Find the AI work that is actually worth automating.

Look for work done every day or every week: copying data, answering the same questions, preparing reports, following up with leads, summarizing documents, routing requests, or checking multiple systems.

Good candidate: “Every morning I copy new customer requests from email into our tracking system.”

Weak candidate: “I want an AI agent because everyone is talking about agents.”

Frequency: 0 = occasional, 1 = weekly, 2 = daily/many times a day.

Time: 0 = under 10 minutes/week, 1 = 10–60 minutes/week, 2 = more than an hour/week.

Repeatability: 0 = mostly judgment, 1 = mixed, 2 = clear repeatable steps.

Business value: 0 = convenience, 1 = useful, 2 = affects revenue, customer service, cost, risk, or capacity.

Risk penalty: subtract 2 if the task involves highly sensitive personal, medical, financial, credential, or regulated data and you do not already have an approved secure architecture.

A score of 6–8 is a strong candidate for a small pilot. A score of 3–5 needs more analysis. Below 3 usually should not be your first automation.

Use a basic monthly estimate:

hours saved per month × loaded hourly cost = potential labor capacity created.

Then subtract software, implementation, maintenance, and review time. Treat saved hours as capacity—not guaranteed cash profit. Revenue only appears if that capacity is converted into more sales, more billable work, lower staffing cost, or another measurable business result.

Pick the highest-scoring low-risk task. Keep a human approval step at first. Measure time saved, error rate, cost, and whether people actually use the workflow. Only expand after the pilot produces evidence.

Want help? See the $199 AI Workflow Audit →