“YOUR CALL IS important to us,” a recorded voice tells resigned customers as they wait endlessly to speak to a human agent. AI is starting to help companies improve the quality and consistency of their service in order to persuade customers that they do in fact care about them.

Ocado, a British online grocer, receives around 10,000 e-mails from customers every day and uses AI to detect the prevailing sentiment in them. It now replies to the most urgent ones first, and is planning to route complaints to agents with expertise in the relevant field. “Like other applications of AI, it’s about trying to make humans more efficient, not take them out of the process entirely,” says Paul Clarke, Ocado’s chief technology officer. Between 2017 and 2021 the share of customer-service interactions worldwide handled entirely by AI will rise fivefold, to 15%, and by 2019 at least 40% of such interactions will involve an element of AI, according to Gartner, a research firm.

AI will change customer service as much as the telephone did in its day. Before the phone started to spread in the early 20th century, companies handled customer inquiries by post or by visiting in person. Phones helped agents to become more productive, and AI will boost productivity even more dramatically, because it can handle large numbers of customer inquiries more quickly than humans can. This has become more important as communications channels have multiplied to take in e-mail, mobile messaging apps and social media. And consumers have got used to dealing with automated services. Surveys suggest that around 40% of American internet users would rather use digital customer services than speak to someone on the phone.

Virtual agents are on the rise. Some 30% of companies now offer standalone “bots” that can answer questions and solve problems, although their range remains narrower than that of a human. Many of these use some AI. They are trained on logs and transcripts of past customer interactions, and as they are fed more data they become better at solving more complex queries. Such bots enable businesses to deal with many more inquiries without hiring extra people. China Merchants Bank, a commercial bank, uses a bot on the popular Chinese app WeChat to handle 1.5m-2m queries every day, a workload equivalent to around 7,000 human staff. Caesars, the hotel and casino group, offers a virtual concierge, Ivy, at two of its hotels, which answers guests’ queries by text, many of them automatically if the inquiry is simple to answer. This has reduced calls to the human-concierge desk by 30%.

AI will also enhance customer-service agents’ knowledge, performance and speed. Some companies are experimenting with “voice-printing” technology which recognises clients’ voices and alerts agents if a caller is impersonating someone else. This will be especially helpful in financial services.

One Australian bank is experimenting with a standalone smart voice-controlled speaker to listen in on agents’ conversations about loans. If the agent forgets something or makes a mistake, it jumps in. Some companies are also using AI to suggest responses to customer queries which a human agent can approve or adapt before sending. Over the past year this has allowed KLM, the Dutch flag carrier, to double the number of text-based customer inquiries it handles to 120,000 a week while increasing the number of agents by only 6%, says Dmitry Aksenov of Digital Genius, a firm that helps automate customer support.

A few companies have started offering AI-enabled services that listen to calls to judge agents’ performance and send them suggestions for improvement in real time. One startup, Cogito, whose customers have included insurance firms such as Humana and MetLife, focuses on recognising “compassion fatigue” in agents. It takes in details such as how fast agents are talking and what words callers are using to detect emotion and gauge whether the interaction is going well. If there is a problem, it cues agents to act more empathetically. A tool like this can help large companies monitor their agents’ performance, but the agents may also welcome the feedback. Call-centres have a turnover of 30-40% a year, partly because agents have had little help with improving their performance, says Joshua Feast, Cogito’s boss.

Marty Lippert, head of technology for MetLife, reckons that in areas like customer service and human resources AI offers a return on investment of around 20%. Most companies buy AI services from outside providers, but firms with technical know-how often prefer to create their own. For example, a team at Uber, a ride-hailing firm, has built a system using AI to deal with e-mailed queries (there is no telephone option). It sends the agent ranked options for what to do next, which has cut the time it takes to resolve a complaint by around 10%.

One hope for AI is that it will free customer-service agents from routine tasks so they can sell customers other services and generate new revenue. KLM has been able to generate millions of dollars of extra sales since it started using AI because agents now have more time to help customers book upgrades and new flights, says Mr Aksenov of Digital Genius. But not all customers will appreciate more sales pitches.

Services that make customers’ lives easier will generate more customers, who will provide more training data to make the AI systems smarter

AI will certainly change the way selling is done. Many firms are experimenting with developing AI-enhanced recommendation tools, like those used by Amazon and Netflix, to help salespeople with their jobs. Google, Facebook and Amazon have been using AI to target consumers with ads and special offers online for years, with great success. Similar practices could spread to other businesses. For example, when sales staff at Goldman Sachs, an investment bank, take orders for corporate bonds, they can now see instant suggestions of bonds with similar risk profiles to pitch to their clients. Caesars uses AI to work out customers’ potential daily spending, choose the clients who will receive personal phone calls and in what order, and decide what specific promotions to offer them. The company’s boss, Mark Frissora, says that refining marketing to a “message of one” boosts customer loyalty over time.

Don’t call us

Gartner, a research firm, expects the number of phone-based customer-service agents worldwide to decline by 10% by 2019. That would increase the workload of those who are left. But companies need to be careful not to dilute their interactions with customers too much. The rise of virtual communication has left them with fewer opportunities to establish deep relationships, so customer service will become ever more important.

Clever firms will use AI not just to improve existing services but to engineer new ones. Metro Group, a German retailer, is testing the use of computer vision at the checkout: the items in a basket are recorded by cameras and the shopper is charged accordingly. Amazon uses similar technology in a convenience store in Seattle. Timo Salzsieder, chief information officer of Metro Group, reckons these new unmanned, vision-enhanced checkouts can handle 50 customers per hour, more than double the number for a manned checkout.

Some insurers, including Ping An of China, use AI to let customers file a claim after a car accident. Instead of having to phone the insurance company and fill in lots of forms, customers take photos of the damage to their car and submit them through an app for a quick quote for repairs. Building a tool like this is a technological challenge, but getting in early is a good idea. Services that make customers’ lives easier will generate more customers, who will provide more training data to make the AI systems smarter. Ping An gets 15m claims a year and handles 30% of them on its app. “It takes an enormous amount of cost out of the system and puts customers in control,” says Jonathan Larsen, Ping An’s chief innovation officer. Such offerings also reinforce firms’ direct relationship with their customers.

Conversely, voice-controlled smart speakers, as offered by Amazon, Google, Microsoft and Apple, could come between the companies and their targets. Some of these speakers host other firms’ apps. For example, UPS has built a tool enabling customers to track their packages through Amazon’s Alexa, which they might previously have done online or by phone. Companies worry they could be disintermediated, so that the firm that makes the speaker becomes the customer’s primary relationship, says Paul Daugherty of Accenture, a consulting firm, and co-author of a new book, “Human + Machine: Reimagining Work in the Age of Artificial Intelligence”. And, since voice-controlled speakers guide customers to a single answer rather than offering them multiple choices of firms to interact with, those that cannot or do not want to use these speakers may miss out on forming a relationship in the first place. Much will depend on how quickly voice speakers spread. Currently only about one in six American adults owns one, but that is already more than double the figure a year ago. And as speech recognition improves further, the appeal of speakers will grow, especially among youngsters.