What should a small team test before buying a job-ready AI assistant?

Salesforce is selling AI assistants for jobs like customer service, IT, HR and sales with the starting actions and data model already packaged. That can save a small team a lot of setup. It can also make the tool look finished before anyone has seen a messy case. Before buying, I would give it ten real requests that fall just outside the happy path: a customer with two orders, an employee who changed roles, a lead with a stale promise. Then count what happens next. Did the person handling the exception get a usable starting point, or did they have to reconstruct the whole story from scratch? A job-ready AI assistant earns its renewal when the odd cases get easier to pick up too, not just when the first reply goes out faster.