Home > Columns > CRM Columns

AI: The Final Frontier in Customer Experience or the Portal to a Brave New World?


Presented By: CrmXchange

Over the years, the one constant in the customer care arena has been technological change. It has evolved from the dawn of rudimentary call centers in the ‘60s and ‘70s, and grown via advances in touch-tone dialing and toll-free service that led to IVR technology.  It has ridden the wave of domestic and later international outsourcing that began in the ‘90s, taken new paths with the broadened spectrum of service channels made possible by the internet, and reached higher levels with the introduction and universal adaptation of customer support software and sophisticated routing solutions over the past two decades, the increasing need for mobile customer service and more.

Throughout all this progress, the common thread has been the need for human agents to manage the interactions, deliver the answers, and be the front-line link between companies and their customers. Now the landscape is shifting rapidly and the relentless stream of advanced algorithms is changing the equation. Is this the time that the long-predicted mass migration to AI-based entities – referred to as intelligent virtual assistants (IVAs), chatbots, virtual agents, and a variety of other terminologies – will begin in earnest?

It’s no secret that AI entities can enable organizations to increase efficiencies and reduce costs, making it possible to deliver quality service without having to maintain 24/7 contact centers staffed by employees and contractors all over the world. Gartner has made the bold prognostication that by 2020, 85% of all customer interactions will no longer be managed by humans. In an often-cited 2016 worldwide survey by Xerox (now Conduent), 42% of respondents predicted that with the move toward wider use of AI, the contact center as we know it now will cease to exist by 2025.  “AI is the next wave of the industrial revolution,” said Yi Zhang, CEO of Rulai and an early proponent of the technology.

Billions of dollars are being poured into the development of AI-based customer service solutions and next-gen IVAs with more self-learning capabilities are already making their way to the market. These emerging entities rely on natural language processing (NLP) and machine learning technologies that enable them to grow smarter over time and adapt to customers’ individual preferences as they “learn” from past interactions to improve their understanding of customer expectations and needs.

  luminoso.july2017 According to Dan Mitus, Technical Solutions Manager, Luminoso, AI and natural language understanding specialist, the most effective solutions use AI and NLP to understand language the way a human does. “The goal is having people be able to converse with a machine in a way that makes them feel comfortable … offering a solution that can interpret  historical experiences and maintain context,” he said. “The products that helped introduce AI to the public such as Siri and Alexa allow people to do things that are very transactional in  nature. Ultimately, the goal is to introduce more conversational solutions that use natural language more effectively. That’s a big challenge and much of the NLP research is going to provide deeper understanding that will enable machines to conduct more robust conversations.”  

Understanding customer intent is also an important attribute, with Mitus believing that the ability to do so helps to encapsulate the human experience. “Contact centers offer great insights about a product or service. You’re not going to get any better input than the honest feedback that people provide when getting in touch to talk about their problems,” he said.  “A solution should be able to understand not just what people want but the reasons they want it…the how’s and the why’s.”

At this juncture, the most prevalent use of AI solutions in the contact center is to automate routine tasks and enhance service delivery. “I think back to the days of phone trees, when customers were screaming into the phone for a live agent because the machine didn’t understand them,” said Mitus. “Being able to automate the process has a significant business value, but the solution must meet the challenge of always understanding the customer. In most cases, one single negative self-service experience means that the caller will go back to asking for a live agent every time and the opportunity for them to self-serve on an ongoing basis will have been missed.”

Successful implementation of AI in contact center environments produces the desired savings by reducing the need for live agents to handle calls, as well as freeing them for more meaningful work. In financial institutions, for example, agents have been spared such mundane responsibilities as providing account balances or helping people make transfers or payments.

Despite the current and future benefits of AI, many companies remain on the sidelines. While a recent ICMI/Oracle study revealed 85% of respondents said they would like to see their organizations adopt or expand the use of AI. Accenture research showed that 56% of companies are in what they call the “observer” stage. These organizations do not fully see the transformational or incremental value of AI and are undertaking relatively small initiatives with a “wait and see” approach.

“A lot of businesses have existing infrastructure, processes and people based around systems that they’ve already built,” said Mitus. “It’s difficult for them to make the necessary changes to incorporate AI, but it’s the legacy systems that are inflexible, not the AI.” He believes that organizations need to create a clean path that enables them to move forward from treating contact centers as cost centers to supporting them with AI solutions that can reduce costs and create new opportunities.  “Designing the right process to set up a transitional phase where businesses can seamlessly integrate AI and machine learning applications enables them to leverage all of the capabilities that are now available,” he said.

In addition to call deflection and low-level self-service, Mitus believes that companies should explore AI to make human agents more efficient, put better resources at their fingertips, help them better understand the customer, and automatically detect when a customer needs to be escalated to a supervisor. “These are intermediary steps toward more automated self-service but can add a lot of value toward changing the organizational perception of contact centers as strictly cost centers.”

sundownlogo.aug2017 FabioCardenas, CEO of Sundown AI, whose self-learning Chloe automation layer is designed to help companies answer consumer questions and send notifications via voice, chat and SMS, believes that companies need to start by determining what operations they have that are purely transactional. “At the lowest level these are questions that could be answered by FAQs; at the next level are questions that need to be customized and require database lookup to be answered in real time. The next layer would be taking action on behalf of the user, such as closing a case or sending out notifications. These are the base use cases for incorporating AI.” 

