Product Updates: Various Metric Tools, Languages Support And More

1. Measure AI intent prediction accuracy

Virtual assistants are in many ways like people. In order to stay relevant, they need to consistently learn new stuff and adjust their knowledge. To track general progress of AI trainers and virtual assistants, we have included simple metric to the Inbox training view. This number shows the percentage of the phrases that AI managed to successfully link to the topics. As you can see below, the trainer has validated 2710 phrases and the AI got it right 65,83% of the time.


2. Functional in any European language + Russian + Arabic

English speaking regions enjoy numerous options when choosing a conversational AI solution, whereas non-English users have often been overlooked. We believe in flexibility. Our virtual assistants can operate in any European language and Russian. In addition to that we support Arabic (including major dialects and Modern Standard Arabic). Users can freely speak to the virtual assistant in the languages that were chosen by the company and the AI will automatically switch to the appropriate language.


3. Co-intent understanding

Customer messages come in all shapes and sizes. It is commonly accepted that interactions with virtual assistants should be carried out in a rather laconic and concrete way. In reality, this is not always the case. To improve AI’s ability to understand longer texts, multiple intents and unclear messages, we have implemented co-intent understanding. When the virtual assistant is not quite sure, what the customer exactly had in mind, he/she will be presented with topics that will most likely match with the message.



4. Intent ranking and prioritization based on confidence scores

As mentioned in the previous feature, there needs to be a rigorous way of making sure that the human and the AI talk about the same topic. We have solved this challenge by precise confidence scoring system. An ensemble of algorithms calculate unique confidence score for every phrase in a user message. This score then determines, which intents will be prioritized and presented to the user.


5. Customers can provide comments as feedback to agents

Last month, we talked about the feature where customers can rate their agent after the session. We have now included a possibility for customers to give feedback as comments as well. This extra insight into agents work can be accessed and analyzed through AlphaChat agent stats view.

Leave a Comment

Subscribe to our newsletter for the latest in AI