aftenposten subscription purchase prediction · 2017-01-29 · campaign 1: 35% lower cost per order...
TRANSCRIPT
AftenpostenSubscription Purchase Prediction
Ciarán Cody-Kenny (Data Scientist, Schibsted Products and Technology)Eivind Hjertholm Fiskerud (Head of Data and Analytics, Schibsted Norway, Subscription Media)
• We built a prediction model to identify news readers likely to buy a subscription
• By measuring news readers' digital engagement on Aftenposten over 4 weeks, we can predict if they will buy a subscription in the next 2 weeks with 78% AUC performance
• Targeted Facebook campaign experiments based on the model showed a 22 % lift in conversions and a 35% reduction in cost per order
Summary
Growing digital subscriptions is a strategic priority for Schibsted’s publishers
Digital subscriptions grow rapidly Schibsted’s publishers have ~500 000 digital subscribers
We built a prediction model to identify the news readers most likely to buy a subscription
Observational period
Subscription convert Y/N?
28 days 14 days
What behavior here makes a user more likely to purchase here?
Behavioral data is a good predictor of purchase, more engaged users are more likely to buy
Logged-in users that …
• have visited recently • read more articles • visit more days • visit from >1 device • visit at weekends • come to Aftenposten from multiple sources
… are more likely to buy a subscription …
The prediction model performs well in ranking users on their likelihood to purchase
The 10% highest scored users bought 51% of subscriptions in the training set
The machine learning pipeline is site-agnostic and can scale to all of Schibsted’s publishers
Campaign 1: 35% lower cost per order for highly scored vs. lower scored users
Campaign:• Two segments targeted, one with users likely to buy and
one with users less likely to buy • Targeted users saw an advertisement to purchase an
Aftenposten subscription in their Facebook news feed
Result:• 35% lower cost-per-order for highly scored users than
lower scored users
Campaign 2: +22% increase in conversions for targeted users vs. the control group
Campaign:• A segment of users likekly to buy was targeted and
compared with a control group of users with the same purchase probability on average
• Targeted users saw a sponsored post with a paid article in their Facebook news feed
Results:• +22 % increase in conversion for target vs. control group • Campaign ROI 47 095 NOK
Campaign 2: +22% increase in conversions for targeted users vs. the control group
10 %
20 %
30 %
40 %
2016-11-23 2016-11-24 2016-11-25 2016-11-26 2016-11-27 2016-11-28 2016-11-29 2016-11-30
Cumulative difference in conversion, target vs control group
+ 22 % conversions
Day of campaign
Team
Stefano CostantiniData Scientist
Schibsted
Ciarán Cody-KennyData Scientist
Schibsted
Rebecca WiborgData Scientist
Schibsted
Albert FranziData Engineer
Schibsted
Eivind Hjertholm FiskerudHead of Data and Analytics
Schibsted Norway
Kristina StamnesSenior AnalystAftenposten
Stig BreyholtzHead of analytics, Consumer sales
Schibsted Norway
Marius ThorkildsenDigital Marketing Lead
Aftenposten