16

May

By Dennis R. Mortensen
3 comments
Why optimizing content placement on CTR alone is suboptimal

Some editors will use the Click-through Rate (CTR) of an Article Box* on the front page as a gauge for whether a piece of content deserves the exposure it’s being given. CTR is certainly better than not using any data at all, and surely more optimal than looking at how many readers are viewing my Article right now. I would like, however, to introduce two problems that discount CTR as the only gauge for success – whilst adding to the complexity of our data modeling, it’s also something that we’ve solved when predicting which Article to place where on the front page. Let me elaborate.

Let’s take a look at the following two articles from the New York Daily News, a Visual Revenue Customer; both published on May 13th, both local and both stories from Brooklyn. Also note that they will, for the most part, fall under the same Editorial Instructions which we have in place for our recommendations.

Problem 1:
If the above two local Brooklyn stories are appropriate for promotion in a given Article Box on the front page, and the predicted CTR is exactly 4% for both, which story do you choose?

With traditional real-time performance data only, the first issue is that you cannot really compare the two Articles and their CTR, as they will reside in different positions on the front page. Leaving the true comparison challenge aside, what is of the utmost importance, is that you know how much engagement the individual articles drive down stream – after the click. Knowing that one article drives much higher engagement than the other should certainly be a strong signal for favoring it – even though they have the same CTR.

For your geeky entertainment; the construction story has an engagement score and drive of 1.605506888, where the Immigration story has an engagement score and drive on 1.000000371. A fact that in essence almost doubles the revenue opportunity on our construction story, a fact that would even force the construction story on top with a lower CTR as it drives much more value after the click. Super interesting isn’t it!? :-)

Problem 2:
If the above two Brooklyn stories are competing for promotion in a zone where the editorial instructions allow other story types and categories; how do you compare and choose between a click into a Local section versus a click into a Money section?

With traditional content recommendation systems you might see everything flattened, something that unfairly allows low value categories to win over high value ones, because the immediate CTR is higher. Let’s imagine the above two stories compete with the following story from the Money section:

What do you do if the value (section wide RPM) is so disparate that 4 clicks into Local is of lower value than 1 click into the Money section? If that’s the case a 1.5% CTR on a Money story with the same engagement will win over a 4% CTR.

I hope these two basic examples highlight the challenges and limitations of just using CTR and the accompanying value of thinking above and beyond.

In conclusion – and sprinkled with a bit of advertising – the Visual Revenue platform is more than just click-through, more than just real-time data, more than just in-page analytics! We truly understand the full value of any type of content, both as a result of its direct revenue opportunity and its ability to engage and generate revenue down the road.

Cheers :-)
/ Dennis (@dennismortensen)

*Article Box – Definition
An Article Box is made up of one or more Articles Excerpts. Articles Excerpts within an Article Box are grouped and cannot be separated (e.g., A feature story on the financial crisis with links to detailed stories on individual banks). Only Articles Excerpts that cannot be separated are grouped into an Article Box, nothing more. Most Article Boxes contain only one Articles Excerpt.

  • http://uk.msn.com Chico

    Interesting post, thanks.
    Do you also take into consideration the ‘tone’ of the story? I.e. if you want to be known for more hard-hitting editorial how do you factor this in? How do you then know if this is working, do you measure the propensity for a particular article clicker to return more frequently?

  • http://visualrevenue.com/ Dennis R. Mortensen

    Hi Chico,

    >>Do you also take into consideration the ‘tone’ of the story?

    The primary purpose of the predictive models which we have in place – is to maximize the output – or in other words, we assure that the Articles which fall within the editorial instructions are monetized to the maximum extent. The Editorial instructions together hold the ‘tone’ of the publication.

    >>I.e. if you want to be known for more hard-hitting editorial how do you factor this in?

    That is a great question. We see this as a strategic decision which is to be implemented at deployment and not as something which is to be tested with on the fly. If a strategy is in place to e.g. focus more on local news, more on politics and see less use of galleries as a content tactic – then a set of persistent editorial instructions are applied and recommendations will take this into consideration in real-time.

    As a side note; we do have Editors who wants to override Persistent Editorial Instructions, such as us not being allowed to show Articles older than 18 hours in the hero spot and that it can only be from a set of hard-hitting traditional news categories. We call that temporary editorial instructions and they can be appended on the fly.

    Cheers
    d. :-)

  • http://tumbleweedmarketinganalytics.com/ Tom Wolfer

    Yes, a very interesting application of your ‘engagement’ metric to determine which story is more valuable and, as such, should be featured in an article box on a media website. Determining which is the most valuable content is key for a media website, and it is also important for any organization that aims to build an online community of loyal followers. I have written an article about analytics, social media and valuable content communities that addresses how valuable content leads to loyal followers which results in business results: sales, donations or ad responses, for example. I wonder if the Visual Revenue platform would be a valuable tool for determining valuable content within this context.