These publishers, not much different from everybody else, derived their revenue from channels such as:

• Display Advertising (CPM)
• Contextual Keyword Advertising (CPC)
• Sponsorships (CPT)
• Lead generation (CPA/CPL)

Think of the above list, as the publishers running internal sales departments, them using a multiple of advertising networks at the same time, having AdSense or similar applied to a subset of their pages, selling random front page take overs for 24 hours and unique lead generation for events as they happen etc.

My first thought, like most of you, I am sure, was to assemble all twenty something revenue streams in an Excel, and try to come up with a model for how we could quantify, the revenue increase of a promised page view increase. Assuring myself, that I took into consideration, facts like, unsold inventory, price differences between sections, the volatility of lead generation payouts etc. Honestly, first try, it didn’t work that well. There is simply too many variables, for which we know too little, which we need to take into consideration. Remembering that some of the variables go beyond those directly inferred from the revenue streams themselves.

I went back and had a second look at this, with the principle of the simplest explanation usually being the correct one. So I simply went with a monthly eCPM value [1].

eCPM = (Monthly Revenue / Monthly Impressions) * 1000

What’s beautiful, to me at least, in the above eCPM approach, is that it takes into consideration ALL known, and ALL unknown variables. AND It worked splendidly :-)

Let’s create an example for Publisher X:

Monthly revenue: $350,000
Monthly impression: 110,000,000

eCPM = ($350,000 / 110,000,000)*1000
eCPM = $3.18

Evaluating an initiative, which promises to provide an additional 22,000,000 page view impressions per month, the revenue opportunity is $ 69,960 ((22,000,000 * $3.18)/1000).

In conclusion. I suggest the possibility of replacing sophisticated online revenue opportunity models with simple monthly eCPM values, to calculate believable publisher revenue opportunities. I am super eager to hear about your models, so please share, and please refute my simplification.

[1] eCPM: Effective Cost Per Thousand

Update: Perhaps it’s time to baptize this not per the tone of Google (eCPM) – but as the more appropriate RPM (Revenue per Thousand)

Cheers :-)
/ Dennis (@dennismortensen)

Find below a pool of excerpts from the article:

• The Christian Science Monitor now sends a daily e-mail message to its staff that lists the number of page views for each article on the paper’s Web site that day
• Bloomberg News and Gawker Media, now pay writers based in part on how many readers click on their articles
• The New York Times, The Washington Post and The Los Angeles Times all display a “most viewed” list on their home pages – (Me: Not overly data driven, but it still counts)

And the usual, and still lovable example, of the perceived ultimate journalistic sweatshop:

• At Gawker Media’s offices in Manhattan, a flat-screen television mounted on the wall displays the 10 most-viewed articles across all Gawker’s Web sites. The author’s last name, along with the number of page views that hour and over all are prominently shown in real time on the screen, which Gawker has named the “big board.”

I personally think Gawker got it right for the most part, they should probably get slightly more sophisticated on their metrics (which they might well be internally), but the philosophy and thinking is well within my own comfort zone.

I recently commented about similar endeavors; Every news-writer has a Dashboard with Metrics determining his compensation and Analytics is building the Newsroom of the Future.

Picture: Gawker scoreboard for Reporters, which list most-viewed articles throughout the day. (Michael Appleton for The New York Times.)

Cheers :-)
/ Dennis (@dennismortensen)

Rational and maximizing in the sense that people should always act in a way that leaves them better off. A premise which is assumed accurate for most automatic optimization models; models which takes a set number of fixed variables as an input, and provides a calculated output, or Rational output if you will.

My point is this, some verticals, such as News Media, are not rational in nature, and as such, we cannot fairly expect our rational models to be instant true. I still believe that we (the industry) can and will succeed in increasing performance – and the above notion, is thus nothing more than a cautionary note-to-self on assuring anything I do, take that into consideration. If not directly built into a model, we should certainly accept a margin of, perhaps not inaccuracy, but conceivably a performance decrease from what was to expect.

I am sure you can imagine a number of scenarios, where an optimization model will predict, with indicated high accuracy, that you or a pool of people like you, will click a given Article Excerpt, but you do not!

