Start with a decision, not a tool
The best first automation is rarely the most spectacular. It lets the business learn within a bounded risk, creates an observable improvement and fits a process the team already understands.
In a small business, look for repetitive work with recognisable inputs and an output a person can validate: classifying requests, preparing a first draft, extracting fields or gathering context before deciding.
Assess four dimensions
Compare opportunities using the same criteria before deciding. In an SME, the fourth dimension is usually decisive: without a person available to review, the best automation fails.
- Value: time consumed, frequency, cost of errors and effect on customers or employees.
- Feasibility: whether data exists, may be used and can be integrated reasonably.
- Risk: consequences of errors, human review, limits and retained evidence.
- Adoption: a clear owner, available users and a path into daily work.
Design a pilot that can finish
Define one input, one output, a user group and an observation period. Decide from the start how errors will be recorded, which cases will be escalated and who can stop the system.
The first implementation should answer a concrete question: can this intervention improve this process under these conditions? Measure a simple baseline first: time, errors or response time.
The final choice
Prioritise the case that combines observable value, accessible data, manageable risk and a team willing to test. An estimate helps compare opportunities; it is not a savings guarantee.
If two cases compete, choose the one you can observe best: what the first one teaches will make the second one better.