I got an email today about FreshDesk’s Customer Sentiment Predictor, a new feature their Data Science department just installed. Its primary task is predicting your customer’s mood. It displays an emoticon on the customer’s profile image depending on whether the customer is Very Happy, Happy, Neutral, Sad, or Angry. Imagine your inbox with each email having an emoticon on it. The emoticons made me smile. It is simple and visual.
I tried to look at their predicted moods and then read the customer’s ticket to see if I agreed with their prediction. They got some of the moods correct, but a lot of the tickets seemed neutral – or maybe I’m just not good at reading moods from what our customers write.
The new feature lets customer service administrators correct the mood for a particular support ticket. It then learns from these corrections to improve its prediction model.
The feature can help prioritization of which tickets need immediate attention. Also, customer support teams can have more experienced reps handle the angry tickets.
Language is a tricky beast to study though – especially for a machine. It’s not as simple as numeric data – if the customer rates your product five stars, you know for sure that they really like it. IBM’s Watson had to study years worth of Jeopardy questions to be able to develop a prediction model for getting the answer correctly. I imagine that it has tons of processing power to use as well.
My 2 Cents
I am excited to see different industries finding practical applications for Data Science – specifically Predictive Analytics. It is really an amazing field! We collect so much data now and this resource can teach machines to recognize what normally only people can determine. Studying past support tickets to determine if the current one is of a certain sentiment is a good feature.
It helps our customer service staff be more efficient and effective in what they do. And that is a great goal for any project – not just Data Science ones!