If there’s one guy knowing all about Strossle’s products, it’s our own Chief Product Officer Joakim Stenberg. During the last 12 months he’s been working hard to ensure Strossle’s platform was ready for the GDPR. The strategic decision was to move from personalized distribution to contextual distribution - or “Green Data” as we like to call it. The questions is, what is it? We asked Joakim:
CONTEXTUAL DISTRIBUTION OR TARGETING SEEMS TO BE ON THE RISE, BUT WHAT DOES IT REALLY MEAN?
Basically you take everything you can learn about an article and its surrounding and use that information to figure out which other articles that should be relevant, given the context the reader is in. This is really something we’ve learnt together with our publishers. After all, we are all in the business of telling stories and we must strive to make both editorial and commercial recommendations as complementary engaging as possible.
HOW IS CONTEXTUAL DISTRIBUTION DIFFERENT FROM PERSONALIZED?
Personalization is all about understanding the readers and their consumption patterns. Personalized article recommendations are based on historic behaviors such as: what type of articles do you typically read? Do you read the full articles or do you just skim through? How do these patterns differ depending on time of day or day of week? Etc. You will look at pretty much the same metrics for contextual recommendations as well, the main difference is that contextual is focusing on what to recommend when you’ve read a specific article, whereas personalization is all about what to recommend to a specific reader.
HOW CAN A MACHINE DECIDE BETTER THAN A HUMAN WHAT CONTENT SUITS WHERE AND IN WHAT SITUATION?
In our experience, humans are actually pretty good at predicting what articles other readers might appreciate. When we first started out with cross-site recommendations we successfully experimented with manual recommendations. But there are two problems with humans: first and foremost, we do not scale. It takes way too long to manually select recommendations for a given context, especially if you want to base your assumptions on data. Second, you know others as you know yourself, meaning that you will probably do a good job recommending articles to those who share your views, but not necessarily to everybody else.
Machines, on the other hand, scale and they have all the necessary data available at all times. Imagine that you are tasked with selecting 4 recommendations that will nicely complement a specific story and you have millions of possible articles to choose from. How would you decide and how long would it take you? Powered with a nice analytics tool I believe you could come up with a decent set of recommendations within 5 minutes.
Compare that to a machine able to decide within milliseconds taking all the data you had available in your analytics tool into consideration and all the learning collected from billions of recommendations produced in the past.
HOW DO CONTEXTUAL RECOMMENDATIONS BENEFIT THE USERS?
Through better privacy. Privacy means surprisingly little to a lot of people, but the kind of user profiles many platforms created before GDPR, is nothing to joke about. Strossle is present on a large number of sites throughout Europe and there’s a lot we can learn from studying consumption patterns on a daily basis. With GDPR we decided to drop all personal data and solely focus on context, a bold move you might say, but the fact is that our recommendations perform extremely well. Still there’s a ton of things we can learn and improve.
Another thing I enjoy with contextual recommendations is that they do not only confirm your existing opinions, they challenge you in a healthy way. Journalism fills a crucial role in our society, keeping us informed of what is going on, in the world and in our communities. Personalization must be treated with care when you deal with journalism or this important mission might be defeated.
DO YOU REALLY BELIEVE GDPR WILL STOP AD DISTRIBUTION BASED ON PERSONAL DATA?
No, personal data is considered too valuable and sometimes for good reasons. Imagine you are selling a luxury car, you know lots of people will find your ad interesting, but only a few will be able to afford the car. How do you determine who to display the ad to? Depending on what personal data you have available you should be able to make a qualified guess.
At some point I think Strossle will start utilizing personal data again, but then it’ll be in a very transparent way, collecting real consents from people rather than tricking them into it. I believe this can only happen when you are able to offer real value in return.
DO YOU GIVE CONSENT TODAY OR DO YOU DENY IT WHEN YOU VISIT WEBSITES? AND FOR WHAT REASON.
Professionally I experiment with consent solutions, privately I consequently deny them. I recognize that my personal data holds a value and if I’m going to share it I want something in return.
DO YOU USE AN ADBLOCKER YOURSELF?
Sometimes yes, sadly I find many sites unusable without one. The hunt for paid impressions is an interesting one, publishers are creative at finding new ad slots, supply goes up, prices go down, user experience is suffering.
WHAT WAS YOUR LATEST TECHNICAL WOW! EXPERIENCE?
I’m currently visiting Majorca, an Island in the Mediterranean. We are 10 people and at least twice as many devices on a single wifi connection. Leaving the house we enjoy free 4G connection across the Island. Think of how different it was only 10 years ago and the change in behaviour this technical leap has caused. And all the opportunities that followed and will follow. It’s really quite amazing and a healthy reminder that you must always continue to innovate.