What do you like best?
Watson Conversation is one of the best NLU service out there. Watson has a friendly web app that make it easy to start building a virtual agent, although you need some programming and machine learning skills to build a really good one. When data is not good enough (most common case), Watson performs better in intent detection with higher precision and recall scores compared to others providers. I believe this happens because he can do very well the linguistic expansion from a few given examples. You can start using the free tier up to 10,000 calls, which turns the adoption easy and cheap. I also really enjoy the ease of consuming Watson API and integrating into backends and internal systems.
What do you dislike?
Fuzzy matching dont works so well in others languages as it does in english, although its still in beta.
Recommendations to others considering the product
You can find a lot of information on google about how to start with Watson Conversation. Go ahead and try it by yourself.
What business problems are you solving with the product? What benefits have you realized?
We are primary using Watson Conversation to build virtual agents and chatbots on customers from different industries (from customer service and bots to HR to manufacturing troubleshooting). We use it in different channels: customs web, iOS and android interfaces, Facebook Messenger, customer service and servicedesk tools.