Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management

Unlocking the future of customer service: AI-powered virtual agents and the science of intent

Read time:

P.V. Kannan

More than 20 years after the Internet went mainstream, you believe a new transformation is beginning. Please describe it.

In the 20 years since companies started creating websites, virtually everything about the way we interact with businesses has changed — from movies to books to how we book hotels and rent cars. The only thing that hasn’t changed dramatically is customer service — until recently. Today, the smartest of chatbots — virtual agents — are being powered by AI and connected to a customer’s complete past history. As a result, they can anticipate just what the customer is looking for and answer questions through chat, on the phone and through smart speakers like Amazon’s Alexa.

This will change everything about how companies and their customers interact. Virtual agents will enable brands to have unlimited conversations with customers and allow customers to get what they want very quickly — which, after all, is what everyone wants from customer service. The business world will be transformed with efficient, scalable service that is available 24/7 and gets smarter every day.

Virtual agents will be tasked with determining customers intent. Please define the science of intent for us.

This is the science of figuring out your consumer’s intent as quickly as possible and addressing their needs accordingly. Every customer calls you or starts a chat on your website for a reason. That’s pretty obvious, but not many companies systematically track all the permutations and combinations of the requests they get. Today, because of all the data and signals flowing through, the possibility exists that when you see a customer approaching your website in real time and clicking on a particular area, you can accurately predict what they are there for and offer it up proactively — even before they ask for it.

For example, say you go to TripAdvisor to research a vacation to Greece. You have the dates in mind and you find a property you are interested in, but you wonder if you can use your points to cover the accommodation. So, you head to the points section of your credit card company and automatically, a message appears on your screen asking, "is this about your upcoming trip to Greece?" This sort of thing will become very common in the coming months and years.

How can intent be determined?

Intent is knowable. Based on context — who is calling, where they are located, what time of day it is, what the weather is like and lots of other information — a virtual agent can rapidly identify what you are looking for. Of course, anyone who has ever tried to figure out another human being’s intent knows how easy it is to be wrong. That’s why someone needs to teach the system powering the virtual agent about the characteristics of the business in question and how to model it. Then that system needs massive amounts of data to practice on. After all that, the system can develop an algorithm and use it so the virtual agent can make a good guess about what you, the customer, are asking for. Of course, even if that algorithm is effective, it can never be static. As business conditions, products and pricing change, the algorithms powering virtual agents must keep up.

What is an example of a company that uses these new tools effectively?

United Airlines does a great job with its 1-800 number. If people call on the day of their flight, virtual agents can use data to identify these callers and recognize that, chances are, they are calling to cancel. The agent can open the call by saying, Hi Karen; I see you are travelling to Boston today. Do you want to change the flight time or cancel this trip? This saves time and makes for a smooth experience — as opposed to calling another carrier and being asked for your booking number, which you then have to scramble to find. Upon taking the call, United already knows who you are and which flight you are booked on.

You write that “identifying problems is much harder than solving problems.” What are the implications for AI?

Companies have started to invest in and attempt to understand each component of the customer journey. For instance, once they know that the average Greek vacation-seeking customer starts out on TripAdvisor and then proceeds to a hotel website, they know that you have picked your destination — so they aren’t going to ask, "how about a trip to Italy?"

You believe that virtual agents will actually make human agents better. How so?

The tools exist so that, if you need to cancel a trip and are feeling emotional about it, the AI will direct a human agent to take the call rather than an automated system. Obviously, a machine can’t provide empathy or say, "I am so sorry you have to cancel your vacation." Human agents are going to be better equipped than ever to handle customers and their needs.

What percentage of customer-facing companies are embracing these tools?

I would say about 50 per cent of companies are either actively engaged in this or are trying to figure out how to make it work for them. But sadly, many of them are doing it wrong. They aren’t taking the time to ask, what is the problem we’re trying to fix here? This is not about simply plugging in a bot on your website.

Just because you can now have an automated conversation in real time does not mean that you should apply it to every transaction. When customers need guidance, conversational commerce can be very helpful. In these situations people want to be given options; they might want some expert advice, or to know what other customers are doing. But in 90 per cent of cases, when we visit a website we know what we want to find or what task we want to accomplish. There is no need to have a conversation about it.

Many people are wary of AI. Should we be wary of virtual agents?

Every technology comes with its own problems. Without the Internet, we could never have so many data breaches. Social media vastly elevated the spread of fake news. Mobile device addition has raised rates of teenage anxiety and depression and increased auto accidents. While the Internet and mobile phones have created far more good than harm, a responsible thinker has to ask, what are the potential downsides of virtual agents and chatbots?

Just as with email or social media, bots can be used for negative purposes. That is a worrisome concept, because with AI behind them, they may evolve to be very good at charming people out of their bank account information, for instance, or persuading them to vote against their interests by manipulating their emotions. 

If you’re on Facebook, you have likely received several random friend requests. These are fake friends, and they’re there to create a fake relationship with you in order to get you to send them money. Right now, an intelligent person can spot these, but in the future, AI might be able to make them particularly attractive to you based on your personal tastes. These sorts of risks have already led to a 2018 California law that requires all chatbots to identify themselves as such. I support this idea. No one should be confused about whether they’re talking to a human or a bot.

Another risk is that thieves could hack chatbots to encourage them to request people’s passwords or bank account information. Embedded in a conversational flow, such requests might seem natural. To protect against this, companies must prioritize security in their bot design and development protocols — just as they ought to be doing with their storage of customer data.

What is the best way for an organization to get started on this journey?

Before you start applying any AI, you should first take the time to understand how well your current channels are working. Be very objective about it, and identify the problem areas for both your company and the consumer. Then, build the technology around that set of problems. The fact is, some issues can be solved with an automated agent, while others will always require human intervention. The challenge is to figure that out for your customers to make their experience seamless.

This article originally appeared in the Winter 2020 issue of Rotman Management magazine.

P.V. Kannan is the CEO of San Jose-based [24] and author of The Age of Intent: Using AI to Deliver a Superior Customer Experience (Amplify Publishing, 2019). He holds more than 30 patents and has been featured in books including Thomas L. Friedman’s The World is Flat and That Used to Be Us.