Pick the most painful recurring task in your operation. The one someone updates at 11pm because if they don't, Monday breaks. Plug in four numbers. See what that work is costing you per year — and what an AI agent doing the same job would cost.
Manual cost = hours × cycles × hourly rate × (1 + error rate). The error multiplier captures the time spent re-checking, correcting, and re-running. Most teams underestimate this; in our experience the true rate is usually higher than the first guess.
Agent cost = one-time build + annual API spend. A typical scheduled agent for a single workflow costs €4-15k to build (depending on the number of data sources and the complexity of the output) and runs on €100-500/year in API costs for typical small-business volumes. We anchor on €8k + €500 as a middle-of-the-road default.
What this calculation does NOT include: the qualitative wins. The report not getting skipped because someone was sick. The numbers being consistent week-to-week instead of formatted differently every time. The team member doing higher-leverage work instead of the manual roll-up. These tend to dwarf the dollar savings, but they're harder to estimate, so we leave them off the headline number.
If the year-1 net is comfortably positive, this task is a strong candidate for automation. Tell us about it and we'll figure out together whether a 3-week build makes sense. Bring the spec, or just describe the spreadsheet.
If the year-1 net is negative or close to zero, the task is probably too small to be worth automating on its own. Worth bundling with one or two other tasks of the same shape into a single project.