Paid content, local news and variable rewards: the unraveling of user behavior
The relevancy of content
Chapter three: defining true relevancy (a simple model).
This is part three in a series of posts on relevancy. View part two here.
Going by a dictionary, something being relevant is defined as ”bearing upon or connected with the matter in hand”. In other word, to say that something is relevant is to say that it is important for some current context. This is completely true when talking about news consumption, there is a context (for example: where you live) for which the content must be relevant.
Leading up to this blog post I have written about information need and the habit forming process from the perspective of a local news publisher. Relevancy is the glue that binds this together, but before we go any further I would suggest that you read the previous posts (if you missed them). There are summaries at the end if you only want the essence.
A model for interest in news
First a short note regarding the approach to relevance I will describe below: this is only to be perceived as a way to look at news consumption in a way that I hope will encourage you to think deeper about the usage of digital news products. Even though it is possible to quantify the parameters I list and use them for basic predictions, I do not claim that this holds up to a true academic standard. This post is an conceptual way to understand what key factors leads to the user choosing to interact with a local news service.
As we go I will break down the components of this model into three parts.
Part one: True relevancy
A simple approach on how to measure news relevancy
Let’s try a scenario.
You are very interested in local soccer. One night your favorite local team plays a game, but for whatever reason you are not able to watch it live. The day after you still don’t have any information at all about how the game went, but someone have emailed you a direct link to the article about the game, as reported on your local news site. You click the link and your web browser starts to render the page. But at this exact moment the browser stops, preventing the article to load, and you are instead prompted with an input field that pops up in the middle of your screen.
The pop-up survey asks you to write down the first five things that comes to mind about what information you hope to find in the article. After you have completed the list the survey also asks you to rate the importance of what you have written down. Let’s say that in total you have ten ”importance points” to distribute over you list.
Your list might look something like this:
- Final score (4)
- Best player on the field (2)
- Table standings after the game (2)
- Size of the audience (1)
- Standings at halftime (1)
As you finalize the list the prompt disappears and the article finally loads. Since you now have a list to go by you are able to compare the content provided to what you identified as the most important information you were looking for.
When you read the article you are able to check three items of your list. By consuming the article you learn about the final score, the best player on the field and the standings after halftime. Going by how you have weighted the different parts by importance, this amounts to seven (4 + 2 + 1) out of ten. In this very simplified example the article would (in this scenario) have a relevancy of 70%.
Relevancy in connection to the need for local news
As mentioned in my earlier post the need of information is major part in what drives news consumption. Whenever the drive (the urge) to search for and consume news gets strong enough it leads to the individual taking action.
Whenever the user decides to turn to the local news publisher to satisfy a information need the user interactions that are required should result in the user feeling rewarded. This is how a habit or a routine is created. Whenever there is a high enough reward the user will be more inclined to use the product again. Another way to express that you are satisfying a need is to say that you are being rewarded (as in, you feel that your actions renders a return). When it comes to news consumption, relevancy is the measurement of the value returned on user engagement.
If a product consistently is 100 percent relevant, the user will very quickly form habits around that product. Of course 100 percent relevancy might be impossible to achieve, but another way to look at this is to compare two different local news publishers products. Looking at the products from the perspective of a user, if one of them consistently outperforms the other in regards of relevancy you will over time always default to using the more relevant product. After a while the choice of product will be done subconsciously whenever there is a need for a news update.
Part two: Friction
What keeps you away
Let us do another thought experiment.
You are scrolling through your Facebook feed (suppose that you use it) and in the midst of the various post passing by you see that someone you know have shared a link to an article that looks interesting to you. You stop scrolling and focus on the headline and the introduction (the text connected to the link). Maybe there is a picture connected to the article link, and you also spend a moment looking at it. After just a few seconds you have assessed that the linked article really could be interesting for you. You shortly hover over the link, trying to decide if you should engage. But then, instead of clicking it, you resume scrolling for new content.
Of course, you might not feel that the behavior I describe above really correlates to you, but there is a good chance that you recognize the scenario. We constantly see this happening when content spreads over social media. Most interactions regarding news articles shared on Facebook actually takes place between individuals that have not consumed the content itself.
Why do we consistently see this behavior?
To describe it using one word we can label it as ”friction”. Basically what is going on is that when content is scrolled into view in a split second your mind will asess the information you have at hand (in this case the Facebook post). A decision will be made if you expect the information reward to live up to the level of engagement required.
