Automation promises time back. But automating the wrong process can lock in a bad way of working, frustrate the people who rely on it and cost more to maintain than it saves. Before building anything, ask seven questions.
- Does it happen often? High-volume, recurring work pays back fastest. A task done twice a month rarely justifies the effort.
- Is it repetitive and predictable? Clear rules, or patterns AI can learn reliably, make good candidates. Work that changes every time does not.
- Are the inputs digital and consistent? Emails, forms and structured documents are easier to automate than handwritten notes or free-form phone calls.
- Can success be measured? If you cannot measure time, errors or backlog today, you will not be able to prove the benefit tomorrow.
- Is there a clear owner? Someone must decide how the process should work and handle the exceptions automation cannot.
- What is the cost of a mistake? Low-risk tasks can run with light review. High-risk decisions need a person to approve every outcome.
- Is the process already sound? Automating a broken process only produces bad results faster. Simplify first, then automate.
A simple scoring grid helps. Rate each candidate from one to five on volume, predictability, data quality, measurability and ownership, and subtract points for risk. Start with the highest scorer, not the one with the loudest sponsor.
Finally, design for exceptions from day one. Every automation meets cases it cannot handle. Decide where those go, who deals with them and how quickly. A good automation does not remove people from the process; it gives them the unusual, judgement-heavy work and takes away the rest.