Cheers :-)
/ Dennis (@dennismortensen)
- Image by: http://jamin.org/, post idea by: @jdaysy

I was spending my weekend in Albany, having good positive day long discussions with a publisher, about whether one should optimize for more Page views per visit (PPV) or more Article views per visit (APV).

I have strong opinions about the importance of setting proper KPIs (Key Performance Indicators) and applying appropriate subsequent optimization goals. In my assumption, that we are somewhat in agreement on the former, I believe it is fair to suggest that Article Views are a much stronger metric than Page Views. Some of my reasoning for choosing Article Views (AV) over Page views (PV) are:

• Article Views are more true to the actual atomic product of a News Media publisher
• Article Views are more directly aligned to the actual publisher cost structure
• Article Views are more directly related to the value the reader obtains
• Article Views are likely to be items in for-pay structures in the not too distant future – and thus more directly aligned to Revenue.
• Article Views are not as easily gamed as a metric as page views are

It is easy to come up with a dozen ideas on how one can game page views, and we see this all the time, from aggressive pagination of articles to aggressive use of galleries. Not only that, there is a strong possibility, that by gaming page views, you might subsequently de-optimize more true KPIs.

If I were to run a News Media organization, I would choose Daily Unique Article Views as my primary success metric. Think about it – and do let me know why your organization is not using Article Views over Page Views ?

Cheers :-)
/ Dennis (@dennismortensen)

My primary point is that, in any optimization activity, you must serve the overall purpose of the organization – and if so, can you honestly say that your optimization goals are KPI duplicates? and if NOT KPI duplicates, can you confirm them as KPI derivatives? And if NOT KPI derivatives, you must certainly be able to confirm them as KPI derivative proxies ?

If none of the above, then I believe it is fair to say that the further away your optimizations goals are from your organizational KPIs – the higher the risk of applying direct harm. (turning decent marketing initiatives into organizational de-optimization exercises).

My conclusion was (is) that you want to create marketing scenarios, where you can predict an organizational aligned outcome with a high level of confidence.

Find my slides below (which of course doesn’t do justice to my presentation, as the text is only pointers to the subject). Anywho, thanks to Andres, Luz, and team for a splendid day in Madrid (*even with the challenge of a power outage, mid-presentation)

Cheers :-)
/ Dennis (@dennismortensen)

The worth of an online non-subscribed news reader per month: $0.58
(16.2 page views per month * 3 Ads per page * $12 CPM)

Which makes pay-wall conversion rate debates more fun. Envision a $12 per month pay-wall was erected today, in a world where the above is true. This would force you to do roughly 5% unique visitor to subscriber conversion rate, to be as well off as before (aka 1 in 20 people need to sign up for your paid service before you supersede advertising revenue).

Cheers :-)
/ Dennis (@dennismortensen)

Sources:

• Project for Excellence in Journalism Analysis of Nielsen NetView Data.
  (Data are averages of September, October, and November of 2009)
• New Business Models for News Report, August 16, 2009
  City University of New York Graduate School of Journalism

I am by all definitions a capitalist and tend to believe that free markets solve all problems. Still do! But I also believe that it is a fair question to ask, whether we should tolerate that e.g. the New York Times, serve a pool of their likely less educated articles, to sub $30,000 a year in household income visitors ?

Even worse perhaps, we could foresee that publishers who do not have access to such dimensions for their visitors, will use proxies or possible use US-census information data on a ZIP code level. In such a scenario, personalization for a poor neighborhood will affect the information given to everybody in that area. I am sure we all see how this is not optimal or fair.

Let me provide an example, from a recent research paper by a colleague of mine, which provided facts on this type of personalization on a search level. Search represents content selection and presentation very well. When searching for “Wagner”, one segment was thinking of and was presented with Richard Wagner (the brilliant German composer) – the other segment was presented with Wagner, the paint sprayer company!

This example, imbalanced as it is, and by itself, is of course not disturbing and deconstructive to the social fabric as such, but I am sure you can envision a large pool of machine learned choices, activated to the extent where we move beyond just a digital divide, but a much more harmful information divide.

In conclusion, if we are set out to bridge the information divide in our society, aggressive news personalization, with pure revenue optimization for eye, might actually end up extending this gap. Does this mean that news personalization is not suggested, of course not, but it does mean that we have to think about this subject, on a more sophisticated level than which existing personalization and recommendation technology to use.