To provide service for more complex products or higher-level information requests, Cardenas believes that companies need to get more advanced systems that can employ machine learning and discern customer intent, analyze patterns and customize information from specific customer profiles. “In higher level systems with machine learning, you are able to train the system to identify intent. The beauty of that is that once it is trained, it is able to deliver intent on any future data that it receives without anything done manually; the system can then provide suggestions on how to improve customer satisfaction.” He cites QA as an example of one area where this is particularly valuable, with machine learning enabling companies to more quickly examine and share scorecards of conversations, rate them all (as opposed to the small percentage reviewed by manual procedures) and compare agent-to-agent performance. This can produce exceptional value in preventing negative outcomes.

Perhaps the most pressing question about implementing AI solutions is how quickly they can produce ROI. While acknowledging that results depend on the size of the training set, on the complexity of the applications and the use cases, Cardenas said that the costs of the Sundown.ai system… which is mainly used by SMBs as opposed to enterprise clients… starts at about the salary of an employee for two months and claimed it could begin producing returns over a 3- to 6-month period. Both ROI and efficiency are improved by an AI solution’s ability to provide 24/7 service over a variety of channels and clear a backlog of cases and service tickets that have been received. For example, Sundown cited improving resolution rates from the mid-40% range to 85-90% over a two-month period for two international clients.

rulai.aug2017 To Rulai, which takes a different approach by providing companies with an Interaction Design Console that enables non-technical users to design chatbots without the need for coding, the decision to implement AI should be based on a cost-benefit analysis. This should be used to determine which use cases are more frequent and which ones would ultimately be the most successful ones for deploying bots. They offer a build-a-task application to help businesses test the use case and offer an adaptive learning mode to help companies improve bot effectiveness. 

Rulai believes in empowering the business user to develop bots without having to rely on internal IT. “The idea is to put the domain expert—the CX leader-- in control of AI,” said Rulai CEO Yi Zhang. “Most mid-market companies do not have much in the way of IT resources nor the budget to implement solutions without knowing whether they are going to produce results. CX leaders want to manage the strategic imperative of integrating AI over the long haul; they want to set the pace for rolling out virtual assistants in conjunction with live agents and how to augment the capabilities of human agents. The idea is to remove the friction in the process.”

“Deep learning originates from a company’s networks and the amount of data available to pre-train the model,” said Zhang. “It starts with general understanding of syntax, semantics and dialogue flows to help in determining customer intent. The additional computing power now at our command makes it more effective. In some areas, deep learning is achieving better performance than human beings in such tasks as sentiment analysis based on text and speech. It’s better at predicting survival rates in medicine, in gambling and computer games. Natural language processing is the holy grail in AI,” she said. “It takes static customer information and integrates it with dynamic changing information and combines it with ongoing customer behavior to help predict intent and maintain context.

Many of the real-world implementations of the Rulai build-your-own bot solution begin at the top of the business chain with a VP of customer experience or contact center operations. Once on board, the executive then appoints a “service design manager,” creating a new opportunity for someone who is viewed as the one of the best and brightest in the contact center, according to Rulai customer evangelist Jim Diaz, “This person is charged with keeping tabs on the best practices of implementing and managing bots. He or she is also responsible for knowing where the data is and how to use it.” Rulai helps to further empower these service design managers with training that helps them effectively become what Diaz called an “AI project manager.” They are taught how to manage the console, building out tasks using simple drag-and-drop techniques and pulling in data. According to Diaz, their main responsibility will be approving recommendations into the console that affect the training and deployment of the bot. Essentially, service design managers monitor the performance of the bot and are aware of situations where the bot needs tweaking to facilitate necessary adjustments.

Zhang cited a recent McKinsey report called “Artificial Intelligence: The Next Digital Frontier?” which estimated the annual investment in developing AI technology last year at $26 to $39 billion (which has tripled over the past three years). She noted that a significant portion of this spending often goes to bringing in AI specialists to make it work. “The average cost of hiring an AI expert for a company was between $5 and $10 million,” she said. Zhang believes that not only can this expense be dramatically reduced by putting AI in the hands of a domain expert who is focused on the customer experience, but the results will ultimately be superior. “The domain expert needs to be the leader, not the engineer.”

Of course, even if alternative opportunities are created for some workers and there is a greater need for higher level agents to address more complex inquiries, implementing AI technology will have a serious negative impact on contact center employment. “Even when a self-learning system needs to escalate for a question that is not currently in the database, the resulting answer is then added to the system’s body of knowledge, which lessens the need for humans to handle these tasks.” said Sundown’s Cardenas. “This will ultimately affect overall headcount, which will also have an effect on rates charged by SaaS providers. Companies don’t want to manage acres of cubicles when there are cost-effective alternatives.” Over the next 3 to 5 years, Cardenas sees a reduction in live agents of 30% to 45%, particularly in areas like India and the Philippines where contact centers handle more repetitive transactions. “Jobs that require a greater degree of expertise should survive in the short run,” he noted. “But over 10 years, who knows?”


There are numerous business cases that can be made on how AI can deliver significant value to companies of all sizes willing to use it in customer-facing applications. In addition, by providing live agents the resources they need to answer more complex inquiries more quickly and accurately, it can contribute to greatly improved service on all levels. For now, most industry observers believe that while the technology has already disrupted the way contact centers are operated and staffed, AI is not yet positioned to displace the need for human agents.