A retail company plans to replace a manual returns triage process with a Claude-based assistant. Leadership has asked for a defensible answer to whether the project is worth funding, and the team has only a general belief that the assistant will be faster than the existing staff process. Which step belongs at the start of the life cycle, before any build work begins?
- ADefine the use case and its success criteria, including the baseline the current manual process achieves and the measurable target the assistant must beat. Correct
- BSelect the model and set its generation parameters first, since the choice of model determines what the assistant is capable of doing for the returns team.
- CBuild a Model Context Protocol server exposing the returns database, so that integration work is finished before the assistant's behaviour is designed.
- DWrite the production deployment and rollback procedure now, so the operations team can approve the release path well ahead of the first build milestone.
Why A is correct: A defined use case with a measured baseline and a target is what makes the funding decision defensible and gives every later phase something concrete to test against.
Why B is wrong: Model selection does bound capability and has to happen eventually, but choosing one before the success criteria exist means there is no bar against which the choice can be judged.
Why C is wrong: An MCP server is the right shape for a reusable integration, but building it first commits engineering effort to a data surface whose requirements have not been established.
Why D is wrong: A rollback procedure is genuinely needed before release, so this feels prudent, but it is a later-phase artefact and cannot answer whether the project should be funded.