Customer service software maker Zendesk on Wednesday launched Satisfaction Prediction, a machine learning and predictive analytics feature for helping companies spot customer satisfaction problems before they spiral out of control.
By analyzing billions of data points from real-time and historical customer-interaction data, Satisfaction Prediction assigns each customer with a score, and that score tells the customer service team whether an interaction is headed toward high or low satisfaction. Thus, the low satisfaction cases can be prioritized, especially for high-value clients.
While it may seem like it's solving a problem that could easily handled by any moderately intuitive customer service representative, Satisfaction Prediction is trying to go a step further by shining a light on customers that are most at risk.
For instance, if a customer service caller uses a specific series of words or phrases, it might make them more likely to take their frustrations out against a company on public platforms such as Facebook or Twitter.
According to Zendesk, the machine learning underpinning Satisfaction Prediction learns from signals that may come before negative customer satisfaction, such as the amount of effort required to solve a ticket, the latency between user and agent responses, and the language used within a ticket.
Those aspects are then paired with the customer's satisfaction rating of the ticket, which allows the company's machine learning infrastructure model to learn from the signals. In turn, the analytics feature is able to predict whether new and updated tickets will elicit a good or bad satisfaction rating.
Satisfaction Prediction is now available in limited beta, with general availability planned for early 2016. Pricing details were not yet revealed.