Home > Columns > CRM Columns

Blurred Lines - How AI and ML are Enabling Human Agents to Be More Effective and Responsive


Presented By: CrmXchange

The opening lines of Robin Thicke’s hit song from several years ago, “Blurred Lines,” echoes a common complaint of many consumers, “If you can’t hear what I’m trying to say…if you can’t read from the same page.”  Throughout the history of contact center engagement, agents have had to rely mainly on their instincts to interpret what customers were asking for and use their multi-tasking skills toggling through multiple screens to try and find the right information to resolve issues.

While there have always been traditional call recording solutions to monitor a portion of conversations, coaching and training applications to provide guidance, and rudimentary knowledge base solutions to help locate answers, in most cases agents conducted interactions with limited insight. 

A steady flow of technological advances continue to influence how business is being done, companies are actively exploring emerging AI, machine learning, and predictive analytics tools to give agents the wherewithal to operate more efficiently and eliminate issues that have traditionally hampered their ability to respond quickly and accurately.

Dani Apgar, Executive Vice President and Co-Founder of RapportBoost.AI, a Los Angeles based, 2-year old start-up focused on optimizing chat, SMS and other non-verbal communications, put it succinctly. “There are millions of different directions that a customer service agent can take a conversation,” she said. “Agents had previously been limited by their own cognitive abilities and personal biases. Machine learning algorithms have been trained by millions of interactions (in our case, chats), more than any human could possibly digest.  These algorithms help establish a set of best practices.  This means that companies are no longer shooting in the dark or having to throw darts at a dartboard. They’re using guidance informed by science to determine the best ways to assist a customer.”

However, she also pointed out that existing best practices should not be construed as license to provide one-size-fits-all service. “Every company is unique and every customer is unique. Machine learning helps personalize interactions by placing individuals within a broader group of customers, not just demographically but in terms of what they are asking for. It can help determine what type of personality they have, and what their communication style is. Machine learning helps show agents the right way to engage a person on a one-to-one basis. For example, you wouldn’t engage a 40-year old businessman in the same conversational tone as you would a 20-year old millennial.”  

Kaan Ersun, SVP Marketing, Solvvy, sees two big changes that will allow companies to move from offering what he sees as reactive service to advancing to proactive service. Ersun sees the first difference coming at a macro level.  “Companies will be able to see the emerging trends from their ticket/case history and be proactive in resolving customer issues as well as solving problems at the root cause (such as fixing a faulty product design) instead of treating the symptoms (replacing individual units that don’t work)” he said.  “We provide this functionality to our customers by auto-categorizing incoming tickets and identifying emerging trends and surfacing strategic insights.”

He sees the second change as taking place at the micro level. “Instead of responding to the support issues/requests by an unsatisfied customer, companies will be able to use predictive analytics to determine which issues are likely to arise or whether a customer may churn and be able to proactively respond in a personalized fashion, which provides a competitive advantage. Since these issues often vary from week to week, it’s important for business to keep up with the changes.” He envisions a future where companies will be in a preemptive position where they will see an issue arise early enough to be able to call the customer before the customer contacts them but believes that is two to three years down the road.

Ersun also believes that it is important for contact center operations to deploy technology that puts all the necessary portals and processes into one simplified platform. “It creates a single repository of clean and accurate data that allows accurate record keeping and provides the necessary training data and contextual data for the AI systems to learn from and utilize to operate correctly,” he said. “It will also allow agents to operate more efficiently as they need to train on and operate out of a single unified platform.” 

One aspect of interactions in which AI and predictive analytics can make inroads is in determining customer sentiment. “Learning to recognize and measure the emotions, behavioral and interaction patterns in the conversations that affect their bottom line can make major difference for businesses,” said Jason Ferrell, VP of Product and Strategic Advisor for Behavioral Signals. The company, based in Austin, Texas, has developed algorithms that can analyze human emotions and help to predict behaviors.  “When provided with a higher level of insight, a $20-to-$30 contact center representative can sound like the CEO of Apple in a conversation with a customer,” said Ferrell. “This doesn’t mean that a customer will necessarily come away thinking that they’ve been talking to Tim Cook. When emotional intelligence is properly gauged, a business can pick up unconscious vocal signals that reveal agent incitement, customer intention, and can accurately predict the outcome it needs to make real progress.”

Ferrell believes that most organizations are still measuring only basic metrics, such as talk time, when they could be going much deeper. “Only a few companies are digging into content in a meaningful way,” he said. “Too many are still relying on manual labor to evaluate a fractional representation of calls. I spoke with one company who said they were planning to bring on four or five more people to listen to calls…in New York, that’s about $250,000 a year. By contrast, an AI-driven quality monitoring solution can evaluate 100% of calls.” Ferrell understands that it’s easier for some companies to justify such added salaries than to make the commitment to a technology investment. “Even when it can make a significant difference in operational efficiency, cost is a deterrent,” he said.  He analogizes it to companies being wary of moving to VoIP five years ago. “It’s a scary transition.”

