Practical guide
How to Decide If a Process Should Use AI
A practical way to decide whether a process should stay manual, use rules, add AI assistance, or be automated, and where human review belongs.
Last updated
2026-09-14
Start with the process, not the tool
The question "should we use AI here?" usually starts in the wrong place. It invites a yes or no answer before anyone has described the work. A better first question is "what outcome do we want to improve, and what is getting in the way?"
Once the process is clear, AI becomes one option among several. A rule, a checklist, a form change, or a simpler handoff may solve the problem with less risk and less maintenance.
The four approaches
Most processes fit one of four approaches. Choose the simplest one that reliably reaches the outcome:
- Manual: keep a person doing the work when volume is low, judgement is high, or the process is still changing.
- Rules-based: encode a clear decision as an explicit rule when the inputs are structured and stable.
- AI-assisted: use a model to draft, classify, or summarise messy inputs, with a person reviewing the output.
- Automated: connect the steps end to end when the work repeats, the inputs are stable, and the risk is manageable.
What changes the answer
Four factors move a process between approaches. Repeatability decides whether building anything is worthwhile. Input consistency decides whether a rule can hold. The consequence of an error decides how much human review is needed. The clarity of the existing rule decides whether AI adds anything at all.
Risk is the factor people skip. High-consequence work does not need more automation. It needs a named reviewer and a recorded approval before the output is used.
Write a short decision record
A defensible decision fits on one page: the problem, the options considered, the chosen approach, the reason, the review boundary, and what would change your mind. That record is also strong portfolio evidence because it shows how you think, not only what you built.
Keep the record short enough to explain in a few sentences. If it needs a long document to justify, the scope is probably too large.
The four-way approach test
- 01Outcome: define the decision or work to improve.
- 02Repeatability: check whether the inputs and steps are stable enough to systematise.
- 03Risk: assess the consequence of an inaccurate or delayed result.
- 04Approach: choose manual, rules, AI-assisted, or automated work with a review boundary.
Common questions
Should every repetitive task be automated with AI?
No. Repetition alone is not enough. Consider process stability, data quality, risk, review capacity, expected benefit, and whether simpler rules or manual work are more appropriate.
What is the difference between rules-based and AI-assisted?
A rules-based approach follows an explicit, deterministic rule that you can read and test. AI-assisted work uses a model where the inputs or decision need judgement, so it should be paired with human review.
What is useful portfolio evidence for this decision?
A bounded decision brief that shows the problem, the options, the risk level, the recommended approach, the review boundary, and the next test.
This resource is maintained by Iteretta. It is educational information, not legal, financial, medical, employment, or other professional advice.