Somehow, Airline Customer Service Is Getting Even Worse

In early 2020, when the coronavirus was still a distant concern, my wife and I booked an AirAsia flight to Bali. Big mistake. At the start of lockdown, we scrambled to secure a refund. We called the airline’s customer-support line: no dice. We pleaded with its online chatbot, a lobotomized character named AVA. We sent a Twitter message to the brand on March 17 and received a response seven weeks later that read, in full, “Twitter Feedback.”

Those were dark days in airline customer service, with so many travelers desperate to figure out alternative plans. The present is not much brighter. In recent months, airlines around the world have changed how they engage with customers who need help. Frontier will no longer take your call, encouraging fliers to make contact via chatbot. Alaska Airlines is removing check-in kiosks at certain airports, driving people to its app. Air France, KLM, and Ryanair have all suspended customer service on Twitter, which for a time may have been the quickest way to summon a living, breathing employee.   

As Twitter melts down and people flee Facebook, social media just isn’t as useful as it once was for airline customer service. At the same time, airlines are leaning into AI, betting that the latest wave of chatbots will be the most cost-effective way to support customers. The long-standing truth is that companies don’t want to talk to you. First they didn’t want to do it in person, then they didn’t want to do it by phone, now they don’t want to do it online, and soon they won’t want to do it at all. It’s not personal—it just costs money. But hype-fueled AI products have yet to pick up the slack. “A chatbot being able to talk and to learn and to suggest and to persuade and do all of these things that humans do? I haven’t seen it in action, personally,” Eva Ascarza, a co-founder of the Customer Intelligence Lab at Harvard Business School, told me. Airline customer service is caught between two eras of the internet: one built on social media, the other on machine learning. The transition promises to be rocky. If you’re traveling this summer, you better hope that you don’t need help from an airline.

Airlines belong to a category of consumer-facing businesses that marketers call “high-touch”; they deal with customers whose needs are constantly evolving. Flights are delayed, bags get lost, people have to change their plans. And customers feel they deserve a certain level of care: After all, despite its gradual democratization, air travel remains quite expensive, especially in this period of high inflation. Multiply passenger expectations by the total number of seats—Delta flies something like the population of Sacramento every day, on average—and you start to appreciate the sector’s complexities.

These daunting customer-service demands have pushed airlines to automate since the dawn of mainframes. In 1960, IBM and American Airlines launched the first computerized reservation tool, based on a program developed for the Air Force. Customers would call a travel agent, who would then call an airline ticketing agent, who would then input the trip particulars. By 1964, the system could process some 7,000 bookings an hour, at a time when ticketing agents working manually could process one or two. The problem is, we’re still using it. “The basic systems which said ‘Box A talks to Box B via telex’ have largely remained unchanged since the 1950s,” Timothy O’Neil-Dunne, an airline-industry consultant, told me. (He paused to make sure I knew what a telex was. It’s a fax for text messages.) “So we are dealing with very, very old tech,” he added.

That old tech speaks in short codes: confirmation numbers, airport initials, seat numbers, passenger types. Customers rarely know all of the data that apply to their itinerary, which meant that until the advent of more advanced AI in recent years, changing a flight or locating a bag required a human intermediary, someone fluent in airline and English who could translate a question and input it as DL754, ATL, 19B, and Y. But call centers are expensive—even in Manila. Mindsay, a company that develops conversational AI for the industry, estimates that each support call costs airlines $2.20; in 2017, Harvard Business Review pegged the average cost of a live customer-service interaction at three times that amount.

Over the past decade, Facebook and Twitter emerged as efficient alternatives, allowing airlines to automate their response to certain posts and messages while paying special attention to the most urgent issues (or in some cases, the highest-profile users). In some ways, airlines demonstrated the viability of extending customer service over social media—if they could do it, any brand could. A study last year by the customer-experience company Emplifi found that among 23 industries, airlines had the second-fastest average customer-response time. In many cases, tweeting at an airline can really result in shorter wait times than sternly repeating “representative” on the phone or running a gantlet of scripted if-then scenarios with an online textbox.

Until recently, that kind of automated sorting was the best that chatbots—which many airlines offered early versions of—could muster. The predominant use case for AI in customer service was “the prioritization of calls, the prioritization of requests,” Ascarza said: software that decided how long you could wait for human assistance before the big vein in your head popped. On the other side of Twitter sat a flesh-and-blood airline agent whose voice was never heard but keenly felt. You could tweet something salty, tag the airline, and soon get an invitation to DM from an agent, who invariably signs their name.

