There have now been four Data Science Hackathons held at the I-COM Global Summit, where groups from around the world meet to engage in collaborative data science and innovate to solve challenges. This year, the Summit was held in the coastal city of San Sebastian from the 9th to 12th April, exploring the theme of ‘The Reality Disruption Field’.

This is the second year that Ebiquity has taken part in the I-COM data science hackathon. This year, we took a team with a balance of consultancy and data science skills from Ebiquity’s Effectiveness and Advanced Analytics practice areas and made up of myself, Esther Etim, Miles Wood and Yan Zhang.

The Lufthansa challenge

This year’s hackathon was divided into two challenges, one posed by Lufthansa and one by Intel, with data provided by multiple partners including Kantar, YouGov and Google, while Domo provided a platform that contained all the data and allowed us to visualise and interrogate it. We decided to go for the Lufthansa challenge. There were two parts: a quantitative part where we had to build a model and come up with a prediction, and a qualitative part that involved solving a business question.

The qualitative challenge asked what new KPIs ‘were a strong indicator of competitive digital strength and long-term sustainability as measured by market share and sales for Lufthansa?’

How we tackled the challenge

We began by dividing the data sets between the team for further investigation and analysis, which allowed us to get the most coverage across the datasets without overlapping the work. We could submit predictions throughout the day to assess how well our model was performing, so our first challenge was to get a model up and running so that we had a baseline to work from.

For the quantitative challenge, we were presented with three years’ worth of airlines sales data originating in both the German and US markets. Our task was to predict weekly Lufthansa sales for each of the provided booking classes, from a specified 20 week time period using prior sales data across multiple airlines, as well as any of the provided auxiliary data we consider fit for the task.

The flight data we were presented with was very granular, so we needed to perform some aggregation before modelling. We decided to group the data into flight “corridors” – for instance Germany to US – and cabin type – economy, premium economy, business class and first class – as we felt this would give us the most robust learnings, while also keeping the number of models we had to build to a sensible number. After aggregation and cleansing, we had close to 200 models in total.

What we found

The data showed that the airline industry is a competitive market place undergoing rapid change. Low-fare airlines like Norwegian Air are undercutting established airlines, such as Lufthansa, with cheaper no-frills fares to capture price-sensitive travellers who are looking for a mode of transport, rather than a travel experience.

While established Airlines are seeing a decline in Economy class bookings, we identified Premium Economy as a market opportunity. Airlines looking for opportunities to grow yields could explore this sector to capture less price-sensitive customers who are looking for a more premium experience.

Business and First Class are much less competitive as customers are less price sensitive, however understanding just how price sensitive they are could be key. Planes frequently fly with partially full premium cabins, so lowering the prices on these could yield significantly more upgrades or bookings, in turn driving up overall yields and operating margins.

As a final point to consider, the industry, at least in the lower booking classes, operates with a strong seasonal pattern, so the best time to capture customers is when they’re in the market to make a purchase. Some media lines, such as TV, see media cost seasonality, and travel is well placed to exploit this. Some of the cheaper months to advertise are January and July/August – exactly when people are in the market for a flight, cruise or package holiday. Aligning advertising with seasonal demand and media cost seasonality, where it exists, is key to driving marketing return on investment.

What we learned

While in our client work we always strive to be clear on the question the client is asking and their expectations of the outputs. We felt that we could have done an even better job of interrogating the brief and understanding the questions in more depth, as this year’s challenge was a particularly ambiguous question and brief.

Further to this, while working in the analytics space, building models and working with numbers is often not enough. The most successful projects are those that manage to take a complex question and develop a solution that can be easily delivered to a non-technical audience. This communication is key to a successful outcome. Next year we’ll be focusing on how we can deliver our work to a non-technical audience in a much simpler way.

We look forward to returning to I-COM 2019 where we’ll be even more prepared to take on the next tough challenge!