Which processes to automate first in an SMB

Every SMB that looks at AI faces the same question: where to start? The instinct is to attack the biggest pain point. It is usually wrong. The biggest pain point is often the least structured process — the one where automation is hardest and failure most visible. The right first target is picked with two questions, not with emotion.

The two-question matrix

Question one: how often does the task happen? Daily beats weekly beats monthly — frequency is what turns a small saving into a real one. Question two: are the rules known? If an experienced employee can explain how they handle ninety percent of cases, the rules are known. Automate what sits in the top-right corner: high frequency, known rules. Email triage, follow-ups, data entry, reconciliations and standard reporting almost always live there.

On administrative work specifically, market studies talk about time savings close to 40% once the repetitive layer is automated. Treat that figure as an order of magnitude, not a promise — your own number depends on your volumes and will only be credible if you measure your baseline before starting.

The three prioritisation mistakes

• →Starting with the exception instead of the rule: automating the complex edge case first, when the volume — and the payback — is in the boring ninety percent.
• →Starting with a process nobody owns: if no one can state the current rules, the project becomes a process-definition project, and stalls.
• →Starting with the customer-facing process: begin where an error costs an internal correction, not a client relationship. Earn the right to automate outward.

This week, draw the matrix on one page: list ten recurring tasks, score each on frequency and rule clarity, and circle the top-right corner. The task you circle first should be running with an agent — under human validation — within a month. That first visible win is what funds, politically and financially, everything that follows.