Virgin Holidays - Data Science
Package holidays are not impulsive purchases and as such it can take up to three months for a media interaction to result in a booking. This presents an enormous challenge for performance marketing channels who rely on timely and current data to make real time decisions, and where most technology is built to serve simple 30 day cookie windows.
To solve this problem, we built a bespoke forecasting tool which can, with precision, predict the likelihood of any of our media interactions leading to a booking.
This allowed our teams to make more informed optimisation decisions, leading to increased investment in non-brand paid search by 40% year on year, ultimately delivering a 65% increase in bookings year on year.
In our first year of working with Virgin Holidays across paid search, paid social, SEO, performance display and data science, we delivered performance improvements across the board. This was achieved through best practice, account hygiene and basic optimisation. However, this is one-off, unrepeatable work so our second year was going to be more challenging.
We were tasked in our second year working with Virgin Holidays with ‘simply’ delivering further significant performance increases from what was now a stable, mature account.
Innovation and Creativity
We identified that the lack of timely data to make optimisation decisions due to Virgin Holiday operating a 90 day attributed purchase window attribution model, with corresponding cookie lags, was a major obstacle to us being able provide timely, accurate optimisation and drive significant performance increases on their account.
With no solution in the market available for us to use, Forward3D built a bespoke forecasting and attribution model to overcome the challenges of the purchase window.
Strategy / Execution
We split the project into three stages:
Analyse and visualise historic booking data broken down by booking type, destination, campaign type, audience profile (demographic, destination and stay length) etc.
Use this data to build a forecasting model that could predict the outcome of current media interactions.
Test the impact of optimising to the forecasted return rather than the existing methods (a small number of confirmed bookings).
Using a combination of our own data from the advertising engines, raw data pulled from Virgin Holidays’ attribution partner and their own 1st party booking and customer data, we first built a dashboard dubbed ‘Look to Book’. This visualises all bookings, broken down to a highly granular level such as demographic, destination and stay length.
All of these are tied back to the media interactions which played a part in that booking. Importantly, this allowed us to understand changes due to seasonal trends and peaks such as sale periods, and ensure our forecasting is accurate.
Our data science team worked with the paid search team to produce and refine forecasts predicting the return of campaigns. Over time, the forecasts were refined to a point where we were confident they were highly accurate.
Using machine learning technology ensures this refinement continues as long as the tool is running, without the need for human intervention. When we were comfortable that our forecast was accurate, we could begin to test optimising to these new numbers.
Starting out with a small number of campaigns, we A/B tested optimising to the forecasted numbers and reviewed performance. As the actual results came in, we grew in confidence about the effectiveness of the new approach and scaled the test across more campaigns. In order for this process to be as efficient as possible for the team, a new version of ‘Look to Book’ was built for purely optimisation purposes.
This enhanced version could visualise the forecasted return, the current actual return, and our targets, allowing us to make longer term decisions about strategy and budget phasing rather than being restricted to just bid and keyword changes. Finally, our teams used our own technology platform ‘Stage’ to build automated processes allowing their optimisation steps to be as streamlined as possible, meaning that they could move on to focusing on the next test.
Forward3D Q4 2017 results for Virgin Holidays show significant year-on-year improvements, clearly meeting and exceeding the challenge set by Virgin Holidays.
We were able to confidently increase spend by 40% across competitive non-brand campaigns, and this delivered a 65% increase in bookings. The increase in spend is a measure of the confidence Virgin Holidays have in us and the tools we have built, as this required additional budget sign-off before being spent. The return on investment reinforces the accuracy of our approach and the benefits it brings.
This optimisation approach is now being rolled out across all performance channels we manage, and is being considered for implementation on their sister business, Virgin Atlantic, which we have won in the meantime.
As well as being the basis for our forecasting model, ‘Look to Book’ has now become a widely used tool across Virgin Holidays for budget and campaign planning as well as real-time analysis of performance.
“The work carried out by Forward3D has greatly enhanced our ability to optimise our paid search with our current attribution set-up. This has always been a challenge in the past, and with this new model we can now react quicker to market and trading conditions as well supply the right insights to our trading functions, increasing business confidence in this channel.”
Jamie Marchant - Senior Digital Performance Marketing and Technology Executive, Virgin Holidays