Think about it.

Please note, that in regards to advertising and it’s use of targeting, such as a demographic profiling and behavioral targeting, I see less of an issue, if any, by applying aggressive tactics.

Cheers :-)
/ Dennis (@dennismortensen)

Again, and as last time, I’ll leave the article to you, but have a look at the following Statements that are, not just, future fantasy, but very much existing data driven processes already in place.

• Every writer has a “dashboard” where the metrics determining his compensation — any scoops, hits an article attracts — are tracked.
• Writers’ salaries are tied, among other factors, to how many “market-moving” articles they have produced

Further to this, there is the acceptance that;

• Any breaking items from the magazine will appear elsewhere first

The above three bullets certainly indicates a very data driven culture and strong views on news media content valuation. Unquestionably something I personally support. Whether you are to one or the other side, this is surely an exciting experiment. An experiment of converting a hundred something old-school BusinessWeek writers to data driven Bloomberg writers – obsessed by metrics.

This is super relevant, even to those news media, that doesn’t run financial news related content (distributed on terminals like Bloomberg).

-AND when you ask such questions, I suggest you ask the reverse; why shouldn’t you have a real-time and accessible writers compensation dashboard ?

Cheers :-)
/ Dennis (@dennismortensen)

The first commentary I have, is the somewhat immediate need for online news media publishing metrics standards. Let me provide a simple set of offline publishing metrics, a set of quotes that show lack of standards and finally a suggestion on how we could perhaps look at this instead. All with the intent of moving us forward.

For as long as we can remember, Circulation in many ways defined the health of news media, such as daily newspapers or weekly news magazines. Circulation being the SUM of copies distributed to subscribers plus copies picked up at newsstands etc. However; a circulation count does not equal a Readership count, as most publishers will take into consideration that every copy is read by more than one person. If you agree on this layout, we then have a set of basic offline publishing metrics that look like this:

• Subscriber = Customers paying for a publication
• Circulation = SUM of copies distributed to Subscribers + SUM of Unsubscribed copies picked up
• Readership = Circulation * Count of people reading it on average

The above offline publishing metrics are fairly easy to understand, even though you probably, like me, spend most your day around online specific metrics. Please note that the concept of paying for, does not necessarily equal a payment in money. I identified earlier, that a newsreader is likely to pay in one of three currencies.

The first decent question is likely to be, If we are ever to compare offline to online; does the above three basic offline metrics even translate into a set of online publisher metrics or specifically online news media publisher metrics? Before we consider if that’s possible, and even a valid question, I would like to present a few quotes (beginning with one from Ken’s Newsonomics book.)

Ken Doctor quotes Ann Moore, CEO of Time Inc.

- “She made the point that every Sports Illustrated print subscriber is worth $118 a year to the magazine, mainly in advertising. Every SI.com customer, though, is only worth $5 per year”

In trying to understand the above quote, and assuming an error in using the metric, Customer, I looked up a previous quote from Ken:

- “I keep coming back to the numbers recently laid out by Time Inc. CEO Ann Moore: Each Sports Illustrated paid reader is worth $118 a year, mainly in advertising, while each SI.com visitor fetches a measly $5”

The  above quote then again links and reference the following quote from a Forbes article:

- “In 2006 Sports Illustrated generated about $118 in revenue for every person who paid for the print magazine, compared to $5 per online reader.”

Feeling a tad unsure on the legitimacy of the above metrics, I wanted to see if Ann Moore was quoted elsewhere, and Rafat Ali did indeed do an interview which was published on paid content, and in it you find the following quote:

- “The average reader of Sports Illustrated delivers about $118 to the bottom line in Time Inc. The average very engaged user of SI.com can generate about $5 in advertising contribution”

I am sure you, the web analyst, see a pattern, and with that pattern a number of issues. Some of those issues might even make the above $118 versus $5 incomparable, and render it invaluable, beyond the ability to use it as a punch line.

First. Let’s have a look at the 4 quotes and their accompanying 2 supposedly comparable metrics – and as they all reference the same set of proofs, they should essentially be identical.

