What are acceptance criteria?
Acceptance criteria are the specific, testable conditions a piece of work must meet before the business agrees it is finished, written down before work starts so the buyer and the builder share the same definition of done.
Without acceptance criteria, projects end in arguments about whether they are finished. With them, the conversation becomes a checklist. Good criteria describe behavior that can be observed: when a new web enquiry arrives, a CRM contact is created within two minutes with source and service type filled in; when a field is missing, the record is flagged for review rather than skipped.
For automation and AI work, criteria should cover normal cases, edge cases and failure behavior. What happens with a duplicate client, an attachment in the wrong format or an answer the AI is not sure about? It also helps to set a quality bar, such as the share of drafts a reviewer can approve with minor edits, measured against a set of real examples.
Criteria belong in the written scope before work begins, and they should be tested with real users and real data, not demonstrations. When a vendor or contractor is involved, tying sign-off or payment milestones to acceptance criteria gives both sides clarity and protects the business from paying for work that does not do the job.
Example
For example, an accounting firm commissions an automation that reads client bank statements and prepares categorization drafts. The acceptance criteria require that drafts are produced for 50 sample statements, every transaction is flagged when confidence is low, nothing posts to the ledger without staff approval, and client files never leave the firm's approved storage.