How AI is rewriting content creation for travel

Content is the foundation of a brand’s interaction with its customers. Whether in the form of a website, mobile app, emails or a social campaign, the words a brand uses provide both information and inspiration to drive conversions.

SDL is a global language and content management firm that provides services across a variety of industries, including travel where it works with two of the top three largest online travel agencies, six of the top 10 airlines and five of the largest hotel chains.

Traditionally, content creation has been a human-centric process – requiring the perspective and skills of people, not machines, to determine what to communicate and how to write it most effectively. But artificial intelligence and machine learning are changing this work, bringing not just automation but personalization and cost-savings.

PhocusWire spoke to Kevin Ashbridge, vice president of SDL’s global travel and leisure solutions, to find out how it's using these emerging technologies now and what lies ahead for the future.

Can you start by telling us how you are using AI and machine learning in relation to content?

We’ve been using machine learning to translate content for about 15 years. But now we’ve got incubator projects in the lab at SDL where it’s actually creating the content in the first place - that’s both inventory content like hotel descriptions as well as looking into a marketing assistant.

Can you explain more about how machine learning can create something like a hotel description?

If you think about the machine translation process, you’re going from let’s say English to French. And what happens is English comes in, it gets deconstructed into an internal pattern that gets reconstructed into French.

There’s two parts - there’s the decoding and the encoding into another language. We took that encoding bit and we fed it not English but a whole bunch of hotel facts: WiFi, minibar, things like that, and we trained the engine to write a hotel description based on those facts alone. It was actually feeding in the travel industry codes for these facts.

The engine took all these codes, used its understanding of what a hotel description actually looked like around the web and then wrote it based on the information we were giving it. At the beginning it was a tiny sentence about the facility. But if we gave it a bunch of things like geolocation, codes at the hotel level, codes at the room level, it constructed an entire hotel description - exactly the same thing you’d find on Expedia.com or Booking.com.

Then we could take what this machine generated and immediately translate it into 45 languages. All of this happened in the space of about one and half seconds from writing it to translating it.

What was even more cool, by feeding it slightly different sets of codes it would create a slightly different description. If you fed codes that were more relevant to a business traveler or codes more relevant to a family or romantic couple, then for the same hotel it would write a description in real time in 45 languages that was relevant to the end consumer.

So this is enabling the automation of customized marketing?

Yes. Because this can be done in real time, it really opens the doors to individual, personalized, relevant descriptions being delivered to an individual on their individual device at the right time given the context of why they are doing something. This is the game-changer.

In travel, content has always been treated as blob. It exists as a prewritten description in a database. It may be translated, but you get served the same description. It doesn’t matter who you are or why you’re going somewhere. You always get the same piece of text.

With the inclusion of machine learning in it and its ability to understand what to do, it’s ability to act in seconds to produce a piece of content in real time, in the language of the end user, is something that is changing the face of content.

Content in a few years from now won’t exist anymore. It will simply be created and delivered in that moment. It won’t exist as a prewritten piece any more.

And how would this content - on a website or in a mobile app for example - how would it become personalized to me?

It’s based on what you are doing on the interface and the choices you are making. And if you have any sort of purchase history or preferences, those elements can be included as well.

Relevant content is more important than personal content, because relevant content is believable.

Kevin Ashbridge - SDL

Share this quote

Then there’s a whole bunch of metadata – the airport you are leaving from can sometimes dictate the sort of content you are after. How many people are traveling with you? Is it business or pleasure? What’s the weather like when you’ll get there? That can all be blended to create something super-relevant to what you are doing.

In my opinion, relevant content is more important than personal content, because relevant content is believable.

What is the status of your content creation product today?

We’re in beta to gauge the market, working with a few partners privately to prove out the technology – final touch-ups before looking at a launch.

The initial use case is that full inventory - hundreds of thousands of hotels that need to be written and translated into multiple languages. At SDL we did a calculation where a 100-word hotel description, which is pretty common, in eight languages takes roughly two days to do - written, sent off for translation and that translation is turned around. And it costs about $22. For 100,000 hotels you spend $2.2 million for that content for that hotel.

These rapid content technologies here take not two days but three seconds to do, cost in the region of $5 instead of $22 – so you are looking at half a million for that inventory. It doesn’t need to extend to personalization. I’m simply talking about dealing with the huge amount of content that they need to deal with in these bulk inventory businesses such as online travel agencies, wholesalers and aggregators.

You mentioned machine learning functioning as a marketing assistant. Tell us more about that.

We really believe that all of these technologies should aid the human process of creativity. In the marketing side what we want to be able to do is aid the marketer with ideas of what to write. Often if you are given a large brief for a travel campaign, there’s a lot of longform material that needs to be digested. What’s the location like, what’s the hotel like, what would work for families, for couples.

From that longform material you as the marketer are often trying to get down to small elements: landing pages, sending mail shots, even tweets. What this marketing system does is it analyzes those longform bits of content, using machine learning technologies and suggests inspirational content for these short-form outputs.

It will summarize a long document, it will understand the topics available across that document – spa, food, beach for example – and suggest here’s a paragraph about the food, here’s a tweet about the spa or a deal we’ve got, taken from the longform stuff. It enables a marketer to work much more quickly. It can come up with 20 variations for that longform content in a few seconds and you can choose what works best. And all translated in the language of the end user.

And I think in just a couple of years as machine learning catches up in its sophistication in being able to create content, it will actually start to move into the inspirational area of travel content as well. So it’s creating not just something that’s inspirational, but something that’s inspirational, highly personal and highly relevant to the individual reading it. It’s what we call agile content. Something that was simply unimaginable a few years ago.

We’ve been talking about content primarily for marketing materials, websites, emails, etc. But could this machine learning-created content also be applicable to voice interfaces?

As a content concept, I think the biggest use is around the ambient computing environment. If you really think about it, when you are having a conversation with Alexa about where you want to go on a particular business trip, what you are simply getting in the moment is almost a facsimile of what you would get if you did the same interaction on Booking.com for example. Where the relevant content technologies come in is taking that conversation to something that’s much more personal to you in that voice environment.

In travel e-commerce traditionally there have been three major conversion factors. One is images and video, another is social proof and the third is what we might call the digital experience - what does the site look like, how well laid out is it, is it trustworthy.

In the voice commerce world, which is supposed to be growing immensely over the next couple of years, all of those conversion factors are gone. You don’t have videos and pictures, you don’t have social proof and you don’t have a digital experience.

All you are left with is words. That’s it. So the words being delivered through that device need to become more and more relevant in that conversation you have with Alexa to make you trust the conversation. Relevancy means trust.