
Which Processes to Automate First: 5 Criteria
· by Panayot Dimkov
Which processes should you automate first? The ones that repeat often, follow stable rules, receive their data through a tidy channel, can tolerate an error without serious consequences and have an owner who can validate the result. In practice, though, the choice is nearly always made on enthusiasm: people automate whatever sounds most modern, whatever the boss asked for or whatever the vendor knows how to build. Choosing by criteria takes an afternoon and completely changes the outcome of the first project — which is the one that decides whether there will be a second.
1The five criteria for prioritising processes
2The effort-impact matrix, properly defined
The effort-and-impact matrix is useful if both axes are defined precisely. Left undefined, everyone puts their favourite projects in the top right.
| Axis | What it measures exactly |
|---|---|
| Impact | Hours per year it frees, plus the cost of the errors it prevents. In numbers, not adjectives |
| Effort | Number of systems to connect, number of known exceptions and whether the process has to be tidied first |
With those two axes properly set, something uncomfortable usually appears: the highest-impact processes are nearly all in the high-effort zone. That is why the first project is not chosen for maximum impact but for the best impact-to-effort ratio in the low-effort quadrant.
3Three good first candidates in almost any business
4Four traps when choosing
- The process that is about to disappear. Before investing, ask whether that system is being replaced next year or whether a rule change is about to reshape the workflow.
- The one the boss asked for. It may be the right one, but score it with the same criteria as the rest. If it comes out badly, that conversation is easier before than after.
- The one nobody wants to touch. Sometimes the reason is technical and sometimes it is political. If it is the second, the tool will not solve it.
- The one that only exists in one person's head. Tempting because the saving looks huge, but if it is undocumented it has to be documented first, and that is another task.
5How to score it in an afternoon
No methodology required. A table with the candidate processes and five columns will do, scoring each criterion from one to three.
What gives the exercise its value is not the final score but the conversations it triggers. The frequency column forces measuring instead of estimating. The stability column reveals that two people do the same process differently. And the "who decides" column discovers, more often than you would expect, that nobody owns that workflow.
6Why the first one should be small
The first project is not chosen to maximise savings. It is chosen to prove this works and that the team adopts it, which are the two real unknowns. A small, stable, visible process answers both within weeks.
And there is a side effect worth more than the saving: when someone sees their repetitive task disappear, they start pointing at others. The best inventory of automatable processes does not come from a consultancy, it comes from the team after the first success.
Frequently asked questions
Where do I start with twenty candidate processes?
By scoring them against five criteria: frequency times unit time, stability of the rules, quality of the input data, cost of error, and whether an identifiable person suffers it. That reduces twenty candidates to three or four in an afternoon, and the conversation the exercise triggers is usually worth more than the score.
Should I start with the process that consumes the most time?
Not necessarily, and it is usually a bad idea. The most time-consuming process tends to also be the most complex and the one with the most exceptions. For a first project you want a good impact-to-effort ratio, something that shows results in weeks and builds the confidence to tackle the big one afterwards.
What if the process is not defined?
Define it first, which is business work rather than technical work. If everyone does it differently, automating only makes the disorder circulate faster. Sometimes that definition exercise already solves half the problem without touching any tool.
How do I know it is worth it before spending?
With the baseline: count one real month's volume, time the task once and note the error rate. With those three numbers and a quote, the return can be calculated without guessing. How to do that honestly is in process automation ROI.
What if the process disappears within a year?
Then it rarely pays off, unless the saving in that year already covers the investment. It is a question to ask explicitly when choosing: if the system is being replaced or a rule change will reshape the workflow, that candidate drops to the bottom of the list however painful it is today.
In short
There are five criteria and the most predictive is rule stability, not volume. A frequent but shifting process produces a flow that needs endless tweaking; a stable, boring one produces savings for years.
And the first project is chosen for something else: to prove quickly that it works and that the team adopts it. After that first success, the team itself points at the next candidates better than any analysis. That is how we approach AI process automation.
Several processes and no idea which to tackle?
We score them with you against the five criteria, measure the baseline of the two or three finalists and tell you which one shows results soonest. No obligation.
Request a process analysisNexum Blog
Related articles

6 Jul 2026
Digital twin in industry: a 2026 guide
What a digital twin in industry is and isn't: types, real use cases, the technology you need and how to launch a pilot with measurable ROI.

11 Nov 2026
RPA vs API: when each one is worth it
RPA vs API: if there is an API, use it; RPA pays off without one or as a stopgap. When it is a patch, attended vs unattended, and how to harden it.

6 Nov 2026
Intelligent Document Processing (IDP) vs OCR
Intelligent document processing (IDP) goes beyond OCR: it classifies, extracts and validates delivery notes, certificates and contracts without templates.