How do we stop a quick AI-generated change to our internal quoting tool from breaking pricing or approval rules?

Question

Tyler Smith · Accepted Answer

Treat a pricing change as a controlled business-rule change, not a casual prompt. AI can modify code quickly, but it does not know which commercial commitments your team must preserve unless those rules are explicit and tested.

Name one business owner for pricing policy and one technical owner for releases. Use the Accountability Chart to clarify responsibility, while keeping approval decisions separate from the mechanics of deployment.

Before requesting a change, write down the current rule, the intended rule, and what must remain unchanged. Include examples covering standard orders, boundary values, exceptions, and unauthorized actions. These examples should become repeatable checks, not just notes in a chat.

A practical release process should include:
- A separate test environment with representative, non-sensitive records.
- Review of the actual changes, not only the generated explanation.
- Checks for discount limits, rounding, permissions, and approval routing.
- Business-owner approval of the resulting quotes.
- A saved version and a tested rollback path.
- A clear decision about whether existing quotes retain their original pricing rules.

With Lovable or Replit, confirm how your project handles version history, deployment, and database changes. Restoring application code may not reverse a data migration or undo quotes already sent to customers.

Keep the interface simple and train users on the changed behavior. A technically correct release can still create mistakes if salespeople do not understand when approval is required.

The recommendation is to make small, reviewable changes against documented rules. Use AI to speed up implementation, but never treat a confident generated response as proof that pricing remains correct.