“This call may be monitored or recorded for quality assurance purposes.”
Customer service calls often begin like this, but in practice only a small percentage of calls are analyzed. “Call center calls are recorded, but on average supervisors only have time to review one to five calls per month per agent, which represents only a small percentage of calls” says Cogito cofounder and CEO Joshua Feast, a graduate of MIT’s Sloan School of Management. “Agents may go weeks without feedback, and it’s often subjective, making it difficult for agents to improve their performance” Cogito’s solution is an AI-driven coaching application that analyzes conversational dynamics within phone calls and provides live guidance. The application extends agent intelligence, helping them better recognize and respond to the subtle behavioral cues expressed by customers. It also delivers an instant measure of customer perception for every phone interaction.
The employee feels reassured that they have a means to improve, and the customers are happier.
“The software helps agents become more charming and emotionally intelligent – to listen better and build rapport,” says Feast. “It addresses the most challenging part of being a front-line service professional: dealing with difficult, emotional customers.” Cogito reduces both customer churn and employee churn, says Feast. “If agents don’t get timely feedback or help build rapport with customers, they can burn out very easily,” he says. “With Cogito, they have a window into the customer’s sentiment and live guidance to help them properly adapt. The employee feels reassured that they have a means to improve, and the customers are happier.”
Sharing the caring Feast didn’t start out with the goal of improving call centers, but after growing up in friendly New Zealand, he felt the world could use a bit more caring and charm. “When I came to MIT, I was interested in the notion of caring, and how can we have more of it,” he says. “How can technology help create better relationships and help each of us be a better version of ourselves? I grew up in an entrepreneurial environment, and was interested in building a business that combines the best ideas in science and technology to drive positive human impact.” The MIT Media Lab was the ideal place to advance these goals. There, Feast learned about research on relationship dynamics that had been under development for years by Professor Alex (Sandy) Pentland and his Human Dynamics Lab. Feast was intrigued by Pentland’s “Honest Signals” research into understanding human behavior and interpreting psychological states. Pentland, who eventually co-founded Cogito with Feast, “had shown it was possible to analyze voice and understand the markers of distress,” says Feast. “The research indicated voice and other behavioral feedback provided excellent insight into a person’s intent; for example to detect if someone was experiencing distress or losing interest.” Based on this research, Feast and Pentland received funding from DARPA and the National Institute of Mental Health to launch Cogito into R&D mode in 2007. Four years later after extensive development and the capturing of millions of data points, the first version Cogito’s commercial software was released. The initial focus was to analyze conversations to offer real-time coaching for clinical use cases. “We wanted to help nurses and psychologists recognize distress in vulnerable patient populations,” says Feast. To help commercialize the software, Feast called upon the MIT Venture Mentoring Service and the Martin Trust Center for MIT Entrepreneurship. After an initial roll-out in healthcare, Feast decided in 2015 to extend the business to sales and service conversations in call centers. Today, Cogito is used by a variety of large banks and insurance companies. “Our customers have seen very substantial improvements in customer and employee satisfaction,” says Feast. “We have also demonstrated efficiencies like shorter call times and fewer repeat calls.” The Boston-based startup was recently named as one of the prestigious MIT Startup Exchange STEX25. “MIT ILP and STEX25 are helping us build relationships with large customers,” says Feast. “Cogito sits on the shoulders of giants in more ways than one.” Conversations are like a dance Every day Cogito streams millions of calls into its high performance computing cloud platform, and scans for behavioral signals based on hundreds of measures across the voice spectrogram. The signals are synthesized into behavioral models, or as Feast puts it, “specific things that are psychologically relevant.” The application then presents in-call guidance to help agents modify their behavior for better outcomes. The Cogito software can detect pressured speech based on agitated, accelerated speech patterns, or it can detect a voice under distress. It can also identify and encourage positive indicators such as tones and patterns that reflect empathy. One goal is to guide agents toward speaking with consistency “which helps you come across as confident and in control of your topic,” says Feast. “You also need to control the tension in your vocal cords -- relaxing them makes a big difference.”
