I’d be interested in how these platforms handle situations where the AI can’t confidently answer. Good escalation seems just as important as getting the first response right, especially when a customer is already frustrated.

The biggest thing I’d want to test before paying is how accurately the chatbot handles real customer questions. A cheap chatbot that gives confident but incorrect answers could create more problems than it solves.

I’d look for tasks where people spend a lot of time collecting information, making decisions, or doing repetitive work. If AI can reduce that friction in a way that’s actually measurable, that feels like a strong starting point for an AI SaaS.