Calculate ROI by comparing what the manual process actually costs today (fully-loaded labor, error correction, missed volume) against the one-time build cost plus ongoing per-run inference cost, then solve for the volume where those lines cross. If that break-even point is inside your realistic 12-month volume, build it. If it isn't, don't.
Most founders skip this and instead ask "is AI worth it," which isn't answerable. The answerable question is narrower: does automating this specific task, at this volume, pay back faster than the cash and attention it costs to build and run. Here's the framework we use before scoping any agent project.
Step 1: Price the manual baseline honestly
Take the task the agent would replace and cost it out the way a CFO would, not the way an optimist would.
Fully-loaded hourly cost of whoever does the task now (salary + benefits + overhead, not just base pay)
