Mezi Talks AI, Mezi For Business And Global Expansion
Earlier in the year, The Blue Swan Daily introduced you to the artificial intelligence (AI) personal travel assistant, known as Mezi. At the time Mezi had announced its entry into the corporate travel market by offering a travel-as-a-service solution for Travel Management Companies (TMCs), Corporates and Travel Agents.
Simply put, Mezi is an AI-powered chatbot capable of automating more than 60 percent of the conversations initiated by travellers and then gathering availability, curating recommendations, and making reservations at the request of the traveller with limited human intervention. In fact, in most cases users aren’t aware of when they are chatting to the robot or agent.
In these few short years the technology has already been used by more than 100,000 travellers and impressively processed over 50 million words across 500,000 conversations. The introduction of this technology into the corporate travel world is making an impact with major TMCs already signed up to offering the product to its clients.
We recently caught up with Mezi Vice President, Travel Strategy & Partnerships Johnny Thorsen on the sidelines of the 2017 ACTE-CAPA Global Conference Sydney.
Johnny ThorsenMezi is an AI for travel platform. We've effectively created a solution that makes it possible to automate conversations with travellers. The service is completely wide-labeled, which means our customers, our B2B customers. It can be travel agencies, it can be corporate agencies, TMCs, it can be concierges, operators, or credit card, or other kind of programme owners. So they get our solution wrapped in their branding, and then they decide the rules for the communication and the content that is activated through their user base.
So we have a nice mix. Our largest and most well-known customer is called American Express credit cards. Most people know them. They deployed our solution back in February of this year, in a small scale, and are now accelerating. And they have basically called our solution, Ask AMEX, so it's literally a branded service under their name, but our technology 100% underneath.
We then have three corporate TMCs: Adelman Travel, W Travel, and Casto Travel, all three are U.S. based TMCs. And interestingly, all three have decided to create a new brand around our offering. So Casto Travel called the messy version Marco, which is short for Marco Polo, the global traveller. W Travel has decided to call it Seat 1A, which is where you want to sit when you travel. And Adelman Travel have not announced their name yet, but they've also created a new brand. This is fascinating because we're literally seeing our technology creating a new line of business in an existing operation.
Our plan, this year, was to stay in the U.S. and keep it tight. We're still in the early stages of getting our AI engine and our NLP, natural language processing, engine right. We are running at 60% automation for conversations in the U.S. market. But if we go to Australia, we need to understand the Australian travel dialect, so if people say arvo, we need to know that it's afternoon. We don't know that, today.
We're two and a half years old, as a company, but we are only one year old, in terms of focusing 100% on travel. And as we go forward, we think we will get to 80% automation rates. And automation rate, in this case, means that the software understands what you are asking for and can generate a relevant response back to you, with no human involvement.
However, that still leaves 20% that is not responding automatically. And instead of telling you to drop, end call, we now hand it over to a human agent. They will look at the conversation, they will generate the results, and send it back in the same channel. So for you, that means you stay in one channel, and you have 100% success rates. And for the agency, it means for the first time, they can focus on customer service delivery. They don't have to be a TDS expert. They're working our point of [inaudible 00:03:05] system, and they interact with emojis, with fun symbols, it's a human conversation, rather than a transaction conversation.
Early findings from our end users are that they literally love this channel. We have an NPS goal in the high 70s, and it's been consistent there, since we started tracking it. And we are seeing people constantly reacting with surprise when they find out that it doesn't matter what they ask, there will be a response. It's not a, "I cannot help you." It'll be, "Hey, please wait a few minutes, and somebody will come back to you." So we can see that the more they use us, the more confident they become in our ability to help them. And we literally have users who start asking us if we can do shopping for them, whether it's shoes, clothes, or tickets for events. They expand outside travel, and that's where it really gets interesting because now you're actually increasing your touch points with the end user.
In 2018, first of all, we are going to go international. Most likely, it will be in other English-speaking markets. I'm in Australia, scanning for potential launch customers, down here. And we want to go to two or three other markets, next year. We also want to increase the AI capabilities, so we're looking at things like predictive search activity. If we see patterns in your past behaviour, why shouldn't we be able to say, "Are you going on your usual Chicago trip, next week?" We're also looking at AI-driven policy, coming into the corporate arena. So instead of a travel manager trying to sit up front and figure out what that policy is, why not take a given sample of bookings, send it in, and ask the software to generate your policy, based on real behaviour? So we're really trying to change how things are done by simplifying and optimising.
Copyright policy: All transcripts on this site are the copyright of CAPA - Centre for Aviation. Our reproduction policy is as follows: you may quote up to 400 words of any transcript on the condition that you attribute the transcript to CAPA - Centre for Aviation and link to the original video page. All other use is prohibited. While we aim for 100% accuracy in the transcript, there may be some minor transcribing errors.