Always start with your airline web analytics! This is what I say to everybody when we’re discussing how to increase conversion on your website. Web analytics is the foundation of every successful conversion optimization program.
If you’re at the beginning of your conversion optimization journey and want to get started with your airline web analytics, I suggest you check out these two articles first:
However, today I’m not going to talk about the basics. Today I want to show you how a particular airline is using data science to increase its conversion rate.
By now you’re probably tired of articles that tell you to use data science and artificial intelligence – because most of these articles don’t tell you how to do it.
Don’t worry! At Diggintravel we’re always hands on and super practical. This is why I talked to a data scientist who is actually working with airline data to increase conversion rate and ancillary revenue.
Transavia is certainly one of the airlines that recognizes the importance of Conversion Rate Optimization (CRO). They have a special role within the company, a CRO coach, to help build their culture of experimentation and continuous optimization.
In addition to recognizing CRO as a crucial process, they’ve changed the organization of their commercial department from a traditional, functionally managed department into a role-based, agile team. Their Data Fleet is a cross-functional team that supports all core units (they call them Bases) and also supports the CRO process.
You can see how data specialists fit into Transavia’s CRO team organization
Data and airline web analytics experts are certainly an integral part of Transavia’s online growth team.
Vincent works as a Data Scientist and is a part of this agile organization. He’s part of both the data fleet and the CRO team. So, when I talked to him, my first (obvious) question was: how does this setup work in practice?
The data people (analysts, BI specialists, data scientists, data engineers) are all scattered across the organization. So we are decentralized. I work in Base Direct, where we’re responsible for marketing and sales, and all the other data people are in different teams responsible for different parts of the organization. Together we’re part of an overall Data Fleet, where all people who work with data are put together. Once a week we get together and basically tell each other what we’re working on and get help from each other. It’s decentralized, but we get together and benefit from each other’s experience.
My next question was, how does a data scientist end up in an airline conversion optimization team in the first place?
Two years ago I graduated with a Master’s degree in economics, and after that I started with a Data Science Traineeship. I didn’t have a lot of experience in math or programming, so I learned a lot by myself and online, but also from my Traineeship.
Transavia works together with the trainee bureau that works with the University of Amsterdam. Professors from the university teach us twice a month for a full day to get [us] up to speed with our data science projects. The idea is to get the background from the Master’s level (academic study) and then put it into practice on the job.
Now, I promised we’d be practical here. The first practical application of Vincent’s data and analytics skills was a project they called automated marketing:
We actually started with a problem: that specific flights for specific destinations and time slots don’t sell out. Demand for some combination is just not high enough. This was a problem for our revenue management team. They asked us to get more friction than we normally do for some of those flights.
We decided to check out for whom these specific flights might be interesting. Normally we would send an email with these specific flights to the whole customer database, but we wanted to be a bit more relevant for our customers. We wanted to see if we could pick up just a couple of people in our database that would be specifically interested in the flights we have on sale, because they were cheap compared to other flights for the same destination.
So, Vincent and the data team wanted to use data to help the marketing team go from a one-size-fits-all offer to more targeted messaging?
Yes, this offer was fully based on your history, the frequent flyers data. We have to know that you either go to the same destination a lot during the year or that you go to a lot of different destinations, so we can build recommendations based on that.
If you’re going to Barcelona many times a year, then we know maybe you have a second house or you’re on pension and visiting your children. You don’t need to get the full list of flights we have on sale, but [we] just want to push Barcelona. We want to keep it really relevant to you and keep the number of emails to a minimum, but when we email you it’s really relevant. For the people who go to different destinations, we want to push the destinations based on data and the locations that they’ve already visited.
What kind of algorithms did they use for these recommendations?
It was based on collaborative filtering, meaning we compare one customer to other similar customers to see where they go as well. Then we can recommend the new destination based on that. If you’ve been to three destinations and someone else has been to the same three and then a fourth destination, then that fourth [place] might be interesting to you.
You’re probably curious about the results of this project, right?