• subscriber
• paid reader
• every person who paid
• average reader
• customer
• visitor
• online reader
• average very engaged user

I am not sure, if given this list, that I would immediately conclude that any number given for any of the above metrics would or should align, to the point where I could compare performance amongst them. If anything, this list shows that there is indeed a need for metric standardization, at least to the minimum of agreeing on a shared vocabulary. Trying to be a bit more fair, I paired up the metrics, as they were presented in the 4 quotes above:

• subscriber = customer
• paid reader = visitor
• every person who paid = online reader
• average reader = average very engaged user

Given these 4 metric comparisons alone, I would not be able to say that they reference the same subject. Beyond the obvious difference in-between them, we also have the actual issue of trying to e.g. equal a paid reader and a visitor. I believe I know what a paid reader is (Subscriber) and I am quite confident in knowing what a Visitor is. BUT I am also very confident in saying that those two are not the same – and that translation from the offline Subscriber to an online Subscriber is not done by just using a Visitor metric. If anything, and this certainly needs more work, one should align the following:

• Offline Subscribers = Online Users
• Offline Readership = Online Visitors

Where online users could be as firm as an online subscription to e.g. the Financial Times (which makes for a very good comparison to offline subscribers of the paper). But Online users could also be the SUM of Feed and Fan subscribers to e.g. the New York Daily News (who has a more ad driven monetization model).

Please note that I am not disputing the values per say, I am however a bit hesitant in believing the off-line $118 amount compared to the $5 amount being of any real analytical value.

Again, the book is great and it got me thinking about a number of issue.

In conclusion and as I continue down the rabbit hole, trying to figure out how Data and Analytics can help publishers increase their current performance. Standardization might not look like much of a performance or optimization technique, but not knowing clearly what your are optimizing against, makes the task not only difficult, but also dangerous to the point where you can make more harm than good on certain initiatives. So the laissez-faire metric attitude, should be exchanged for well defined and understood online publisher metrics.

Cheers – AND finally a post where picking the thumbnail picture took longer than writing the post ;-)
/ Dennis (@dennismortensen)

It’s already been a year since Dennis announced the launch of Yahoo! Web Analytics 9.5 on this blog. And one of the shiny feature announcements then, was the addition of demographic data, such as gender and age. Dennis, being such an awesome analytics expert he is, went on to suggest a few exciting ways of using these data in his screen cast overview of Yahoo! Web Analytics, which I recommend you watch before you read on if you are not familiar with the tool.

With the announcement, however, a few analysts from European countries questioned Yahoo‘s ability to deliver this kind of data in markets where Yahoo search market share in particular is almost non-existent. The Czech Republic like many other small European countries may well serve as a case in point. Take this purely as a rough indication, but my experience tells me that for Czech websites with as little as a couple thousand visitors, Yahoo! indeed is able to provide demographic data with a confidence level of about 80 % and for websites with more than 50 000 visitors you easily get the highest confidence level of 95 %.

Yahoo’s ability to collect this demographics data is down to the fact that it comes from the Yahoo! ID / Yahoo Cookie which you need in order to use the vast array of Yahoo! web properties, most notably Yahoo! Mail and the photo sharing website Flickr, which is popular in about every country in the world – and so seems to account for a bulk of the visitors to Czech websites with Yahoo! ID / Yahoo Cookie.

So for me, and I hope for many of you as well, no matter if you’re a fellow YWACN member or Yahoo! customer, the question is no longer about whether Yahoo! can provide demographic data about visitors to your website, because the answer is resounding yes. The question really is if you can trust the data, use it to analyze behavior and business outcome of different demographic segments and make decisions informed by such analysis.

I therefore set out to examine the demographic data, gender to be specific, provided by Yahoo! Web Analytics and compare them with data from NetMonitor, the official audience measurement platform in the Czech Republic which provides authoritative data that drive demand in the local online advertising industry.

Okoun.cz, the website in question that the data come from, is a traditional kind of message board that has been around since 2001. All data published below come from a period of 4 months from November 2009 through to February 2010.

Data are worth a thousand words

First and foremost, before we take a deep dive into the numbers, let’s be clear that all kinds of demographic measurement (maybe except for official census) are based on some kind of approximation. The idea is that if we determine a large enough sample, which exhibits the same qualities as the total population it suffices to analyze data from the sample and assert that any analysis outcomes are likely to be true of the whole population as well. It follows from there that the margin of error from such measurement approximations depends heavily on the sample quality and the confidence level is decreased with bigger total population and smaller sample dataset.

