What AI Automation Costs

AI automation cost breaks into five parts, and only one of them appears on a vendor price list. The assessment is a fixed, bounded piece of work that establishes which processes are worth automating and what they currently cost — it should be priced before it starts and should pay for itself in the first workflow. The build is priced per workflow and is driven by integration complexity, not by AI: reaching an old ERP costs more than reaching a modern API, regardless of the model used. Model usage is usually the smallest line, often a few dollars per thousand documents, and it falls as models get cheaper. Review design — confidence thresholds, queues, escalation — is a real cost that most quotes omit, and skipping it is what produces the automations that get switched off after an incident. Maintenance is the last line: interfaces change, so budget for a small ongoing allowance rather than assuming a build is permanent. The honest way to compare quotes is per workflow with measurement included, not per seat.