Bethanie Garfin

At Autotrader, our consumer data is increasingly being used to personalise the on-site experience, and to provide insight to our retailers. Four years ago, we built an identity stitching solution, allowing us to join together user activity across mobile web, desktop and app, but nearly turned it off because of the complexity and low proportion of users who were being stitched. Then we launched Buying Signals, a product which tells retailers what the consumers enquiring on their adverts are actually in the market for. Suddenly, knowing that two sessions belonged to the same person in real-time became important and we needed to rethink our approach to identity stitching. We decided to run an experiment: how would our in-house Spark Streaming solution compare to the Snowplow Identities product? We will share this journey, from the complexity of implementing real-time graph traversal in Spark, to demonstrating the business value of stitching with consumer segmentation. Learn: - How we ran a build-versus-buy test on live production traffic and our findings and conclusions - Why real-time graph traversal is complex, and the challenges of doing it with Spark streaming - How we discovered that stitching matters most for consumers who are close to transaction