My understanding is completely from a 'how hard can this be to...

My understanding is completely from a 'how hard can this be to achieve' perspective in terms of customer value.

In the case of selling insurance, ask yourself what exactly does an online insurance selling bot really need?

Well from the buyers perspective, there's just a few pieces of information I need in tabulated format. If I need anything else I'm going to make a call.

Designing a fancy machine learning system isn't really going to add much to this process. They're going to navigate to the page, check what's covered, check the price and either hit 'back' and go back to google to look for a better deal, or click sign up.

People go to websites to buy things as opposed to real-people because they don't want to be salesman'd into shit, having a robot doing the salesman spiel isn't going to help.

The best case value scenario for this company is if 4chan or something gets hold of it's machine learning and teaches it to be a Nazi insurance salesman like:

http://arstechnica.com/information-...e-neo-nazi-millennial-chatbot-gets-autopsied/

And the story goes viral.

That's the best case scenario.