The Future of Customer Service Is AI-Human Collaboration

Customer service is traditionally considered a cost center, so many organizations have focused their customer improvement efforts on reducing costs. This proves to be a critical mistake, as everyone is left unhappy. Even as customers are sick of pressing two for reservations and three for service, service reps are sick of answering the same questions over and over.

The latest technology for service is virtual agents: Automated systems, trained on service transcripts, that can use AI to recognize and respond to customer requests whether by phone or chat. Given the history of customer service, you might think the focus here is once again on cutting jobs and saving money. But it’s not.

Our combined experience in customer strategy and running a company that helps build virtual agent systems has demonstrated two fundamental — and counterintuitive — facts about customer service and automation.

First, the most significant gains from virtual customer service agents are from improvements in customer experience, not cost savings.

And second, successful virtual agent systems depend on bots working with humans, not replacing them.

Can automation actually improve service? When customers think of automation in customer service, they remember what happened with interactive voice response — a telephony system that made them navigate phone menus for gathering information but often did not provide shorter wait times or better service. Online chat with agents can be similarly frustrating, because the typical agent is juggling two to six sessions at a time, causing slow responses.

But virtual agents signal a shift for customer experience. If a virtual agent can interpret the intent behind your chat or phone request, it can get you an answer more quickly and efficiently than a human agent. For most common queries, this delivers a better experience than interacting with an actual human.

For example, at SiriusXM Satellite Radio, a virtual agent with infinite patience will walk you through the process of initializing your new car radio. It will look up your account based on your phone number, wait while you make sure the car has a clear view to the sky, and connect to the system that sends codes down to your radio to allow it to start working. The virtual agent doesn’t care if that takes three minutes or 25 minutes because it is designed to act efficiently on your intent, not to minimize the agent’s time on the phone.

But not every customer service call can be easily answered by a bot alone. They succeed best because they’re part of a system that includes human agents as well, capitalizing on the strengths of each.

How customer service agents and bots work together. Humans and bots have different skills. Human customer service agents easily recognize when someone is frustrated and can respond with empathy. AI-powered virtual agents, on the other hand, are wizards at assembling data from disparate systems to render a judgment instantly, even if they lack the emotional intelligence to know why such a decision might be right.

That’s why companies are now designing service flows in which virtual agents and people work together. Such workflows help free human representatives from the drudgery of rooting through computer systems to find information. The human agents can instead concentrate on the exceptions — customer problems the system has not encountered before, or frustrated customers that demand empathy. The virtual agents solve the common problems efficiently, while the human agents deal with the high-stakes issues that don’t fit the usual pattern.

In customer service, there are three basic modes in which virtual and human agents work together: bots initiating the conversation and handing off to a person; human agents serving customers with help from bots; or bots serving customers with human agents supervising. We’ll show examples of each below.

Bots can save human agents time. Based on our experience with deploying virtual agent systems, those systems can typically handle 80% of incoming questions without assistance. But what happens when the question is too complex or the customer too upset to deal with a virtual agent?

Even then, virtual agents can save time. Some virtual agent systems are actually designed to collect information, such as the customer’s name or account number and a description of the problem, and suggest resolutions as they hand off the call to a human representative. For example, at the rental car agency Avis/Budget, virtual agents collect customers’ names and the places and times for pickup and drop-off, cutting 30 seconds off the call time for the human agent and eliminating repeated questions for the customer.

Handoffs between bots and people can improve efficiency. For example, at the marketing service company HubSpot, a chatbot qualifies leads, delivers content, and connects potential customers with HubSpot’s (human) sales staff — all through Facebook Messenger. Having been prequalified, such leads are 40% more likely to be willing to engage with salespeople.

Bots can make human agents smarter. Some companies — like high-end professional services firms — would rather have human agents handle all service calls. But those agents can appear smarter if a bot is whispering in their ears. For example, when there’s a new discount or an out-of-stock product, the bot will bring it to the attention of the agent. It’s like having an assistant scurrying around to figure out the best things for you to say to the customer.

ABIE, a customer service bot at Allstate Business Insurance, was designed to help salespeople. Allstate’s independent insurance agents use it to look up specific pieces of information, like what it would cost to insure a landscaping business and or what rules apply to medical labs. In the past, finding that information would require searching through hundreds of documents, but ABIE quickly brings the correct knowledge to light to make the sales staff more efficient. This allows ABIE to handle more than 25,000 inquiries every month.

Bots can improve with human supervision. In cases where virtual agents can’t figure things out, a human supervisor can be a great help. When it comes to customer service through online chat, a human agent can efficiently manage eight or 10 conversations between chatbots and customers — a far higher level of productivity than if the agent were answering the questions herself. And when a virtual agent gets stuck, the more experienced human can step in and solve the problem. Not only that, but the human agent can tag the problem when the virtual agent bogs down, which allows the AI system to learn from that scenario and become smarter for the next question.

In this mode of interaction, both the bots and humans are able to focus on what they do best. For the virtual agents, that is handling routine cases quickly and efficiently. For the human supervisors, that is solving more complex problems, using empathy and emotional intelligence, and training the virtual agents on cases they didn’t recognize at first.

Bots can free up people to work on more interesting challenges. At InterContinental Hotels Group (IHG), virtual agents are working to solve technical problems for employees who call the IT help desk. With a workforce of more than 30,000 employees, IHG recognizes significant gains in employee satisfaction.

IHG used machine learning to review chat transcripts from customers contacting the help desk and constructed a virtual agent that could answer common questions. The result was a system that worked with the 150 help desk staff instead of replacing them.

IHG’s help desk can now scale to meet spikes in demand that might result from putting a new technology system in place. As Scot Whigham, who designed the system for IHG, explained, “We can scale up to meet what comes into our systems, but then scale down so that we don’t have excess capacity that we are paying for.” Instead of paying extra staff in times of low demand, the organization can redeploy resources to perform other important things on the IT to-do list.

For the tasks that it was trained for, like password resets, the virtual agent system can now handle 80% to 85% of the volume of questions coming in. Even when the virtual agents can’t answer a question, they can help. The virtual agent system is able to collect basic information from the worker, regardless of whether it’s answering the question itself or handing it off to a help desk staffer. Whigham reports that among hotel workers who have engaged with the system, satisfaction rates remain high.

The future of customer service is human-machine collaboration. As all of these examples demonstrate, the AI-driven transformation of customer service is not about getting rid of workers — it’s about making them smarter. When machines handle routine inquiries, customers are happier. And when service staff can concentrate on more complex questions — or on answering questions with a bot making suggestions — they can deliver far better service.

People doing knowledge-based work have always been more productive and thoughtful with the help of computers and information. Customer service is no exception. Now customer service workers can leave the drudge work to the machines and concentrate on being the sympathetic ear and clever problem solver for the customers who really need them.