Practical guide / Process automation
Which business processes should you automate first?
Start with a frequent task, a recognisable input, a verifiable result and someone who can resolve exceptions. That combination lets you measure an improvement and decide whether it is worth extending.
The next task, where it belongs.
- Input
- Rule
- Action
By Kenea · Published and reviewed: .
1. Describe one operation from start to finish
“Automate administration” is too broad to price or assess. A useful description would be: “when a service request arrives, check the mandatory details, create a draft in the system and assign it to the person responsible”. This identifies what triggers the process and where it ends.
Walk through several examples with the person doing the work. Note where they look up information, what they copy, what they decide and what they correct. Include a normal case and an incomplete one. If people interpret an approval differently, agree the rule first: automating a disagreement does not resolve it.
- Input
- A form, a file, a change of status or a scheduled time. Specify which information must be present.
- Transformation
- The checks, calculations, searches or decisions required.
- Output
- A record created, a task assigned or a report made available. “The workflow ran” is not enough if the result is missing from where it is needed.
- Exception
- What happens when information is missing, two matches are found or an application does not respond.
- Owner
- Who authorises the change and who handles unresolved cases.
2. Choose a valuable task that is ready to move forward
Frequency matters, but a small task repeated often may rely on unreliable data or a closed application. Compare opportunities using the same worksheet before choosing technology.
| Criterion | A good sign for a trial | What to clarify if missing |
|---|---|---|
| Workload | Volume and manual time have been observed. | Measure a sample, including corrections. |
| Rules | Two people reach the same result with the same data. | Resolve conflicting criteria. |
| Access | The software provides an authorised way to read or write. | Check interfaces, permissions and the provider’s terms. |
| Exceptions | They are recognisable and assigned to someone. | Define a review queue and how work resumes. |
| Control | A limited trial can verify the result. | Prepare test data, boundaries and recovery. |
Do not turn the table into a score that conceals a blocker. Large potential time savings do not compensate for missing permission to process data. Among tasks that meet the conditions, prioritise observable benefit against implementation and maintenance effort.
Common candidates include registering requests, preparing drafts from validated data, reminding people about pending approvals or gathering information for a report. These are possible applications, not Kenea client cases. Each needs its own analysis.
3. Measure the whole task, including exceptions
Separate active working time from the time a request spends waiting. Automation may remove a two-day delay without saving two working days. It may also accelerate data entry while shifting effort to the person reviewing errors.
Microsoft proposes combining business indicators with feedback from people using the solution. Its examples include time, errors, adoption and cost. Choose measures that explain the particular problem. Source: Microsoft Learn, measuring business value.
Available time does not automatically become cash savings. The effect may be greater capacity, fewer delays or better quality. Total cost includes licences, service usage, maintenance, changes to the source software and the time of those supervising the process.
Also record how many operations finish correctly, how many need review and how long resolution takes. A process that accumulates pending work needs attention even if its execution count is high.
4. Decide which parts need rules, integration or AI
When the input is structured and the decision exact, a rule may be enough: check a field, apply an agreed threshold or copy an identifier. If the task moves data between applications, examine systems integration and ownership of each data item.
AI can help interpret documents or variable language. That part needs examples, quality evaluation and a route for cases it cannot resolve. Our AI or automation guide develops this decision. The rest of the process can continue to use explicit rules.
Before building, review what your existing software already supports. An existing feature, a configuration change or a maintained connector may meet the need. If the only available route operates through screens, also check what happens when the interface, session or permissions change.
5. Define a trial that can end with a decision
Limit the trial to one input type, one output and one owner. It may prepare a draft for comparison with the expected result without taking the final action yet. The aim is enough evidence to proceed, adjust or reject that approach.
- Normal case: the result contains the expected data and appears in the correct destination.
- Incomplete input: missing information and its reviewer are identified.
- Repetition: receiving the same request twice does not accidentally create two operations.
- Interruption: a failure or restart leaves a state that can be explained and recovered.
- Exception: the responsible person understands the problem and can continue the work.
Set criteria before running the sample: which errors block progress, what a person must review and what evidence to retain. There is no universal success percentage for every process. Internal classification and changing a customer’s data have different consequences.
A worksheet for the first conversation
Copy these questions and answer them with your team. Mark anything not yet measured as pending.
- Which operation do we want to improve, and what result should it produce?
- How often does it happen, and how much active work does it require today?
- Which applications, versions and people are involved?
- Which exceptions are common, and who resolves them?
- Which access is authorised, and what data does the trial actually need?
- How will we assess quality, time, cost and recovery?
At Kenea, a commissioned technical assessment turns these questions into a scope and recommendation. Implementation, testing and operating conditions are agreed for each case. Explore our process automation work and the full range of Kenea solutions.
To begin organising your ideas, the free assessment maps a path from your choices. You can use it without registering; it provides initial guidance.
Explore my process Compare rules and AI
Source and editorial approach
Source consulted on 8 September 2026: Microsoft Learn, methods for measuring business value, updated 14 May 2025. The matrix, worksheet and calculation are Kenea’s editorial proposals for preparing a decision; they are not a benchmark or implementation results.