But customer service through social media has become strained. “Can you calm down and allow me some time to work please ??” Delta tweeted to an inquiring customer last year. During the pandemic, airlines struggled to handle the unprecedented volume of passengers upset by endless rescheduling, and they doubled down on their automation efforts. In 2020, Delta temporarily suspended its customer service on Twitter and Facebook amid agent shortages and increased wait times. KLM, which was fielding 50,000 Facebook messages a day that March, enhanced its chatbot with machine learning; the discount carriers WestJet and AirAsia leaned into their existing ones.

Not all bots were created equal: AirAsia CEO Tony Fernandes recently called AVA, my erstwhile nemesis, “the most hated AI chatbot” in Southeast Asia. Nevertheless, it worked in the aggregate—at least from the brand’s point of view. “During COVID, AVA helped to clear millions of cases that would not have been possible given the amount of requests,” Fernandes said in an emailed statement. That was all before the implosion of Twitter under Elon Musk: Twitter has begun charging companies $1,000 a month to integrate their customer service into the app, prompting Air France and other companies to reconsider the site altogether. Meanwhile, the rise of platforms like TikTok, which aren’t as conducive to customer engagement, have undermined social-media support even further.

Those trends, along with recent strides in generative AI, have emboldened airline executives. Air India has committed $200 million to update its digital systems, which will include ChatGPT-driven features. Frontier claims that its self-service model requires less labor and delivers better customer-service experiences (although in a recent investor report, the airline baldly states its interest in limiting “avenue[s] for customer negotiation”). In February, AirAsia replaced AVA with a new bot called Ask Bo, which it promises will be “more proactive and attentive” thanks to “enhanced” AI. Technical details are scant, although a spokesperson for the airline claims that since Bo’s launch, 95 percent of all customer queries have flowed through it, and 73 percent of those queries were resolved with no follow-up. Considering how eagerly airlines have leaned into automation, expect other carriers to soon follow suit: “Airlines look on the face of it to be an ideal place to start to deploy an AI-based chatbot” powered by the latest large language models, O’Neil-Dunne said.

These days, even when you appeal to airlines over social media, you’re likely triggering some kind of machine-learning program. Both Twitter and Meta have invested in automation features for brands fielding customer messages. “I just get a link, and then I’m starting this WhatsApp conversation,” Ascarza said. “Sometimes I know it’s a bot, sometimes not.” She pointed out that although research suggests consumers prefer human interaction, it’s often because they lack compelling alternatives. That’s starting to change—and the ancillary features of AI support may suit us just fine. On a text-based interface, “it’s okay to be short,” Ascarza said. “And it’s okay not to be polite. And it’s okay to just get to the point.”

But the freedom to jettison social niceties is a shallow benefit. The simple reality is that customer service has eroded on social media, while the AI programs meant to replace it don’t yet meet the burdens of air travel. The last wave of chatbots, such as SWISS’s Nelly and WestJet’s Juliet, could clear the most straightforward cases with brute force, but they could also be blundering and ineffective. Airlines have iterated on those models and introduced new and improved versions, like Ask Bo, looking to capitalize on the fresh interest in AI as so many other companies are. Still, sophisticated bots on the level of ChatGPT don’t widely exist in air travel, and the way that airlines will actually deploy them—however many months or years from now—is an open question. For now, as the social web recedes from view and AI stumbles into an uncertain growth spurt, consumers everywhere are falling through the cracks.

In the long run, AI might improve customer experiences more than it degrades them. As airlines build smarter and smoother chatbots, they may free up their dwindling labor force to deal with the smaller percentage of more complicated requests. If chatbots are already capable of so much, why couldn’t they help us deal with a canceled flight or a lost bag? But AI could reshape customer service in more insidious ways. O’Neil-Dunne noted that as their customer-support tools become more nuanced, airline offerings are going the other way—giving rise to unbundled amenities and pared-down services, like some basic-economy tickets that don’t let you pick your seat without a surcharge. “If the product is simpler, the servicing is easier,” he said.

On the back end, AI could assess the value of every request, including if and when customers should receive help at all. “The decision of who not to serve is as important as who to serve,” Ascarza said. One logical outcome for an airline with millions of customers might be to simply deny or ignore a percentage of all complaints, which already happens with maddening frequency. The only thing worse than a feckless chatbot is a chatbot telling you, with perfect cogence and clarity, to get lost.