A useful first automation is usually a small, well-understood task. Start with a process people repeat regularly, rather than a broad promise to automate the business.
Watch the handoffs.
Look for information being copied between an inbox, a spreadsheet, and another tool. Write down what triggers the task, what information it needs, who checks it, and what happens next. A process that cannot be described clearly is difficult to improve reliably.
Pick a task with a clear finish.
Preparing a weekly report, routing an inquiry, or creating a draft response can have an observable result. Pick one task and define what a successful output looks like. Keep a person responsible for decisions that need judgment.
Separate rules from interpretation.
Moving a record when a field matches a known value may only need ordinary software rules. Interpreting an open-ended message may benefit from AI, but the result still needs review. Use the simplest approach that does the job.
Try it on a small sample.
Include incomplete records and exceptions, not just ideal inputs. Compare the output with what your team would have done. Record errors and the time spent checking the result.
Agree on the next step.
Before expanding, decide who owns the workflow, how a failure will be spotted, and how to return to the previous process. A useful starting brief is: “When this happens, prepare this output for this person to review.”
Discuss an automation project ↗