I’d add one metric that’s easy to overlook: what happens to customer satisfaction after the AI takes over. A chatbot might reduce ticket volume simply because customers give up and leave the conversation, which can look good on paper. I’d compare resolution rate with repeat contacts, escalations, and customer feedback to see whether the AI is actually solving problems rather than just reducing workload.

One thing I’d test before choosing is how easy it is to switch or export your data later. A chatbot can become deeply tied to your knowledge base, workflows, and customer history, so being locked into one platform can become expensive as the business grows. Under $100, I’d want something that solves today’s needs without making tomorrow’s migration painful.

I look at how well each platform handles failure, not just successful conversations. A chatbot will eventually misunderstand a question or lack the right information, and that’s where the real difference shows up. If it can recognize when it’s unsure, avoid making things up, and smoothly pass the conversation to a human with the context intact, that could save a small business more time than another long list of features.

I think PocketBase is one of those tools where the real question isn’t “Can it scale?” but “How expensive is it to outgrow?” If it lets you reach paying customers quickly without creating a huge mess, that’s valuable. The important part is having a clear exit plan before you actually need one.

I’d probably judge it by what happens when the AI gets something wrong. A great demo is easy, but support gets messy in real life. If the AI-native system can admit when it’s unsure, preserve the customer’s context, and hand the issue to a human without making them start over, that matters more to me than how advanced the AI sounds.

“one agent everywhere” is only useful if the customer doesn’t have to realize they changed channels. If I explain my problem on WhatsApp and then have to repeat everything when I move to email or a human agent, the unified system hasn’t really unified much. The real win is shared context, not just shared AI.

the real test of customization is whether you’d still choose to maintain it six months later. It’s easy to add a custom prompt or workflow when it solves today’s problem, but nobody thinks about the cleanup. I’d build AI like a house: customize the rooms people actually use, but keep the foundation simple enough to repair.

I don’t think the real question is “waitlist or no waitlist.” It’s what you do with the attention you get. A thousand people waiting means very little if nobody sticks around after launch. I’d rather launch early, watch what real users struggle with, and build from that. You can create hype later, but you can’t fake real user behavior.

The bigger shift is that AI may not just help us find products,it may start influencing what we actually buy. Instead of comparing 20 products ourselves, we could simply tell AI what we need, our budget, and what matters to us, then trust it to narrow things down.

But I don’t think traditional shopping will disappear completely. Sometimes we still want to browse, see the product for ourselves, and make that final choice. AI might become the shortcut, not the entire shopping experience.