Skip to main content

From command to context and conversation: how conversational AI will reinvent the call center

(Image credit: Image Credit: Georgejmclittle / Shutterstock)

It’s a frustration we’ve all faced: we are trying to get help from a call center - let’s say a mobile phone service - and we can’t get what we need. At first, the voice service can’t understand what we are saying: “Please help me reset my password.”  We repeat the question, but we don’t get the answer we want this time. We repeat it again and get the same answer. Growing increasingly frustrated, we start hitting buttons and ask to talk to a human. Even when we reach someone minutes later we don’t get the answer we want. We haven’t been able to reset the password, and we are agitated. 

It’s little wonder that while the call center is ubiquitous it is also rife with problems: poor customer service, skyrocketing costs, constant employee turnover. “Conversational AI” - a system where customers can talk to virtual assistants in a lifelike way and learn from previous interactions to provide markedly better service - promises to end these frustrations and reinvent the call center. The upsides of this are enormous: human workers in call centers will be freed to focus on thornier problems, customers will get quick answers to their problems and companies will save countless millions while actually improving customer service - some estimate a savings of as much as $8 billion over five years. 

Conversational AI is already in use in many households. Alexa might be helping you stream music. Google recently announced Duplex, an AI system able to handle a range of telephone conversations like booking appointments or placing orders. Technology that can power a “conversation” that results in effectively completing a task with multiple back and forths is already with us in some contexts. Soon, it will be widespread in the often disparaged call center. Here are the key reasons this technology promises to completely upend the call center - for the better.

Conversational AI Will Be Able To Mine “True Intent” 

One of the biggest customer frustrations with call center voice assistants or even human employees is that they don’t understand what you want. Conversational AI promises to change that.  We are moving away from first generation simple Q&A bots that respond to basic queries to “virtual assistants.” As a result, the technology requirements are many order of magnitude higher. Simple voice assistants were limited to easy questions like: “I need an insurance quote” or “I need to get a new pair of tires.” Virtual assistants can hold a conversation and understand multiple steps and clarify, authenticate, disambiguate, present users with multiple relevant choices, integrate with backend applications and - most important - generate meaningful responses. 

We call that finding a customer’s “true intent” and it’s one of the biggest business challenges for companies and their call centers. Virtual assistants, however, can engage in full conversations and draw on previous interactions to determine that when a customer asks “can you tell me about my insurance coverage?” that might be an existing customer that needs to file a claim. As a result, the customer call is routed to the appropriate location without human intervention.  The customer is happy and the company saves money.

Drawing On “Memory” To Answer Questions

One of the key aspects of human conversation is “memory” - knowing and preserving what someone has said in the past. In most cases it is also personalized  based on previous conversations or, in the case of call centers, the status of the user (new user or existing Gold card member).  

Let’s take the a simple example of a retail customer who reaches out to a call center with a simple question: “My laptop is not working. Please help.” The response to this question depends on the user and their past interactions and status. A virtual assistant, however, is able to navigate through multiple conversation paths from “troubleshooting” to “returning the computer” to “suggesting offers for new computers.”

The importance of vertical domain knowledge and its applicability in understanding what the user is saying cannot be understated.  For a virtual assistant to be able to hold a conversation the problem set has to be constrained to a domain (even better if it’s a very specific subdomain). Certain terms have different meanings across domains and some of them have different implications if not interpreted correctly. For instance, “term”, “policy” and “copay” have different implications depending on whether customers are reaching out to a bank call center or an insurance call center.

Virtual assistants draw on a rich history in each domain. For example, the Avaamo retail banking domain has more than two million banking specific utterances and more than 1,000 intents. That information is used to understand, comprehend and respond to users in the right way. An average sentence of 13 words requires three trillion variations for deep learning to pinpoint true meaning. By focusing on specific functional areas within an industry (e.g. “quote generation” in insurance) and combining curated datasets with specialized algorithms for sparse datasets, it’s possible to pinpoint thousands of industry-specific intents and entities. What that means is a quick, painless customer interaction with  a virtual assistant that leads to a quick resolution for a customer.

Virtual Assistants In Call Centers Will Understand Language And Context 

Most large businesses that utilize call centers are global. They have customers, employees and partners in various regions where English is not the language of choice. However, many call centers still have limited options when it comes to language. Call centers should expect that conversations might be in Spanish or Portuguese or Mandarin or even in mixed languages like Hinglish and Banglish.  Avaamo in particular has pioneered a “language neutral” approach to representation of training data that helps us quickly make the assistant available in different languages with minimal language training. We also have translators that let users access the assistant through different channels. The more a customer reaching out to a call center can be understood in their desired language, the less the call becomes an issue requiring precious employee time. 

Conversational AI Helps Reach Desired Goals

A key aspect of holding a conversation is ensuring virtual assistants drive users towards their desired goals.  The intent of any virtual assistant  is to perform a set of tasks that are directing the users towards a goal. Whether it is to troubleshoot a problem and resolve a customer issue or recharge their mobile phone or get them through a loan process - the conversation flows follow the natural flow of the business. When the virtual assistant clearly understands what specific goal or task needs to be accomplished it can it can quickly resolve intractable problems. Call center employees will still play a critical role. Virtual assistants will take on the mundane and repetitive and even transactional conversations, freeing up human agents to deal with interactions that require more empathy, complexity and business value.  

The next generation of virtual assistants will help shape contact centers that are truly customer oriented, focused on making each interaction more efficient and effective. The technology to better understand, converse and resolve customer issues is also getting better each day. What that ultimately means is fewer frustrated customers who spend hours getting an answer and better, more cost-effective business across industries.

Sriram Chakravarthy, Founder and CTO of Avaamo (opens in new tab)

Image Credit: Georgejmclittle / Shutterstock

Sriram Chakravarthy. Founder and CTO. Sriram brings over 15 years of engineering experience in designing massively scalable distributed software.