Imagine if you could intelligently automate responses to support requests without ever involving an agent or needing to write a single business rule.
You get thousands of password reset requests daily. The responses are simple but the high volume is consuming valuable agent time. Or maybe you’re an online retailer that’s constantly sending standard follow-up emails asking for more information about customers’ orders. Or it’s your message portal that’s bogging you down. You want to respond with a standard thank you to each and every piece of feedback submitted, but you’re not staffed to devote agent time to that volume of requests.
As Google’s Smart Reply is doing for consumers in simplifying and automating email replies, Wise Auto Response is doing for customer support organizations that can’t keep up with the ever-growing volume of inquiries.
Announced today, Wise Auto Response helps customer service teams answer common customer questions faster, freeing up agents’ time to focus on more complex, higher-value customer issues.
For example, Wise Auto Response can:
- recognize password reset tickets when they come in and automatically suggest the appropriate resolution so that agents are able to handle more complex tickets or take on other customer-facing initiatives;
- detect purchase-related questions and integrate into a company’s purchase database to determine whether there is an order associated with that email address in order to automate and streamline the follow-up; and
- identify product feedback and suggestions that don’t require direct support follow-up, but do merit an appreciative acknowledgement.
How it Works
Using machine learning to interpret the intent of an incoming customer inquiry and mimic the decision making of your best support agents, the application sends the customer the most appropriate templated response without involving an agent.
(1) The customer asks a question via a text-based communication channel, i.e. social, email or chat.
(2) The Wise application reads the inquiry, infers meaning of that customer’s question (based on language and customer data patterns) and suggests possible answers based upon the most common replies that were used to answer similar questions in the past.
For every prediction being made, the application is aware of how confident it is in the response it has recommended to that customer inquiry, and that degree of confidence is used to decide whether a reply should be sent automatically or suggested to a human agent, based upon the cost and consequences of being wrong.
For example, the common inquiries discussed above have a relatively low cost of being wrong, as they don’t require hands-on follow up. In contrast, inquiries such as those requiring a product refund carry a much higher cost of being wrong, and thus bear a higher threshold for confidence in allowing the machine to reply vs. recommending to a human agent.
(3) High-confidence answers are responded to directly within minutes (the application can apply the macro/template automatically without involving an agent.)
(4) Lower-confidence answers and those requiring manual intervention are suggested to agents. If not auto-responded, agents respond.
(5) Agent responses and any customer follow-ups are continuously fed back to Wise to improve future results and learn how to respond as the world changes and new issues arise.
For a demo of Wise Support Applications and Wise Auto Response, or to learn how machine learning applications can work for your customer support organization, download our Essential Guide to Automating Customer Service.