First time poster here, and in full disclosure, CEO of Polly Portfolio.
I think Efi has outlined the major potential applications, but it’s also useful to touch on why chatbots are important, i.e., why are chatbots going to be used for these applications? People talk about chatbots like they’re a unified phenomenon, but in reality, they’re at least four different things:
A natural language interface for the user: this creates the opportunity to service users without requiring them to “learn” your UX. It also gives users the possibility of making a very wide range of requests. Unfortunately, living up to the full promise of a chatbot in this regard requires very advanced language parsing logic, which in turn requires both pretty advanced AI, large sets of training data, and semantic knowledge as well (i.e., something to tell the system words mean and how those concepts are related).
A natural language interface for the service: a chat also happens to be a fantastic format for information that needs to be delivered in narrative form, like explanations. The service can deliver information, and then the user can ask for clarification on specific elements. (Just like a real conversation!)
Low threshold, near-native application access: no website to register for, no app to install, so users can access it immediately.
Contextual knowledge about the user: the chatbot can know about the user’s transaction history, so the user can say things like “I want to dispute the credit card charge from yesterday.” The tricky thing here is all about getting the permissioning for user data right. I’d be unpleasantly surprised if, say, a trading chatbot started scolding me for spending too much money on my credit card!
Transactional applications (executing trades like AJBellYouth or our friends at Capitali.se) and reporting applications (like myKal) primarily rely on facets #3 and 4.
Marketing applications primarily rely on facets #3 and maybe 1 and 2.
Engagement applications rely on facets #2, 3, and 4.
And customer service applications end up relying on all four facets.
I would say that our chatbot is essentially an engagement application that gradually builds up knowledge about the customer to better surface customized investment ideas. We explicitly decided to rely almost entirely on structured interfaces for our application because we didn’t think the technology was ready for allowing natural language requests from the user (and in our case, that wasn’t critical to a good user experience).
I’m curious as to what others think the limitations are on the expansion of chatbot applications.