Why Mittmedia simplifies the startpage and is building a personalized newsfeed
Loyal customers are not something given to you, but that you deserve. If you want users to pay for content you need to create a value greater than the effort it takes to consume it.
That was what made us simplify Mittmedias sites. And, while at it, we came to personalize the content and at the same time minimize editorial work with the classic startpage.
Mittmedia’s reason to be is qualitative local journalism. Not only to report on relevant things but to rise above the media noise in order to connect users with our content. It is also the baseline for the part of our business as we connect advertisers and active users.
We are in the same game as players like Facebook, Instagram, Netflix, Youtube and all the publishing houses of the world. What unites our competitors is that they are all super skilled when it comes to serving the single user a unique content. And by personalizing they can continue to customize by adapting the product to the context in which the user is consuming. A game changer that has made our readers used to interfaces switching depending on when and how they enter our products. And something that they expect from us if we are in the game of winning their attention via smartphones, at desktop or — any day now — voice.
Finding our place in our reader’s everyday life
In order for us to understand our customer’s needs and habits we built a data platform, Soldr.
Through Soldr we began to understand our user’s behavior and we got the possibility to study how they act, from the first scroll through a flow, via longreads over coffee to quick pop-ins after a notification in the app or while tv-commercials are interrupting their movie.
Based on the data that Soldr provides us with we can make a content analysis that considers parameters such as:
- What users are reading
- When they are consuming our content
- How much of an article they read
- How they interact, ie by clicking and enlarging specific pictures
- How they consume content based on its geo parametric
- Which news are relevant to which readers, depending on the channel
- Which content we supply today without the users engaging in it
In addition to this, we have the insights that we get from deep studies, daily encounter with users and quantitative studies such as Net Promotor Score and pollings booth within and outside our user base today. All in which we ask users to tell us about their routines and habits in daily life.
And it is in that layer of insights that we have encountered user feedback in the sense of readers saying:
‘The site is very cluttered and doesn’t feel exactly modern. There is a lot going on at the same time.’
Understanding that gave the site team at our internal development department the idea to do a test to see if a greatly simplified flow would give the users better focus.
Quick and clearly scoped tests have, during the last two years, become our daily routine in order to find time to iterate our product, Mittmedias news sites, apart from all the incoming requests that the organization has for the product.
How we got to it
During the Winter of 2017/2018 the first test with the personalization algorithm was done within some of our apps. One of the things that was done was that we looked at relevance based on readers that live in geographical areas where the region is bigger than one bigger city and suburbs.
And when my team came to me in the spring and wanted to boost article consumption by simplifying our flow the idea of a test came to live.
The first thing we did was that we during a couple of weeks gave a test group with logged in users at Östersundsposten.se a startpage where the editorial team had just hand picked the first three pieces, and the rest was automized on trending content. We also simplified the ad formats in order to give the flow a better harmony.
The test was successful and we could see that both article consumption and ad exposure was good enough for us to go further. As a bonus the editorial team reported on less manual work as they now only had to handpick the top stories (even though the time still was just in theory given that the test only included a small percentage of the current logged in users).
Throughout the years we have learned that it is essential to take the next step towards scaling solutions that are giving us the wanted effect. Therefore, as we heard that our colleagues in the team that has built Soldrs data model were ready to let us use the algorithm for personalization to do the automated part of the flow — we went forward.
During a new couple of weeks the logged in users at Östersundsposten now divided into eleven different clusters based on what and how the had consumed our content the previous 24 hours.
Again the results gave us green light to move ahead. Article consumption was as high or higher than on manually built startpages without personalized content. That meant that regarding the disposition of our startpages the machines were as efficient or more so than humans.
Now the remaining question was: were we ready to scale this to the rest of Mittmedias sites? Will the user behavior stick over time or will it change?
Starting in July and the following seven weeks the test was repeated, and this time also with test and control groups on VLT and Sundsvalls Tidning in order to study how logged in users consumed our content over time. And again we could see that the simplified startpage gets the job done better than the traditional startpage. Hence we know, through inhouse data science, that there is a strong connection between the number of articles a user consumes and his tendency to churn, we know how important this factor is. As a bonus, the automated flow will cut corners for the editorial team that can now focus on producing new or refined content.
Enough with the testing — let the users in
Given that the quick and well-defined tests are our best tool in developing we have used it a lot and learned to let the users benefit of the iterations as soon as possible. In this case, it also handles a need within the data team that, as we scale the personalized flow on all sites, gets the possibility to get real-time training data for the continuous trimming of the algorithm’s different parts. And trimming will be necessary, that we know. Actually, that is exactly what makes our competitors so sharp: their ability to take into consideration what the unique user has in common with other users in the same geo position, at the same age, with the same preferences, income or car. Yes, the list goes on and on.
And by that: as soon as we have personalized our sites with an automated newsflow the work begins with raising the level of relevance. But already from day one, our customers will be given a greater value than before as they are encountered by a newsfeed that is based on not only the users previous consumption but what other users consume and the editorial team has assessed as qualitative local journalism.
The current implementation is set till October and by that Mittmedia is entering a new chapter with our news sites. With this iteration the user will have an incentive to log in and we lower the cost of consuming our newsfeed. Then again, working with driven, creative people there are a number of things and iterations, features and tweaks to be made in order to make it even easier for the user to choose qualitative journalism. But before we get to that we should let the users in so that they can guide us, by their choices and habits, and let us know what to do next.
A special thanks to: Magnus Engström, Thomas Sundgren and Ellen Jernström.
Find a Swedish version of this blog post here.
