The Scientific Method: How Skyscanner and AppsFlyer Experimented for Growth

Skyscanner’s flight searching service has a massive customer base, boasting 60m monthly active users who book 2m flights per month. Achieving that impressive level of success and market penetration hasn’t been easy, and has required the firm to stay on top of the latest methods of digital marketing and growth hacking.

As part of its efforts, the company has teamed up with mobile app tracking and attribution firm AppsFlyer to develop a ‘growth hacking’ method that placed data and scientific experimentation at the centre of its philosophy, using granular measurement to drive downloads and engagement.

“Our mission is to make measurement as easy as possible, with as little manual effort, and to allow our clients to buy with confidence, to make smart decisions, and to improve and grow their business,” said Ben Jeger, managing director for DACH at AppsFlyer.

Mobile has proved as natural fit for the travel sector, and that’s certainly true for Skyscanner, which has found that, over the past few years, mobile has become the foundation to its success. As a result, much of its marketing efforts are centred on moving customers from the mobile web to its native app, where it can offer a richer experience, greater customer engagement and take advantage of even greater data and feedback. 60 per cent of Skyscanner’s traffic now comes from mobile devices, and its app has been downloaded over 60m times.

“Mobile is better for travellers, and it’s better for Skyscanner,” said Ilana Munckton, director of growth at Skyscanner. “We’re still growing on desktop, which is great, but we’re seeing exponential growth on mobile, that’s where customers are really flocking to.”

Skyscanner has driven its huge expansion using a focus on scientific experimentation to power both organic and paid acquisitions. Every marketing effort is based on data that the company has been able to pull from customer interactions, and every change is iterated multiple times, with its impact carefully monitored and measured.

In order to facilitate this, the company has restructured its marketing department into distinct ‘growth teams’, each with responsibility for a specific region. These teams follow a strict scientific model when it comes to accelerating growth:

  • Define: make observations, do research, define objectives and prioritise experiments
  • Design: formulate a hypothesis, define metrics, perform analysis and validate a target audience
  • Develop: build a new product, feature or campaign, implement tracking and determine distribution channels
  • Test: run the experiment, using both A/A testing and A/B testing, as well as multivariable testing
  • Learn: monitor results in real-time, analyse the results and their significance, reject or approve the hypothesis

This model is then followed by one of three results. If there is a clear lesson to be learned from the experiment, the team ramps up efforts, redesigning the experiment and scaling it up to establish how new features or marketing strategies can be employed. If the results aren’t clear, the team perseveres, continuing testing and using current data to tweak its attempts. Finally, if the hypothesis is rejected, the teams pivot, re-defining the experiment and ‘failing forward’ to make sure that the data gained isn’t wasted.

“We run a lot of tests where we look for a causal relationship, and if that’s not really there, maybe we need to test a bit longer, or with a different cohort, in order to truly see if our hypothesis is correct,” said Munckton. “Creative iteration is an art, not a science. It will vary in every market, on every platform, and things that work now may not work in a year.

“Most of the tests you’ll run will fail, so the more tests you run, the higher chance you’ll find something that moves the needle. But you need to make sure you’re running quality tests. Make sure you’re running a robust scientific test, then run as many as you can.”

That focus on creating robust tests led to the firm working with AppsFlyer, whose attribution and analytics technology is integrated with over 2,800 partners to provide a huge wealth of first party data. That data is key throughout every step of experimentation, from helping to identify trends and patterns that can be tested, to building precise audience segments, to tracking and measuring interactions during and after tests.

Munckton described how Skyscanner buys some of its display advertising on a blind network basis, resulting in a mass of different sources and different qualities within that. Working with AppsFlyer to gather anonymised site IDs, the growth teams looked to optimise the blind network traffic that was being targeted.

A test was run, measuring 14 days of standard advertising against 14 days using optimised site IDs in six different markets. Measuring the uplift in ROI, the team found only a three per cent increase – hardly what they were looking for – so they pivoted the experiment, adopting a stricter approach that optimised for only the very top site IDs based on conversion rate.

While volume dropped considerably, the second test eliminated all but the best quality inventory, and as a result, ROI increased by up to 66 per cent in some markets. Now, reviewing site ID on a monthly cadence for all blind network purchases has been adopted as best practice across the company.

Overall, Munckton describes the scientific model as “a process that allows us to increase our velocity and move at pace as we test things”, helping Skyscanner make use of the wealth of data it has access to, and enabling it to zero in on the levers that truly result in improvement for marketing ROI.