If you decide to follow the link you will leave the Facebook feed, but you can expect the user experience on the linked website to have a similar approach to content distribution. You will still (probably) be presented by text, images and video, and on the website you will still be able to access more information by scrolling and clicking. However, there are also potential differences in the user experience, and those differences is what creates the feeling of friction.
Friction in user experience
Examples of potential friction when deciding if to follow a shared news article in a social media feed:
- The website is slow to load.
- There are interrupting ad formats seeking unwarranted attention.
- The text font is to small or large.
- There is a video that starts playing when the page loads.
- The layout is not aesthetically pleasing.
- The content takes to long to absorb/read through.
Friction may also be internal, as in that all users have an individual baseline for friction. There are for example some reasons for friction that are connected to the users real life context rather than the context of the product currently in use.
A few reasons to why a user might have a high degree of internal friction:
- There is not enough time to read everything that seems interesting.
- The language of the content is not presented in the users first language.
- The user have an opinion about the subject and fears getting upset or in a bad mood if consuming the content.
Part three: Expected relevancy
Hoping for reward
The personal relevancy of news content can only be evaluated when it is being consumed. However, going back to my first article in this series, one way or another every potential user/reader will need to use some kind of referrer (a hint about that the content exists) to access the content. This will always happen before the user is able conclude if the relevancy of the content is high enough to trigger the sense of fulfilling a need. Basically, this is what information distribution is.
In short (and obviously): Distribution of content is the required step for a publisher in order to acquire users.
To draw the users attention toward the content “teasers” are needed. A teaser for a printed newspaper might be a headline on the front page of an issue, or a daily placard in connection to where newspapers are sold.
In the digital era, a teaser is the short form description of some linked content. The front page of a news site or app often has many teasers to individual articles or videos. If the user follows a link the expectation is that the content will appear on the next page view. External plattforms are used the same way, when a link is shared on Facebook or Twitter the system automatically picks up some information from the content description (including a headline and most often a image) to be presented as a link.
Going back to the scenario of seeing (but not following) an article link in the social media feed this might be the other part of what makes the user decide to engage or not. In some cases there is possibility that the link description of the content is bad, and even though the user would have felt a reward in consuming the actual article the description used is bad at “selling it”, resulting in that the user is missing out.
However, the real challenge of expected relevancy is when the opposite is true. Any time a user expects a high relevancy (and therefore a good reward) from the link description but the actual content does not live up the expectation there is a sense of punishment.
In the brain, the same thing goes for punishment as for how rewards are handled. A punishment is a punishment, not depending on context. No matter if you put your hand on a hot stove, or if you follow a link in a news feed and gets frustrated by the experience, you will come away with a feeling of letting yourself down. And exactly as in the case of creating user habits by feeding them with constant reward a user habit dies out if the user is punished repeatedly.
It might seem strange that any publisher would chose to punish the user, but there are at least two clear reasons for why this is happening.
The business of punishment
One reason to why the user experience might be causing frustration is often that some business model currently in use is somewhat counter productive.
If a publisher is unable to increase the value of ad exposures to meet a budget the way forward is to increase the traffic volume (and thereby the number of ad exposures). A simplified way of looking at this is that the publishers actually gets payed for punishing its users. The more intrusive the ad is for the user experience the higher the price (measured programmatically as CPM) for the advertiser. This business model is often called a reach advertising model, and if this is the leading business strategy for a publisher the arrival of new users is the primary focus. To be able to secure a steady stream of reach ad revenue the content links (the teasers) need to repeatedly beat other publisher for the user’s attention. This often means aggressive overselling, and today this is most commonly labeled as click bait.
Another common business model that have some punishment “built in” is the paid content approach. This means that a part of the required user engagement consists of creating a user account and paying for the content. The short form description of this is that the publisher relies on that the expected relevancy of the content is high enough to motivate a user to suffer through a (fairly unannounced) paywall experience before reaching the desired content. The flip side of this approach is that the business model rely heavily on that the content, when consumed, lives up to the expected relevancy. The term churn is used for when a user is displeased with the relevancy (and thereby the value received for paying ) and decides to end the subscription.
The business of persuasion
Another reason to why a user might feel let down by some content (and its publishers) is that the information is misleading or that there is a hidden agenda. A common scenario for this take place in connection to political campaigns. Sometimes a publisher is willing to publish counter informative articles with the goal of creating a movement, or to distract from proven facts published elsewhere.
The discussion about how ”fake news” is disrupting democratic processes is constantly ongoing, and put in to focus from time to time. It is important to note that even fake news are produced to serve a need, but that need is in many ways not aligned with the consumers need for trustworthy news services. The fact that fake news exists might be useful (relevant) for a person or an organisation with the need of creating a public opinion. This might be used for driving a change in society (most often by displeasure/hate) but it might as well be a part of product marketing or an individual narcissistic agenda.
