familiar work that happens every week: copying approved information into another system, preparing the same report, checking documents for the same fields, or routing a request to the right person.
A useful candidate has a visible business consequence. It may delay an invoice, occupy an experienced employee, create corrections, or make it difficult to see who owns the next action. Starting from that consequence keeps the discussion grounded in cost, capacity, service quality, and control.
How to know what to automate in operations
Look for work that repeats often enough to create a visible operating cost. The strongest signals are copy-paste between systems, recurring document checks, status chasing, routine approvals, and reports rebuilt from the same sources. These are not automatically good projects, but they are the right places to measure.
Write down the monthly volume, average handling time, waiting time, correction rate, and the role doing the work. Then ask what would happen to the released capacity. A process is more attractive when improvement can absorb growth, prevent a hire, accelerate cash collection, protect a service level, or return skilled people to higher-value decisions.
The six-factor process scorecard
Score each factor from 0 to 3. Use the total to compare processes, not to manufacture false precision. A high score means the process deserves a closer assessment; it does not automatically mean software should be built.
| Factor | 0 points | 3 points |
|---|---|---|
| Frequency | Occasional or unpredictable | Daily or many times per day |
| Time | A few minutes in total | Material staff hours every month |
| Stability | Steps change frequently | Rules and exceptions are understood |
| Data readiness | Missing or inconsistent | Available in consistent digital formats |
| Error impact | Easy to notice and correct | Creates rework, cost, delay, or risk |
| Ownership | No clear decision-maker | One owner can approve the process |
How to assess one process in an afternoon
- 01
Name the start and finish
Use observable events, such as “a signed order arrives” to “the project is created and scheduled.”
- 02
Follow one real case
Record every handoff, system, wait, correction, approval, and manual decision.
- 03
Measure a normal month
Estimate frequency, handling time, waiting time, errors, and rework.
- 04
Separate rules from judgement
Stable rules suit automation. Negotiation and sensitive exceptions usually need a person.
- 05
Define the smallest useful change
Improve one costly step before redesigning the entire company.
- 06
Choose one result to verify
Track hours saved, turnaround time, correction rate, or service-level performance.
A simple time-and-cost example
Suppose a team processes 600 requests per month. Each request takes eight minutes to read, validate, enter into a second system, and confirm. That is 80 hours of direct handling before counting corrections, waiting, or management checks.
At a loaded cost of €35 per hour, the visible labour cost is €2,800 per month. If a focused workflow removes 60% of handling while people continue to review exceptions, it can release about 48 hours per month. That is capacity—not automatically cash. Its value becomes real when the business uses those hours to absorb growth, improve service, avoid overtime, or delay an additional hire.
Three anonymized examples
- A service team re-entered approved customer details from email into a project system. High frequency, stable fields, and measurable corrections made a simple integration a strong candidate.
- A monthly management report required exports from several systems and spreadsheet cleanup. The calculations were stable, but the source data was not. The right first step was to standardize inputs, then automate assembly.
- A senior manager reviewed every unusual customer request. Volume was low and decisions depended on context. A decision checklist and better routing created more value than full automation.
When automation is the wrong next step
Do not automate a process simply because employees dislike it. The work may be unnecessary, the policy may be unclear, or an existing system may already provide a feature that nobody uses.
- The process is about to change materially.
- Exceptions outnumber normal cases.
- Nobody owns the outcome or can approve changes.
- The information is unreliable and there is no plan to improve it.
- The financial or service impact is too small to justify maintenance.
- A simpler policy, template, or existing product feature solves the problem.
What to decide after the scorecard
For a promising candidate, calculate the current cost and the realistic value of improvement. Then choose whether to simplify the workflow, configure an existing product, connect systems, or build a focused internal tool.
AI belongs in the discussion only when the work involves language, documents, classification, search, or other tasks where fixed rules are insufficient. A small, measurable first scope gives the business evidence before it commits to a broader change.