While the Behavioral Signals solution offers the added dimension of gauging emotional response, it is also, as are most of the offerings being used to improve employee responsiveness, a conversational intelligence tool designed to analyze the interplay and measure the quality of the dialogue between customers and agents.  “Many call recording products use one-channel recording but two-channel recording offers superior capabilities. It offers diarization technology that makes it possible to determine who is speaking when. This makes it easier to determine the emotional direction of a conversation. When old-school technology is used that lumps all the words together, these nuances can be missed. When backed by conversational intelligence, an effective agent can turn around an initial negative tone and influence the conversation to a more positive place.”

RapportBoost.AI’s solution takes on the difficult task of interpreting consumer sentiment in non-verbal conversations. “Contact centers have come to realize that not only do customers—particularly millennials—prefer chat, but that chat is a far more cost-effective channel than voice. An agent can engage with 3 or 4 customers simultaneously,” said Apgar. “But establishing trust and building a relationship are every bit as critical as they are in voice interactions. In our own analysis of millions of chat conversations, we’ve identified three imperatives: rapid response time, creating a rapport with the customers and being human. We believe that customers in chat conversations want to know that the agent is a real person and that they are not talking to a bot. This also applies in conversations on SMS and other messaging applications, although chat is a more in the moment channel.”

Apgar points out that sentiment analysis models have existed for decades. She mentioned a recent conversation with RapportBoost.AI’s Chief Data Science Officer and Co-Founder, Dr. Michael Housman. She was surprised to learn that most tools for sentiment analysis were originally built from movie review databases. “From its humble beginnings of gauging people’s feedback about romantic comedies, sentiment analysis has evolved dramatically over the past few years to becoming a readily available tool trained specifically on language for evaluating forms of conversational commerce, such as chat and messaging. It enables us not only comprehend the meaning of emoticons (symbols like a colon and parenthesis), emojis and the shorthand language favored by millennials such as ‘LOL’ and ‘OMG’, but puts them in context within a conversation.” Apgar estimates that about one half of businesses allow the use of emoticons within customer conversations and one third enable the use of emojis. They train their sentiment analysis tools on different models to suit the needs of specific companies. 

But while non-verbal communication channels have gained momentum, voice is still at the top of the heap. Mike Allen, Senior Vice President North America for i2x̅ an AI company headquartered in Berlin, Germany that is now making its presence felt in the US market, cited a Salesforce study that noted that phones still handle around 68% of all contact center communication and another unrelated survey that found that 85% of people who call in hang up feeling dissatisfied. 

Allen believes that the most significant reason for this disconnect is that by i2x̅’s estimate, about half of agents don’t have the information they need at hand to do their jobs properly. The effort put into staff training, and the frustration resulting from not getting consistent quality in customer interactions throughout an organization, is what inspired the creation of i2x̅ in 2010. 

Allen feels the key to better results lies in using technology to improve the human experience. “We strive to find ways to shine the light on the black box that is the voice of each customer. It’s not simple for humans to take the data for insight that makes for more effective conversations,” he said. “We take the best of what machines do: going through data to deliver agents customer insight in real time while closing gaps in knowledge and repeatability to resolve issues more quickly. This can then be combined with human insight, intuition and experience to help our users achieve their full potential.”

The company uses Natural Language Processing software to bring real-time text-to-speech transcription, agent coaching, predictive analytics and human understanding to call center voice communications. Allen credits the dramatic increase in available computational power and network strength for enabling conversational intelligence that allow companies to listen and make more accurate predictions of business outcomes, find patterns in word usage and make coaching suggestions in real time. “There are 5 to 7 key patterns for companies to be aware of that will produce either the outcomes that they want or those that they don’t want,” said Allen. “Conversational intelligence tools can help companies understand what is happening every day on every call. They can then use relevant information as opposed to guessing, improve such crucial KPIs as conversion rate and CSAT score, expand their knowledge base and decrease agent churn.”

Whatever channels a company might choose to provide service on, the right conversational intelligence solutions can have a positive impact on the outcome. But as Solvvy’s Ersun points out, it’s more important than ever for companies to have the most comprehensive picture of the customer across all touch points. “Tools and technologies used on each channel should be powered by and have access to the same customer data and backend, so that service outcomes are consistent wherever the customer makes contact. By delivering a consistent UI and service experience, customers can get the help they need on the channel that is most convenient for them.”