Schibsted Media Group is a leader in using data to guide its interactions with consumers.
Edoardo Jacucci, vice president of strategy and data analytics for Schibsted, leads a unit of data scientists and conversion rate optimisation experts who work with the company’s central leadership as well as with its many subsidiaries.
Each team member working on conversion is embedded within a subsidiary and is focused on solving its problems, but team members all share their knowledge across the entire system.
When developing solutions one of its companies, Schibsted’s data scientists build new features centrally so that they can be applied to any of the corporation’s other properties. And while the building of user profiles for advertisers was once done by each property at the local level, the company is now working to build profiles across all properties in a given market, enabling it to provide rich profiles to advertisers and deliver more relevant ads to users.
Schibsted has traditionally collected a great deal of user data from Web site traffic and from interactions with its classified ads, and it has run analytics on site performance and on purchases and subscriptions, Jacucci says. Its subsidiary companies have progressed at different rates, though, with some developing far more sophisticated programmes than others, he says.
In 2014, Schibsted started collecting data more systematically, with a short-term goal of ensuring that all its properties are examining user behaviour in the purchase funnel and within other specific funnels of interaction.
“We set up metrics and tools to follow exactly what happens, what users do in these funnels, to see how we can make their life easier in the funnel,” he says.
Schibsted is also working to track its users’ reading habits by log-in ID or by cookies to understand their reading preferences and use those preferences to develop relevant article recommendations, Jacucci says.
Its previous efforts at user recommendations have been minimal, he says, though a few of its classified sites have personalisation services. Some sites have experimented with an external provider for targeting content.
But in 2014, data scientists on his team worked to develop an internal system of recommendation features, as well as responding to the growing demand from advertisers for a more data-driven, performance-based product in its display advertising, targeting more relevant ads to users.
Jacucci’s team has made a strong push to help improve the company’s main funnels and ensure there are proper metrics and tools to follow user activity, as well as proper methodology for conducting A/B tests and making data-driven design decisions.
The success of A/B testing for conversion can be scientifically measured, with a better-performing version of the landing page sometimes delivering a 30% to 50% improvement in sales.
Schibsted’s overall success rate in conversion experiments has been about 8 in 10, he says, and its efforts have found a great deal of low-hanging fruit, garnering a nearly immediate return on investment, sometimes as much as a 10% to 30% increase in sales in a single month. One site even saw a 52% improvement in conversions.
“We just go around and try things, and we have some interesting surprises,” Jacucci says.
Determining the return on investment for Schibsted’s engagement efforts is more complex because personalisation doesn’t deliver an immediate economic benefit but improves the user experience and makes the sites more attractive, Jacucci says.
It has been more of a medium-term investment, but its effect is to increase the inventory that can be sold to advertisers, which then increases revenues.
For Schibsted’s classified advertising sites, increased engagement means users are posting more ads and may buy more up-sell products such as bumps that make their ads more visible. In Schibsted’s publishing companies, more engagement can translate into better sales of subscription services and other products.
Jacucci says it can also be difficult to calculate the return on its investment in display advertising, but it is a necessity to follow the trend in the industry. Traditionally, publishers’ products have had a high cost per impression and have not been data-driven. But as advertisers see the huge growth of Google and Facebook, they may demand an increasingly data-driven product.
Schibsted’s aspiration is business-impacting analytics aimed at:
- Predictive metrics.
- Recommendations engines.
- Community building.
- Conversion optimisation.
- Reducing churn.
Frode Eilersten, executive vice president for strategy and digital transformation for Schibsted, says the investment in Big Data is part of the company’s forceful digital transformation. This includes building an internal business culture where data is central to solving complex business challenges.
Crucial to these efforts is Schibsted’s payment ID system, SPID, which is the group’s data supply foundation. This payment solution gets people to log in and input their credit card data, and allows for frictionless payment.
Schibsted captures as much insight about the consumer as it can. This single sign-on experience and environment for all its different services is creating rich consumer profiles and leveraging data. Combined, this helps Schibsted better understand its readers, and learn their preferences in media, classifieds, and e-commerce.
As of late 2013, the scale of Schibsted’s SPID system was enormous internationally: 2.3 million verified users, 4.3 million transactions, and 240 million log-ins.
Schibsted takes this raw data and converts it into action. The company puts a lot of effort into predictive analysis by taking user data and creating consumer segments, Eilersten says. This allows Schibsted to convert consumers from infrequent users to loyal users and prevents them from leaving the Schibsted brands.
Big Data is now being used by Schibsted to innovate with products such as the mobile app for Omni, a news service based on automated content entry.
Not all of Schibsted’s experiments have been successful. One predictive churn model for a subscription product incorporated a hypothesis that churn was related to reading preferences. So the product provided certain kinds of articles with the expectation that readers interested in horoscopes were more loyal and would have a lower churn rate than those interested in gossip.
Says Jacucci: “We, of course, added these variables to the predictive churn model, and when we had the algorithm running, the algorithm told us, no, this is absolutely irrelevant. Your hypothesis is wrong.”
Analysing the data more closely, Schibsted found that the churn was related to the timing of subscription expirations and to reader engagement. The format and presentation of the product turned out to be more important than the content itself.
While Schibsted currently collaborates with a variety of outside vendors, Jacucci says that eventually it plans to phase out those working in areas it considers core to its business. It makes sense to outsource an ad-serving platform, for instance, but understanding how its sites perform, how its users interact with them, and how they can be improved are essential and should be handled in-house.
“What is more important … to an Internet company than understanding how our users behave?” Jacucci says. “User data is so important. It is core to the business, so it should stay in the business.”
To ensure data security, Schibsted has an attorney working as a dedicated full-time data privacy officer who:
- Explores data privacy issues for all its subsidiaries.
- Handles interactions with the relevant regulatory bodies in Norway and other countries.
- Maintains an overview of its security measures.
The goal of this harmonisation process is to create a user agreement that is fair and transparent and allows users to opt in or out of data collection while also allowing Schibsted to collect data across the properties and use it for the user’s benefit.
Jacucci drew an analogy to Google, which began as a search engine but now includes a collection of services accessible under a single sign-on and a single user agreement rather than its users signing on each time they open a new Google site.
“We never lose the beacon, because if you follow the user, the business model will work. If you follow the business model, eventually the user will be unable to. Our beacon is still the user, and we always strive to do the best for the user.”
Jacucci foresees many potential uses for Schibsted’s data, speculating that classified sites could analyse the stock of used items traded within a country and across several countries to better understand macro-economic trends. Though such information wouldn’t necessarily be sellable, it could be used strategically.
He also sees possibilities in analysing public sentiment and general reactions to certain types of news to predict social trends and in identifying gaps in the market where new services could be created to fill consumer needs.
In the end, the motivation for all of Schibsted’s work is to improve user experience and provide relevant services, Jacucci says: “It all links back to our core mission: to empower people in their daily lives. Going deeper into this mission, there is also a mission of empowering democracy. … The mission is filling the information gap that people have.”