Putting it all together
Going through the three main aspects of relevancy and friction described above we can put together a list of components that are in play when a user decides whether to engage with a news product or not.
The four components needed to understand the probability of user engagement (UE):
True relevancy (how well the content suits the information need)
Expected relevancy (how relevant the user predicts the content to be)
UX (product friction) (how hard the publisher makes it to consume relevant content)
User friction (individual features the user has that might prevent content consumption)
Predicting user engagement
Is your news product good enough to form a user routine?
A way do describe the probability of engagement could be to express it this like this:
UE = (TR + ER) ÷ (UXF + UF)
As mentioned in the introduction, this is not a complete model. The purpose of this example is to open up for some discussion regarding how the different parameters affect the outcome. It may also be useful to highlight inconsistencies that in turn might warrant the inclusion of additional parameters.
I will now briefly dissect the model by going through what I have left out of scope, with the purpose of giving good foundation for you to hopefully come up with one or many alternative solutions. And by doing this I will also start to wrap up this blog series.
Constraints and future improvements
Lack of normalization
Since the model above propose an outcome in the form of probability the value should preferably fall between 0.0 and 1.0 (as in 0-100 percent). As of now the result might land well outside of that range. If the total relevancy is 0.6 and the total friction is 0.5 the probability of engagement will be 1.2 (120 percent). To be able to actually quantify the model some normalization will be needed.
For now, all parameters are weighted the same. If we were to actually apply this to real data we would almost certainly need to adjust the relative importance of each parameter. For example, compared to expected relevancy the actual relevancy of the content might have a much higher impact on if the user will create a habit for using the product.
The value of having a low UX friction might also be completely nullified if the users internal friction is too high, or vice versa.
I am also tempted to propose that instead of friction being the sum of (UXF + UF) that friction might be calculated as (UXF · UF). But at this point that would just add a false sense of sophistication, I feel.
Dealing with constants and external factors
To be in the business of publishing digital news content is to operate on a sliding platform. At any given time a user behavior might change due to some unforeseen reason. The struggle for user attention takes many forms, and sometime certain trend can lead to sudden shifts. If a publisher realies heavily on a business model that is fueled by a click bait strategy and a social media platform decides to prevent click bait as a form of content distribution, then that external factor will greatly influence the outcome.
A publisher who primarily produces local news also need to take the local population size into great consideration. Event though this value slowly changes over time it might be used as a static number even when doing long time forecasting. Since there is a estimated (somewhat small) finite number of individuals that might find the content relevant the distribution methods will need to be more fine tuned than when dealing with a larger audience. If enough friction is presented for a small audience the viral distribution will halt, in line with the laws of herd immunity.
When building a model to be used for predictive reasons additional parameters, as those mentioned, should be needed.
In this blog series I have tried to focus in on the underlying process that leads to consistent user behavior in digital news products.
There are three main areas in which we are able to dig for clues for how to really get to know the user behind the statistics:
1. The information need.
2. The habits and routines that drives consumption of news content
3. The expected and true relevancy of news content, and the friction that keeps the user away.
To truly learn how a news publisher is able to attract users in a fiercely competitive digital landscape is a daunting task. More often than not the publisher have a long tradition of being a strong local presence, and from time to time because of that legacy the digital user base is taken somewhat for granted. But to think of the users of a digital news product’s as a transformed group of newspaper reader is both naive and dangerous. The transition to digital distribution was not driven by changing needs on the publisher side, it happened because of a revolution in how information can be found and consumed in everyday life.
More than anything it was the head start with having professional local journalists already at hand that laid the foundation for a stable loyal user base in newly developed digital products. The need for local news did not change dramatically during the digitalization of society, what changed was the number of alternatives for getting that type of information.
To stay relevant in the future a publisher needs to have a clear approach on how to stay relevant. Up until recent years a news service (such as a printed paper or a radio transmission) could do this by producing content well suited for the average person in a geographical cluster. This is no longer true, and today many users identify more closely with online groups than with people living in their neighborhood.
In the era of personalized services and tailor made content recommendations the key factor in adapting to the ever faster changes in habits and user needs is to really understand what made the local news publisher relevant in the first place. With a firm foundation in these basic insights it will hopefully seem obvious how to approach future challenges in the business of providing the type of relevant information that helps build a society.
Magnus Engström is Head of Data Strategy at MittMedia — Sweden’s leading group publisher of community news brands in Sweden.