Support agents have a lot on their plate. Agents must respond to a wide variety of questions, constantly flexing their problem-solving muscles, while also working to keep the customer satisfied.
Sensing whether a customer is satisfied in such a fast-paced environment is a challenge, especially considering the number of factors that can influence how a customer feels, including the amount of time they’ve waited, how many times they’ve contacted you, or even the language you use to resolve their issue. Fortunately, when it comes to customer satisfaction, support agents no longer have to rely upon their powers of intuition and foresight.
Introducing Satisfaction Prediction
Satisfaction Prediction, built into Zendesk, provides support teams with a prediction into how likely it is that the customer will be satisfied with the service they receive. Rather than waiting for feedback to follow an interaction, this prediction can be used proactively to start the interaction on the right note, and to ensure a positive outcome.
The Satisfaction Prediction tool uses machine learning to read signals from your interactions and calculate a predicted satisfaction score between 1 and 100. Hundreds of signals—including text description, the number of replies, and total wait time—are factored into a unique model to dynamically calculate the predicted score at the moment the ticket enters the queue.
Armed with this prediction, you can:
- Save time triaging tickets by using the score to quickly pinpoint tickets that are most at risk
- Go into conversations with more context about the health of the customer relationship and adjust your approach to ensure a positive outcome
- Power workflows using the prediction score by automatically escalating tickets that are at risk
- Analyze your customer support performance alongside your prediction scores to identify areas for improvement
Satisfaction Prediction is now available to customers on the Zendesk Enterprise plan who collect a minimum of 500 satisfaction ratings per month. The tool requires a combination of both good and bad ratings in order to build an accurate prediction model.