Data & Expense Machine learning: what does the industry want next? By BMaaS Contributor Posted on September 25, 2017 11 min read View original post.In this guest post, we hear technology insights and tips from Mariano Albera VP Technology at Expedia Affiliate Network Machine learning is more popular in the travel industry now than ever. There’s a simple explanation for that fact: machine learning is more powerful now than ever before. The appeal of machine learning – essentially a form of artificial intelligence (AI) whereby computers learn without being explicitly programmed with new information – is clear. At exceptional speed, for example, complex algorithms can identify subtle but important data patterns that humans could never have spotted. In ‘learning’ from that information, the ‘machine’ can predict patterns ahead, and then act to process that knowledge to maximise future business. In a sense, then, machine learning is a modern and highly sophisticated technological application of a long-established notion – study the past to predict the future. Machine learning is a modern and highly sophisticated technological application of a long – established notion – study the past to predict the future The practical applications of machine learning, and other forms of AI such as data mining, are many and varied in the travel industry. The rise of the chatbot ‘Chatbots’ are particularly visible examples of machine learning at work. As the name suggests, chatbots are essentially machines – messenger apps – with which customers seem to have conversations. Armed with the knowledge of the customer’s past bookings, the chatbot can offer targeted recommendations highly likely to be converted into sales. Critically, the chatbot keeps learning from each booking the customer makes, so recommendations become more relevant with every new ‘chat’ and customer interaction. That’s a huge benefit in an industry as personalised as travel. Effectively, the machine learns how to close the deal without human help. In many ways chatbots are already better than humans. They: Provide low-cost 24/7 customer support. Deliver real-time message translation, so you’re not on the phone at two o’clock in the morning trying to find an English-speaking sales assistant in Tokyo Are much faster than waiting for a call centre to answer the phone. If you want information about train times, good theatre and the weather in New York, for example, a machine will source and deliver that information to you more rapidly than even the most well-informed human. Admittedly, chatbots cannot always answer complex questions but their sophistication is constantly improving. By definition, the machines keep learning. Icelandair, Lufthansa and Austrian Airlines are three carriers to have seen the potential of machine learning and introduced chatbots. Practical planning, time saving Machine learning also helps in areas such as planning optimal flight routes. Assessing the millions of flight options on, say, a long-haul, round-trip journey, complex algorithms can learn from past booking data to filter those possibilities down to the small number of most practical or appealing options…all in just seconds. Another application for machine learning is in addressing the problem of duplicate listings. Online travel agents, for example, gathering data from multiple sources, face issues of misspelling, punctuation and differing word orders that have historically caused problems for computers. Now, however, machines can analyse data and work out for themselves that ‘Delta Air Line’ is actually the same as ‘Delta Airlines’. No more staff time wasted de-duping and no more frustrated customers seeing two listings for exactly the same flight. How EAN is benefiting Like many travel companies, machine learning is increasingly critical to how we do business at Expedia Affiliate Network (EAN), where we use hundreds of hotel features to rank hotels for our travel partners by relevance to an individual consumer’s preferences. Like chatbots, we learn from every interaction. Let’s say, for example, that a traveller always selects hotels with high-quality gyms but never shows interest in swimming pools. By monitoring each of his selected and rejected options and bookings, our machines learn that fact without being explicitly programmed with those details. So, when the traveller next books, for example, a flight into Atlanta with a partner airline, he is instantly shown a range of suitable local hotels, prioritising gyms over pools, this maximising the likelihood of a conversion to sale. Lessons learnt At EAN, we are working on using a type of machine learning called ‘deep learning’ to rank and sort hotel images and this is what we’ve learnt is that very first thing people glance at within a hotel listing, before considering the hotel name or price, is the image. In fact, it takes us around one twentieth of a second to process an image, so the quality and relevance of the images and the order in which they are displayed to travellers is crucial. In the past, we relied on a manual process to select the featured image for a listing, while the other images were randomly ordered or grouped. EAN has over 300,000 properties, and over 10 million images so, as you can imagine, ranking and sorting the images for these manually is difficult. Enter AI to do this automatically. Looking forward Our aim is to automatically order and sort the images not just according to image quality, but also to traveller types, customer preferences and seasonality, so that the images most likely to encourage a booking are displayed to each individual consumer. The speed at which data can be processed, analysed and actioned, is already exceptional, and is improving daily The good news is that machine learning is advancing fast. The speed at which data can be processed, analysed and actioned, is already exceptional, and is improving daily. Across many industries, not just travel, I’d expect to see machine learning move from niche applications to mission-critical processes. As is so often the case, in issues of computing, the limitations are as much human as technological. Almost every part of the digital user experience can be improved with AI. We all need to think creatively about how machine learning can enhance our activities. What do we, as an industry, want to do next?