Sramana Mitra: What are the parameters against which you do the personalization?
Dave O’Flanagan: There’s a number of different parts. There are three key things that feed into our algorithm. One is that we build a very deep transactional view or behavioural view of the customers. The interesting thing in travel is that I can participate or shop in a different context. I may be a business traveler all year long but this time when I’m on the site and I’m trying to purchase a flight for my wife and three kids, I have very different cost parameters and desired outcomes in terms of my trips.
It’s not just about understanding who the customer is from a historical perspective, but it’s actually about understanding the context that the customer is participating in right now. That’s the customer piece. The other pieces are the product sets that are valid. Travel presents a unique complexity in terms of uncertain routes and certain products are available, uncertain plane and certain products are available. Sometimes, there’s only a finite amount of product available.
As those things change, we may generate a unique offer for our customer. We always get to determine what that offer is and if it’s valid for a customer at a specific route on a specific time. That all has to happen in real-time as we’re changing and customizing the web or mobile experience for the customer. Those are the three key elements. One is the customer. Two is the context in which they shop and three is the available product or content that are right for that customer at that specific time.
Sramana Mitra: When you’re modelling the customer, what are the parameters that you’re capturing?
Dave O’Flanagan: It varies. Boxever, three years ago, was very blackbox. We delivered all the algorithms. We delivered all the clustering and classification on the customers and we delivered that out-of-the-box to our customers. Over the last few years, what we’re seeing is our customers are becoming far more smarter. They’ve got teams of data scientists and real capability in the organization and they want to deploy that on our platform.
What we make available to our customers in building models and building machine learning algorithms is the customer data. We’ve got over 300 different attributes at a customer level about who the customer is, their age, their likelihood to buy, and a lot of different explicit and implicit preferences. That’s combined with all the behavioral data that we connect across different systems. That will be sessionized behaviour from the website, mobile, or from the call centre. There’s the footprint of the customer as they interact with different systems.
The third class of data that we have is the transaction history, which is the purchases that they’ve made, the amount that they’ve paid. All that is available to our audience builder and segmentation engine to classify and cluster customers. That’s the first step. We cluster or classify customers to find like-minded customers and understand how they behave.