So much for theory, here’s the gender split reported by Yahoo! Web Analytics compared to NetMonitor.

From the chart above, it’s pretty clear that while Yahoo’s data show about 70 % of male visitors across the timeframe, NetMonitor’s data for male visitors fluctuate in the range of 55 % – 65 % so the difference between the two measurement systems is up to 15 percentage points.

The metrics are different though. While Yahoo! Web Analytics works with Unique Visitors (which really means cookies), NetMonitor operates with Real Users.

NetMonitor – methodology

I don’t want to go into much detail here, so just in a nutshell: NetMonitor is deployed on something like 95% of the Czech internet (in terms of traffic, not number of websites) and so they can differentiate between good cookies (cookies with a defined minimum lifespan) and bad cookies (below the lifespan threshold) and they approximate the number of Real Users from the good cookies, the number of pageviews generated by the good cookies and the total number of pageviews. This calculation is designed to accommodate the cookie deletion phenomenon.

Demographics data are collected from panel members using two methods: user-centric software based measurement (backbone of the panel, validated data, 1/3 of the panel) and site-centric pop-up surveys (less reliable data that are hard to validate, 2/3 of the panel).

Full methodology description.

Okay, now let’s take a look at the overall Unique Visitors and Real Users data in order to be able to examine the difference in gender split shown above and draw some conclusions.

Traffic differences aside, they’re different metrics after all. What is more important here is the relative sample size. It is quite evident that Yahoo! indeed has a considerable amount of data. In fact it is three times as much as the local audience measurement panel in this particular case. Websites with primarily international traffic may have this sample even bigger – I’ve seen up to 10 % of all website traffic identified with demographics data!

So can you trust the story they tell?

No data are 100% correct, but I believe it is safe to assume that with NetMonitor, more effort has gone into development of sound methodology for the local market – and so its data on overall gender split should be closer to truth. Therefore for top level reporting and in order to attract advertising suitors, website owners are better off using local audience measurement tools that provide rich validated demographics data, that can be easily compared with other websites.

Where Yahoo! Web Analytics demographics data are not very close in terms of overall numbers, the difference is most likely due to the fact that Yahoo! doesn’t have their websites localized in Czech. In effect, this skews the dataset in favour of more advanced users and away from the general population of internet users in the Czech Republic. Not everybody speaks English after all.

The screenshot below bears a lot of evidence to this fact. The Czech Republic are among the very few countries in the world where Google is not the #1 search engine. It is a local player called Seznam but the more advanced users, usually prefer Google as their first choice. From the Search Engines report you can see that the visitors identified through their Yahoo! ID and Cookie are indeed heavy Google users.

However, Yahoo’s demographics data still represent a lot of value to website owners seeking to better understand and communicate with their customers. It is possible to use this data within Yahoo! Web Analytics to isolate segments and analyze their behaviour in contrast with other segments on a very detailed level.

The chart below attests that the gender data are very, very close to reality. Okoun.cz has a ton of different topics you can discuss and of course, some of them are pure male interests and some on the contrary. The chart illustrates individual message board’s relative popularity measured by content consumption in pageviews for both male and female visitors.

If you speak a little Czech I will let you explore on your own from the following table of top 10 “female” boards what topics are relatively more popular with female audience. If you can’t speak any Czech though, just believe me, that reports like this are a lots of fun and insights at the same time.

Conclusion

Demographics data as a dimension,  has been absent among traditional web analytics reporting efforts for years, but now it enables us to talk to our marketing colleagues in their language, because in traditional marketing, demographics segmentation has always played a huge part. It is just about time we discovered its value in web analytics as well.

Demographics data in Yahoo! Web Analytics do not represent precisely the overall traffic for Czech websites, but they do seem to be very precise for the many visitors identified through a Yahoo! ID / Cookie. Therefore my conclusion is that, these valuable demographic data points, can indeed be utilized in segmentation and analysis of your website visitors – even if they primarily come from small countries such as the Czech Republic and especially when combined with dimensions and metrics not available in the audience measurement tool such as campaigns, custom visitor data, conversions and sales.

I’d like to invite you – from whatever part of the world you are – to share your experience on using demographics data in web analytics AND optimization.

Jiri