How to calculate AI automation ROI without inventing a 10× claim.

An automation business case should begin before the automation. Record how often the workflow runs, who touches it, how long each step takes, how frequently it fails, and what delay costs the business. Without that baseline, almost any improvement can be made to look impressive.

Use a representative sample instead of one unusually painful week. Include supervision, rework, handoffs, and time spent finding information—not only the obvious data-entry step.

Time returned is not automatically cash saved. If an automation gives a salesperson five hours back, the value depends on what that person can do with the capacity. Cost avoidance, revenue acceleration, error reduction, and improved service each need their own assumption.

Include design, implementation, model and platform usage, monitoring, exception handling, and maintenance. Agentic systems can trade more latency and cost for flexibility, so compare them against a simpler workflow rather than against doing nothing.

A credible forecast uses a conservative case, an expected case, and a break-even point. After launch, replace assumptions with measured run volume, success rate, human review time, and cost per completed outcome.