The travel industry is falling behind compared to others. These platforms have not changed significantly in many years. Users struggle to book a flight or a hotel room and travel platforms are no help.

We can do better using basic but powerful machine learning techniques to make travel platforms more interesting and attract new customers.

Our world is huge and people want to discover it all, Malaysia, Italy, Japan. That means finding an affordable flight. Add to this the choices of hotels. Planning and booking activities. Arranging for a rental car. This adds up to a million possibilities
Without any knowledge of the user, travel platforms still default to a product-centric approach. Show them what you have. Show the cheapest options on the top.

Know Thy User

One thing these sites are not considering is how the users thinks and feels personally. The usual excuse is that “we have sooo many users, how would we know each.”

But with machine learning, the more data, the better. Personal choices could be arranged to make each user feel more engaged with your application.

Quite a number of successful companies are using this approach. YouTube recommends videos based on what each user’s viewing behavior. Amazon displays possible new items to buy based on past purchases.

The travel industry is getting behind and needs to adapt.

Users Have a Pattern

There is no way to predict accurately which choices the users will take. But you can detect each person’s personalized patterns.

Prices and inventory are the products you offer to your users.

Pools and spas. Eight-passenger cars. Skydiving. After a few bookings, you can better understand each traveler’s preferences. This is because they DO have a preference.

You can save their decisions in the system so we can determine how the customer behaves. We can compare them to others with similar interests. Each time the user comes back, we can provide them with choices based on their past behavior. You can show them something they haven’t seen before.
This is what the magic of machine learning does.

Here, I present two techniques to handle this problem based on how your system behaves.

Solution 1: Recommender Systems

These systems recommend products based on a user’s past choices and behavior as well as the choices and behavior of the categories he may fit into.

This algorithm uses features of your products and compare them among like users.
Conceptually, you can think of it as a table. The columns are the features, like amenities, car choices, airlines, and so on. The rows are the users.

A simple approach would be to input “1” if the user tends to book this feature often and “0” if not.

We compare rows using techniques like Pearson Correlation and dot product. These give us a numeric comparison between two rows, usually something between -1 (weak) and 1 (strong), indicating the relation.

We want to build a table for each user and establish a strong list of features, then start comparing with other users and find a list of the stronger ones. This can help you find features the original customer hasn’t selected yet but the others have. The calculate features show up as recommendations.
This is one way to make things easier for the user. This makes it more likely to create engagement and therefore to sell more.

If you wish to learn more about his, you start by reading this post, Recommender Systems Explained by Pavel Kordik.

Solution 2: Neural Networks

Neural Network detect patterns that are difficult to detect otherwise.

Presumably, they simulate how the human brain works. It receives a input from the outside, process it through the neurons, and it comes up with a response.

In travel, the inputs can be your offering features and the user’s past choices. From that, a neural network will output detected patterns of inputs to behavior.

Unsupervised learning doesn’t need inputs at all.

The goal for unsupervised learning is to model the underlying structure or distribution in the data in order to learn more about the data. There’s no correct answers and there is no teacher.

Jason Brownlee

For these neural networks we don’t need to specify what exact output we need, just categorize how we want them.

An example can be to categorize users by their choice of activity: calm, adventurer, and so on.
This can help us to identify users and group them together.

For example, we can use loyalty data to train these neural network to deal with users who don’t have an account in your site.

One downside of them is the amount of data needed. To train a normal neural network requires thousands of sets. For large companies, this is not an issue, but for smaller ones it would be difficult to aggregate data from different travel sites.

There are companies out there like Figure Eight that can make this process easier and cheaper.

The Good and the Bad

Machine Learning
  • Doesn’t need to be trained frequently
  • Can adapt as your users change
  • Easy to implement
  • Slow
  • Need vast amounts of data
  • Not always reliable (e.g., Netflix)
Neural Networks
  • Fast, once trained
  • Needs few inputs
  • Reliable
  • Training takes time
  • Takes time to implement correctly
  • Trial and error

Start with Your Users

I invite you to consider how to grow your travel business by focusing on your user.
Using Machine Learning techniques and Neural Networks you can make it easier for your users to spend more time and money on your products.

More engaged and happy users will recommend your site to a friend.

Artificial Intelligence can be your friend, in spite of what some people may say.