What TikTok can teach travel brands about content discovery

Despite the hundreds of millions spent on algorithm-related engineering, whether it's a movie from Netflix or a song from Spotify, recommendations can sometimes be hit or miss . This problem is even worse in travel, where the technology is generally behind other industries and the challenge is even harder - but TikTok may have a solution.

Why is TikTok the app to learn from? Because of TikTok owner ByteDance’s deep and successful history in recommendation engines. From its first news app, Toutiao , to the billion-plus-user social media app TikTok (and Douyin in China), ByteDance has successfully mastered recommendation engines, getting the right content to the right person.

While TikTok’s technology may be advanced, it serves as an example of how brands can master the art of personalization.

Here are three lessons travel brands can learn from the TikTok app.

1) Start small, but start now

Recommendation engines are complicated, and understanding how to leverage their ability to add value to your specific business will take time. The learning curve is steep, but fast. This gives an advantage to companies willing to dip their toes in sooner and launch personalized recommendations in some part of their product.

With Medium , you can see how it added a second tab to its articles feed, where it also included hand-holding messages to explain how the recommendations work and how to give feedback. And then YouTube, which followed a similar start-small approach with its New To You feed.

These are two very large media and content companies, which undoubtedly have the resources to hire the best in the business for recommendation engines. Yet they still elected to start deploying this new technology in small parts, allowing for learning time before rolling it out more broadly into their product.

Instagram is a brilliant example of where this start-small strategy can lead. Over the last 10 years it has built up its capability and experience, starting with changing its news feed algorithm, launching its Explore section and more recently launching Reels .

As travel brands launch their own small introductions of personalized recommendations, they'll quickly learn what key data points they're missing to scale recommendations, where the gaps are in existing content and which are the most important traits for the recommendation engine in a user and in the content.

To launch quicker — or if the technical expertise is lacking — they can leverage recommendation engine SaaS providers, like Algolia or Recombee . Of course, over time, travel brands will need people within the company that understand this technology and how to take it to the next level.

2) Enrich content metadata

As humans, we can understand and very quickly categorize things we see — restaurants, video clips and songs — but for an algorithm, this needs to be taught. A recommendation engine can only ever be as good as its understanding of the content it’s recommending.

Video, a good comparable to travel content, is one of the most challenging areas for a machine to understand what is in each clip. This is why TikTok has invested so heavily in building some of the leading machine-learning algorithms to scan each frame, in each video, to categorize it within different subcultures. This enrichment process helps TikTok understand what’s in each video, what type of video it is, what’s the mood and many more things.

In travel, hotels, bars, museums and restaurants, for example, are very nuanced as well. They are difficult to categorize in a detailed way. Consider the question of why someone likes a particular video or restaurant. There are many, many attributes to consider as part of the answer.

Travel brands should start to think early about the data they have for each piece of content, as well as how that data is organized and, importantly, how they can enrich it further. A restaurant, for example, might only have a description and a rating. But it could also have the price level, number of reviews, Instagram follower count, cuisines, etc. Each of these attributes, when combined, is a large factor in why someone prefers one place to another.

The more enriched a piece of content, the more robust the conclusions are from users interacting with it. For example, say users are not just looking at restaurants in Morocco, but rather very popular, medium-rated, cheap eats in the Old Town. This rich metadata will be the difference between great recommendations and average to poor ones.

Travel businesses may have to work with other companies to enrich their data, such as Google’s Maps API , or a new challenger in the space, like Here .

3) Capture and store the right data

This may sound obvious, but there are always holes in the data being captured.

For instance, ensuring that there's a clear and traceable link between every user and every piece of content they’ve ever interacted with is key. And not just if a user has saved or booked a piece of content, but also other intent signals like clicks, views and time spent.

Combined with the content’s rich metadata from the previous point, there starts to unfold a really clear picture of what users are interested in. These are all essential inputs for a recommendation engine.

Netflix now claims “more than 80% of the TV shows people watch are discovered through the platform’s recommendation system.” This is a staggering number and an example of the potential of a good content recommendation system. Netflix has always positioned itself as a data-driven company, and it’s been documented how much usage data it collects on each viewer.

This deep understanding of users and every micro-action and interaction that takes place on the platform is one of the key ingredients in the success of Netflix's recommendation system.

Finally, all this usage data needs to be stored in a raw, accessible format - ideally not buried in a particular analytics or recommendation tool. Data management tools, like market leader Segment , provide a simple way to funnel data to analytics tools as well as to long-term storage containers like Google’s Cloud Platform or AWS. This will give data access to the recommendation engine without any restrictions. And importantly, it ensures a company is not tied to a specific provider long term. Rather, it gives them full control and ownership of their users' usage data.

Travel's big problem

In travel, users face an enormous discovery problem: where to go, where to eat, what to visit, where to stay - it's endless. And as much as travelers enjoy the serendipitous discovery of new places, it can be burdensome.

From our research at Pluto, we've found that - even amongst travelers who go away more than six times a year - 76% say that searching for relevant inspiration is painful. Travelers today have unique interests and want to reflect that in their holidays.

But the overwhelming choice is making it more and more difficult to find exactly what they’re looking for: For instance, Google Maps has more than 200 million points of interest and Tripadvisor has nearly eight million businesses.

A problem worth solving

Solving the discoverability problem gives travelers a strong reason to keep coming back to a product or platform. But it remains a massive challenge for travel brands in part due to the infrequent demand cycles of travel.

This is what we're solving at Pluto , with an app to help travelers find inspiration, plan their next trip and connect with like-minded travelers. We leverage personalized recommendations to help people find authentic and unique places to visit on their trips — drawing on the same patterns as some of the most successful content platforms.

Personalized recommendations are only getting better, and platforms like TikTok have shown it's possible to make significant advances in this space. It's time for travel to get onboard.