"Personalization triples time spent on site at Winnipeg Free Press", Christian Panson, VP Digital

02 Nov 2016 | By Tobias Arns

Christian Panson, VP Digital at Winnipeg, tells how time spent on site triples and page views per visit double at the 140-year-old Canadian newspaper

Up until 2015 pretty much all the content on the website was distributed free. You didn’t even have to log in. It was largely an anonymous audience and it had grown over the years. But most of our readers were only consuming a fraction of the articles we produced.

We took the decision to put up a paywall, but we wanted to add value rather than saying that what you got for free yesterday you’ll have to pay for today. So we thought we could give subscribers something of great value by implementing personalized content recommendations for paying readers.

One important consideration for us a long-established news organisation was not to completely lose content curation. People come to our site in large part for the expertise of our editors in finding the big stories and presenting them in a compelling way.

So we wanted to blend that editorial expertise with elements of personalization. When you are live on our site, editorial can increase or decrease the amount of content that they want to be shown to everybody.

On a newsday where there are multiple evolving stories we might have ten curated stories – chosen by the editors – but on a slow news day, that might be set to three or four. The editorial team select the news that they will put front and centre on the site.  

We implemented continuous scrolling, so after the curated stories you scroll down the page and find content recommended according to your interests and behavior.

Recommending the right content, at the right time

We are creating content for each individual user according to their tastes. We select the best pieces of content over a certain time period. We’ll look at what is trending, take into account what the individual reader has looked at previously and what lookalike readers are enjoying on the site.

We use all of that to determine what the next article should be. As you scroll down the page, you get further away from curated content so we are not limited to breaking news. Once a user has read a curated story, it falls away and another story replaces it.

Implementing the personalization

We identify the user and give them a unique ID that we send over to Cxense when that individual comes to the site. All the content recommendation comes from Cxense.  

Cxense return a list of recommended articles to us based on our content, based on what is trending and based on what the user history has been. We then take that list of recommended articles, we load it on to the page and de-dupe it.

That means we remove any content that editorial has already selected as curated content and we de-duplicate a few other types of content. Then we throw that selection of content into a grid location where we mix in things like ads and editorial.

Every time a new piece of content goes on our site, we ping the Cxense server and they crawl it, so they understand all our content. They do a semantic analysis to understand the sentiment, the topic, to find keywords and categories and to create a “fingerprint” of every piece of content.

On any landing page, Cxense uses the fingerprints that they have about that user and the content and they also compare it to aggregate the fingerprints for all users that match this user in general. Then they use all that information to pick the best story for that user on that page.

About that ROI…

Average time spent on the site went from 282 seconds in September 2014 before we introduced personalization to 632 seconds in September 2015 after the new system was implemented.

Page views per visit were 5.65 in September 2014. After we rolled out the personalized content, this rocketed to 10.54 page views per visit in September 2015.

On desktop, average page views went from seven to 9.4 per visit over the time period. On mobile there was a huge jump from 0.9 pages a visit to 9.4. On tablet, the figure leapt from 5.4 pages to 19.6 pages per visit.

These statistics show just how effective the personalization strategy has been in helping us meet our goals of boosting both time spent on the site and the numbers of pages viewed per visit which all contribute to the bottom line of the business.