Customer service and support has been transformed by technology, and analysts predict it will experience even greater change in the coming years as it invests in and embraces artificial intelligence. That prediction took a big step closer to reality this week when SupportLogic announced they had raised an additional $50 million to help expand their AI-powered customer support platform offering.
Your first thought when you hear “AI-powered customer support” may be that of frustrating conversations with robotic voices asking you to repeat yourself ad nauseum. But the true power of machine learning and artificial intelligence in customer support and success goes well beyond voice activated commands.
Two powerful ways that SupportLogic uses this technology, for example, never even directly interact with the customer. The first is through the use of natural language processing to extract an analysis of the customer’s sentiment and key concerns, to help provide a live customer service agent with real-time support and recommendations. By evaluating the words and speech patterns of the customer, the machine can consider the millions of other conversations it has heard to draw conclusions about the customer’s intent. This is a strong instance of the blend of human and computer abilities, as the customer gets the comfortable and seamless experience of interacting with a person, but also benefits from the world of knowledge and insights the computer can predict about the customer’s needs.
Another way SupportLogic deploys machine learning on their platform is by intelligently routing customer inquiries to the right company point-of-contact. Models can be trained by looking back at many examples of prior customer service requests and interactions, and training a computer to learn about the various inputs that led to successful or unsuccessful outcomes. With regards to inputs, companies likely have a plethora of data points about each customer that may influence their experience with a particular customer service interaction, such as any prior interactions, length of time as a customer, lifetime value, most recent billing, most recent customer activity, etc. The computer can then compare all of those inputs with how the experience was ultimately resolved. Did it end quickly? Did it end satisfactorily? How did the customer evaluate it? Did they churn afterwards? By studying the relationship between all of these inputs- combined with the text of their service request, as well as any analysis of the sentiment of that request- the computer can build models that predict how a customer service experience will end, and route that request to a more appropriate agent or team member if it would otherwise likely have a negative outcome.
Given the rapid growth of large-scale tech-enabled companies, the application of machine learning in customer service makes a lot of sense. Every day, companies can encounter thousands of support inquiries across an ever-growing number of channels. A lot of work has been done to digitize and standardize this process, pulling all the threads of a customer’s identity and previous support needs into one central location for ease of management. That is a circumstance ripe for machine learning, which can then consider this vast amount of data to help learn about patterns and trends that can help address customer needs in a way that has an obvious impact on the company’s bottom line, with reduced churn, longer contracts, etc. Customer service platforms also tend to already create highly “labeled” data; most consumers are familiar with the experience of being asked after a customer service interaction to evaluate the interaction and to indicate their sense of resolution, while the customer service agent is likewise required to formally “close” a ticket. This means there are rich datasets of prior customer interactions with clear and standard outcomes that are perfect for machine intelligence to absorb and from which to help offer insights and predictions.
SupportLogic’s growth, however, is likely indicative of another trend when it comes to the application of machine learning: SaaS verticalization. VentureBeat quotes the SupportLogic CEO Krishna Raj Raja as saying, “Off-the-shelf sentiment analysis and entity extraction machine learning models are trained on a completely different corpus and do not work on these datasets...many of our customers initially started down the path of building their own solutions and SupportLogic often displaces these homegrown projects.” This is a growing phenomenon. While completely generic pre-trained machine learning models can make AI more immediately accessible to a bigger audience, they are- like all models- limited by the body of data on which they were trained. That generic data set may or may not make sense in particular settings, and Raja is arguing that machine learning models not explicitly trained on customer service interactions are not sufficiently useful in the customer service context.
But, of course, as he also observes, it is not practical for every company with a customer service function to try and build their own solution. They often lack the expertise or other resources to do so, but they also only have their own customer service data, and so will not benefit from that of other companies where the context may be very similar. So a vertical SaaS solution could be the perfect fit. We should expect to see a lot of machine learning and AI growth following this model in the future.