Cogito scans for keywords as part of the analysis, but these are far less indicative of a conversation’s success than “interaction patterns,” says Feast. In part, this is because reps often follow scripts that have been carefully vetted for maximum impact. “The reps are generally focused on the words and lose track of how they’re coming across,” adds Feast. “If you go to a café in another country and observe people speaking, you can probably tell how the conversation is going even if you don’t understand the words.” Cogito’s AI algorithms not only analyze each voice in isolation, but also evaluate the conversational dynamics. “A conversation is like a dance,” says Feast. “If you are out of sync with your partner, it is very observable.” One example is that neither side should dominate the conversation. “It’s very important to ensure both parties participate in an effective conversation.” The software can also help sales and service reps improve their “person perception” – the ability to recognize and respond to social signals. “We all think we’re great at our ability to effectively perceive others, but a lot of us are not,” says Feast. “To be more charming, you need to recognize social signals, correctly interpret them, and respond to them appropriately. The most important thing is the recognition – realizing when something is happening and acknowledging it to the other party. Even if you are not exactly correct, it engenders trust and builds rapport.” Cogito’s staff spends a lot of time optimizing the presentation of tips to agents. “Our cognitive psychologists work to understand what behaviors are relevant, and how best to present feedback in real-time,” says Feast. “Since we have united the measurement system with the means to improve, we always know the impact of our guidance has on behavior. We try to offer advice in a positive and consumable fashion, and experiment with different notification strategies to ensure we always optimize impact without overwhelming the agent.” Cogito’s analysis of different conversational strategies and outcomes not only helps the software continually improve via machine learning, but it also provides key insights into customer behavior and their perception of an interaction. Objective behavioral analysis of conversations is far more comprehensive and timely then what can be gleaned from a traditional survey. Cogito’s analysis is language independent, and has been deployed in a number of countries. “Most of the important signals in a conversation are universal properties of humans rather than of language and culture,” says Feast. “The communication of attitude or distress is rooted in ancient brain systems that were developed long before modern language.” Cogito’s goal of encouraging relaxed, yet professional conversations sometimes conflicts with pre-existing call center scripts. “Existing call center tools are focused on helping agents follow policies and procedures and capture customer data, which can often make them appear as if they are just a database on the phone,” says Feast. “If you’re calling customer service, you’ve got a complex problem, and you’re looking for a trusted human interaction. Often, the agent can’t create a connection because of the structure placed around them. The best companies empower front line employees to deliver more human, empathic experiences.”
At its base, Cogito is an extremely high performing behavioral signals processing platform that is agnostic to the type of signal. Our research suggests that voice is the most data rich source of behavioral signals, but our platform can consume signals, execute models, and present feedback in many forms.
When asked if Cogito can remain relevant in a world where chat bots are starting to replace human agents, Feast argues that today’s AI is not sufficiently advanced. “The problems our customers deal with typically concern complex health or finance issues, and the current AI can’t handle that on its own. Many customer service problems are solved by a question answer coevolution process, which computers aren’t great at yet. Even if an AI could eventually handle this, I believe we will always prefer talking to fellow humans about complex or emotional issues.” Cogito mainly targets service, but it can also be used for sales, usually with little modification. “The line between sales and service is very blurred,” says Feast. “We’ve had tremendous success on inbound sales, such as customers calling in to buy an insurance policy or for existing customers wishing to add new services to their existing plans.” Cogito may eventually release specialized versions of the software optimized for negotiations, meetings or counseling sessions. People often ask Feast if Cogito could develop a personal assistant to advise them on speeches or phone conversations. When asked if Cogito could eventually expand to integrate facial recognition for use in video conferencing, potentially even in a Google Glass like encounter, Feast suggests it’s possible: “At its base, Cogito is an extremely high performing behavioral signals processing platform that is agnostic to the type of signal. Our research suggests that voice is the most data rich source of behavioral signals, but our platform can consume signals, execute models, and present feedback in many forms.” Yet, more than technological challenges are involved. “Ultimately, we believe that people should have a real-time coach for all their important business and personal conversations,” says Feast. “But the key questions to ask for each use case are what positive behavioral changes can we drive and what is the best way to deliver information so people can easily make use of it? The potential for expanding to personal use cases is enormous, but right now we are focused on an application that is incredibly impactful for millions of phone professionals and the hundreds of millions of customers they serve.”