batteriesincluded.com · Questions & Answers

Our AI scheduling tool keeps making recommendations our experienced staff disagree with. Who gets the final say?

Give final authority to the role accountable for the operating outcome, and make disagreement a supported part of the workflow. Do not force employees to follow a recommendation merely because the system produced it.

EOS emphasizes clarity and accountability. Apply those principles by defining what the tool may recommend, what it may execute, and which decisions require human approval. The AI is not an accountable seat on your Accountability Chart. A person still owns service quality, workload, and exceptions.

Distinguish three situations:

- A recommendation violates a hard constraint, such as required qualifications or an agreed customer commitment. Block it and correct the underlying rule or data.
- A recommendation misses relevant context, such as travel conditions or an unusual job requirement. Let the responsible person override it and record a brief reason.
- Staff prefer the familiar option but cannot identify a specific problem. Compare the alternatives rather than assuming either experience or automation must be right.

For a Lovable or Replit scheduling interface, ask the engineer to display the inputs and constraints behind each suggestion. Include an override path and a record of consequential changes. Those are requirements to build, not capabilities to assume.

Review recurring disagreements as process-improvement evidence. They may expose missing information, unclear priorities, or a flawed recommendation method. Never turn override frequency alone into an employee-performance measure.

Keep automation limited until the system reliably respects your operating constraints. The goal is better decisions with visible ownership, not unquestioning compliance with software.

Category: AI that stays human

← All questions