How to choose the first process to automate

Start with a task you know

Choose a task your team can explain from start to finish. It might be transferring information between applications, recording a request or preparing a recurring document. These are examples of situations to analyse; the solution depends on how each company works.

Describe who starts the task, what information they use, the steps involved and who confirms completion. Record exceptions too. A seemingly simple task may depend on decisions that still need a person.

Record the current situation

Observe a representative set of instances of normal work. Record frequency, execution time and corrections required. Separate actual working time from waiting time to understand what you want to improve.

You do not need an extensive assessment to start, but you do need a baseline for comparison. If records do not yet exist, first define how they will be collected and who will be responsible. Avoid using an estimate from memory alone as evidence of savings.

Compare the options

Choose two or three possible improvements and compare expected usefulness with the effort required. Consider access to information, system connections, tool costs and the risk of incorrect execution.

A frequent task with clear rules and organised information may justify an initial analysis. A rare task or one with many exceptions may need more preparation. Priority should follow from your company’s conditions and the process’s importance, without an automatic score that hides these differences.

• What problem do you want to resolve?
• Which steps repeat and which require a decision?
• What information is available and can be used?
• Who confirms whether the task was completed correctly?
• Which execution and ongoing support costs will you need to consider?

Choose the approach and control conditions

Improvement can start with organising the process, configuring an existing tool or integrating applications. The use of artificial intelligence should address an identified need and suitable conditions for data and review.

Where AI support is involved, define what the tool can do, which information it can use and who checks the response. Execution speed should be assessed alongside quality, exceptions and review effort. A result that requires many corrections may not improve the work.

Confirm the access required, responsibilities and procedure when execution fails. Your team should know how to recognise an exception and resume the work.

Define what you will test and measure

Before extending use, agree an initial implementation with clear scope and criteria. Choose test cases, who validates them and what counts as an acceptable result. Compare the original task with the new approach under conditions that are as similar as possible.

You can observe time per task, the number of corrections or correct completion of requests. Also record workload and tool and maintenance costs. A gain per operation does not, by itself, demonstrate overall financial savings.

After evaluation, decide whether it makes sense to retain, adjust or extend the solution. The first change should help you understand the process and prepare an evidence-based decision.