The main finding was that it works pretty well and [that] we can scale down on the number of offers we send by email and be more relevant. Recommendations were quite the improvement compared to the normal newsletter we send with all destinations. The one sent to frequent flyers [for] the same destination was a lot better, with 45% open rates. That’s really high.
When I heard Vincent talking about flight recommendations, I started to think about other product recommendations as well. I was curious as to whether the same concept applies to ancillary products.
What we’re doing right now is actually an ancillary recommender based on what the person has already booked and what other people booked on that trip. In a pre-departure email, which is 10 days before the flight, we have three blocks with three different ancillaries. We show these are [ancillary] products for you based on where you are flying and based on what similar people on the same flight will do.
So, yeah we try to use [the same principles] everywhere, not only in emails, but also on the main website and the My Transavia webpage where you can check in and might buy some additional ancillaries.
The website part really caught my attention because Vincent’s team is not only using the data to improve the offer display, they are using it to improve the booking experience as well. You can probably agree that in a time when everyone is talking about personalization, the airline booking experience is still pretty static. You still see the same booking process; it’s the same experience for everybody.
The Transavia CRO team, on the other hand, is customizing the booking flow to increase their conversion rate.
One of the projects I’m working on currently is called the Express Checkout. Basically, based on your flight history, your visit to the website and which flights you’ve currently selected, we determine the chance that you’re going to buy ancillary products. If the chances are low that you’re going to buy ancillary products, then we let you skip a couple of pages in the booking funnel in order we make sure you’re going to complete the booking, because we see that additional pages in the funnel can decrease the conversion rate, and if you’re not going to buy the ancillaries anyway, we’re not going to show you those pages.
But how exactly are they determining who will go through the “Express Checkout”?
The model that we have can predict how likely you are to buy ancillaries. [For example], if it is a last-minute decision, or if you are on a specific flight where you are very likely to buy ancillary products – for example, on a ski trip to Innsbruck, almost everyone buys ancillaries – but for some very specific short legs where you stay for a very short period of time, it’s very unlikely that you’re going to buy ancillaries.
We use algorithms like the famous XGBoost or similar because it just works very well. It combines the linear relations that we see among the groups.
The last case I wanted to discuss with Vincent was website performance. You can find data from Google and many other experts about how your website performance affects your conversion rate. Essentially, with every additional second it takes to load your website, you’re losing customers.
However, we all know how complex airline websites are, especially the search results page where we display all flight and price options. It takes a lot of effort and resources to optimize the performance of the website, and you probably have problems getting the resources. Wouldn’t it be great if your airline web analytics could show the direct correlation between the performance and conversion rate? It would likely make your arguments for resources for your website optimization a lot better.
Well, Vincent and the team are doing exactly that.
Our website tends to be very slow, so it’s something we try to push to improve. To give it the extra push, it’s my job to find the exact relation between the site speed and conversion rate. What we analyze is, first [we] start looking at the loads – so from the beginning to the end of loading the page, the total duration of time, how does it compare to the conversion rate? There is a quite high correlation.
But now I’m more specifically looking at the latency, because this is just a small part of the total loading [time of] the page, but it’s the part we can influence more in comparison to the other parts of the site speed.
We see that the people with a faster loading website, with a quick latency, they’re way more likely to complete the booking. We see that for the people in the top 10% of the fastest level of latency, the conversion is 3 to 4 times higher. This is quite an increase. So, if we can increase site speed for everyone, we can maybe improve our conversion rate drastically.
If you’ve read to this point, you know that smart airline web analytics and using your data effectively is the key to increasing your conversion rate. What can you do next?
Idan
Nice to read but nothing was explained about the way that they used their algorithms, except just a name of one of them. Where is the hands on that mentioned?
Iztok Franko
Hi Idan! Express checkout was one practical example explained, another one was how they used algorithms to send more relevant emails and marketing campaigns. You probably understand the airlines can not talk in details about their data, numbers and projects. You can download the 2019 Airline Conversion Optimization whitepaper to find more examples.