Can a chatbot alone manage travel disruptions?

With travel disruptions causing headaches worldwide, customer service representatives are fielding an influx of messages from travelers experiencing flight delays or hotel cancellations.

Over the years, chatbot technology has evolved to handle a number of these types of traveler queries, but has it advanced enough to replace human agents all together?

Industry experts say that, for now, chatbots and humans still complement each other.

“We use bot technology to be able to understand and classify the intent of what travelers want,” says Patrice Simon, CWT’s VP of new products.

“By putting a filter in front, we can easily direct and route the request to the right team,” Simon says. “And we can do that for everything - if you want to book a trip, change your trip; if it’s urgent, not urgent - we can direct that to the right people.”

Virtual agents are very good at sending out flight updates and answering certain routine questions, such as around COVID-19 requirements, industry leaders say. Plus chatbots can handle multiple conversations at the same time. They can prioritize requests because they have a traveler’s history and know which flight he or she should be on; they can also gather data from the customer through questions, then route the customer to the right human agent.

Virtual agents vastly reduce the number of messages that human representatives receive. Another advantage of bots: They don’t sleep.

But the chatbot still has room for improvement in responding to travel disruptions, and there are certain things a live agent just does better.

‘Every second matters’

American Express Global Business Travel launched a new virtual assistant in its mobile app in late June. The bot boosts efficiency and reduces wait times for customers, says Mark McSpadden, vice president global product and user experience at Amex GBT.

“During disruption every second matters for the traveler and for the travel counselor and the travel ecosystem,” McSpadden says.

During a “really stressful time” for a traveler, the bot saves the traveler’s answers to questions and forwards the information to the human agent.

“If you say, ‘I need to change my flight,’ or, ‘My flight was canceled,’ we know we’re going to need to connect you to a travel counselor. But the [virtual] assistant can gather context on that request before it ever reaches a human,” McSpadden says.

The AI can save the travel counselor time by asking questions such as, “Are you still trying to go to your destination? Are you still trying to get there on this day?”

“It helps us move through more requests quicker and keeps us out of scenarios where we have very, very long wait times,” McSpadden says.

AmexGBT reviews how bots answer questions and makes changes to its machine learning algorithm to improve its responses the next time. One area that the bot stands to improve is with “multi-intent questions;” it may not know to focus first on the hotel or the flight.

“We’re able to take real conversations that are happening with the bot … and make sure that the bot is learning and adjusting over time,” McSpadden says.

Similarly, CWT’s Simon says the travel management company uses an “intentional blend” of automated and human agents.

CWT bots communicate with passengers in two ways: one, by proactively sending out messages with flight updates and asking if the passenger needs to reschedule in the event of a delay; and two, by responding through the messaging channel. If a traveler writes in the messaging channel that they need to rebook a flight, the virtual agent hands off the customer to one of CWT’s human representatives.

Simon says CWT is still at the “beginning of our journey” with bots, and it will take “six to nine months to ramp up to where we want to be. … We want to strive to answer your questions better and faster.”

Teaching tech

Technology solution HiJiffy launched the guest communications operating system Aplysia in late June after six years of development, says CEO Tiago Araújo. Seventy to 80% of hotels are independent and don’t have the resources to teach a chatbot, so they need a plug-and-play solution, Araújo says.

Although Aplysia has more than 140 different topics with answers tailored to each hotel, the majority of queries fall under 25 topics.

Aplysia can understand the sentiment behind a customer’s message in order to prioritize the request of someone who’s having a negative experience, but the aim is not to automate 100% of communications.

“Every time you talk with our chatbot, you always have the option to be transferred to a human agent,” Araújo says.
Roughly 10 to 15% of customer queries may be transferred to hotel staff, but at independently owned hotels, staff are not always available to respond to incoming chat messages in the middle the of night.

Some hotels are opting for “continuous sound notification,” which is similar to a phone ringing at the hotel front desk and only stops when a staffer replies to the message, Araújo says.

HiJiffy is working on expanding its communication channels to include chat on a hotel’s website, Instagram, Facebook, WhatsApp and SMS.

“You choose the interface you want, and we want to automate as much as we can,” Araújo says, adding that right now it’s much easier to automate text than voice. By the end of the year, the HiJiffy lab plans to have built an all-in-one communication hub that includes voice automation.

Another bot that’s learning and evolving is TOMIS (Tour Operator Marketing Intelligence Software), a chat application for tour operators.

Certain triggers prompt TOMIS to route the customer to a person, for instance if someone is going to miss a fly fishing trip because of a flight delay. The bot can’t cancel the trip, but it can get the customer to a human quickly, says Evan Tipton, TOMIS founder and CEO. It gives the customer multiple options for communicating with a person: text, chat, call or email.

The next step is getting the bot to understand the intent of a customer’s inquiry through natural language processing: “If you ask three different questions in a row, which question should the bot prioritize?” Tipton says.

“You can train it to do the first one, but it might not answer the second and third.”

TOMIS’ quality assurance team, the tour operator and Google (whose platform they use) all can provide feedback to the AI to help it improve.

“We have this ability to basically give the bot a pat on the back. So you can hit this button that says ‘had a positive outcome,’” Tipton says. “And then we can also tell it, ‘didn't answer this one.’ And you click that button, and you can tell it how to answer it next time.”