Following its initial launch as a virtual shopping assistant, Mezi has found success as a chatbot-led virtual travel assistant since its 2016 pivot. Since then, it says 100,000 users have downloaded the app to book flights and make hotel and restaurant reservations.
But the startup doesn’t actually market its B2C app. Instead, the app serves as a test bed for the company’s AI capabilities, the end results of which are customizable, white label products. Mezi’s AI technology isn’t outsourced to IBM Watson or similar providers, but rather the app’s deep learning and natural language processing systems are created entirely in-house in order to provide users with three filtered options to each request they make.
To date, Mezi counts travel management company partners including American Express, Adelman Travel and Casto Travel among those third party travel businesses that subcontract their proprietary content needs to Mezi.
Johnny Thorsen, Mezi’s vice president for global travel strategy and partnerships says:
“Our clients don’t sell tech, they sell travel and we can build and deliver to suit their app needs.”
The white label version of the app has been gaining traction with TMCs not only for its customization capabilities, but also for its chatbot reliance; the technology can automate over 60 percent of traveler-initiated conversations while human agents have the opportunity to follow the conversations and interject when required. In other words, TMCs are no longer obliged to onboard agents with GDS experience, an increasingly rare commodity in the travel industry.
Thorsen explains that Mezi’s agency customers tend to feel more productive by a factor of three or four because they’re no longer as dependent on human staff to do legwork. He elaborated that “the agent dashboard will indicate trigger words such as ‘unhappy’ so that the bot will pass the conversation on to a human.”
Travelers turn less and less to search engines once they come to trust a Mezi app it seems. Unlike search engines that at best require a refresh on a search when left dormant for several minutes and more often, force users to reenter all relevant search data after walking away for any amount of time, the app remembers where users last left off so they can pick up in precisely the same place.
Consumers’ past history is always readily available and the search won’t time out, which Thorsen describes as “a massive value in and of itself.” The app can also send users reminders as it will remember when the user last connected and communicated, meaning communications campaigns can be built around users’ past search activity. Users know they’re interacting with a bot, but Thorsen says there’s still an emotional connection with “thank you” ranking as the 24th most used word and “sorry” as the 44th; there are 50,000 words in Mezi’s AI database.
But customizing users’ travel begins just after they download the app, when they create their profile, including their home airport, the frequency of their business and leisure travel, their preferred airline, seat and class of travel. Hotel preferences are captured by brand and style of hotel –chain, classic, contemporary and boutique as well as locations such as city center. Each data element entered will be applied toward the three best options that are returned to users.
App users can also still leverage certain elements of Mezi’s original retail offerings, including ordering flowers on the app and making dining reservations via Opentable. In fact, dining reservations can also be made via Siri. But Thorsen notes that the specifics of the reservation must be specified to a tee. He says:
“Voice is terrible for providing options, but perfect when you want your usual and there are no other options in play.”
He adds that both hotels and airlines are expressing interest in the app to create mobile platforms for their loyalty programs or, in the case of hotels, to showcase their stable of brands.
In the meantime, Mezi will continue to innovate its product, with 2018 objectives to create a new travel protocol for corporate travelers, taking into account how they want to travel –economy or business class—and how much they want to spend.
“We want to create an AI-built policy based on activity and a realistic market situation. Travel policies are normally black and white, but with an intelligent and flexible travel policy, price targets can be enforced with as much or as little leniency as companies want and they can also take predictive services into account for corporate travelers who frequent the same routes regularly.”