To set up tests for a conversational AI, it's essential to define clear goals and priorities, such as improving release times and testing user order phrasing. The next step is to create and prioritize use cases, focusing on intent testing and entity testing. Test cases can be sourced from team brainstorming, user testing results, production data, or machine learning generation, with a focus on brainstorming and user testing data for this project. Once test cases are created, they need to be embedded into the code or configuration files, depending on how tests are being run. After that, it's time to run the tests, which may require writing custom wrapper code for NLU models without built-in testing functionality. Finally, the tests can be integrated with the conversational AI workflow, running them whenever major changes are made to the NLU, and potentially integrating them with CI/CD pipelines as the team matures.