Driving Vertical Content Recommendations

One of the recurring patterns I’ve run into with web content, illustrated here with the upside down triangle, is the inverse relationship between traffic and value.   The areas of a site where traffic is high, value tends to be low and vice versa.  The pattern reflects a basic dynamic: high traffic generally indicates breadth of content appeal while high value generally indicates specificity of audience that advertisers are willing to pay a premium for.  A more specific audience….tends to generate lower traffic.

The inverted relationship is not invariably true; there is high traffic, high value content (and unfortunately low traffic, low value).   But it is a recurring tendency.   And the more that content narrows within a site, from a broad overall positioning to specific topic verticals, the more the triangle visual will apply.

This concept relates to another, the importance of maximizing monetization within each site visit.  In a previous post, Is Increasing Visit Frequency a Realistic Goal? I’ve laid out some public domain Quantcast data to show that the visits per visitor dimension of loyalty has less variability across sites than the pages per visit dimension.    Visits per visitor tend to fall into a more narrow range most likely because of the mind-bogglingly enormous set of possible destinations on the web.    It’s just hard to squeeze another visit per month from a site visitor; getting another page or two per visit is a lower hurdle.   To monetize most effectively you’ve got to monetize them when you’ve got them.

The logic and mathematics of recommendation algorithms are beyond me.  But based on the theories above I believe they can be key to driving increased loyalty along the path of least resistance, getting another page or two from the visitor on the occasions when he/she happens to visit.

Furthermore, recommendations that drive the visitor deeper into the triangle, into content verticals that appeal specifically to a visitor’s profile and command higher value, are worth more than recommendations that drive visitors across the top, to other broadly appealing but lower value content.   The more recommendation algorithms can achieve this goal the more effectively they can monetize for the publisher.

Along these lines, here is a link to a cool visualization from the site Online Behavior that shows how a page can be personalized for a specific visitor:

“A visitor arrives to a website from a search engine using search keywords including “health”.  Why not display a personalized version of the page based on the visitor’s interest?”

Not a recommendation algorithm per se but an example of how a specific visitor could be driven to higher value content that specifically appeals to his/her profile.

 

Are Heavy Online Streamers Watching Less TV?

Nielsen’s first quarter 2011 Cross-Platform Report shows that the top quintile of online video streamers, in terms of time spent, consume somewhat less TV than other groups.    We can’t assume causality from the findings; heavy online streaming may reduce TV viewing or heavy online streamers may be lighter TV viewers to begin with.    But it’s an interesting finding either way, as shown by the squiggly line charts in the Nielsen report that I’ve lifted below:

There are a couple of different threads that can be pulled out of this information; an alternate visual may help to tease them out.   What I’ve done below is plot the online streaming quintiles, ranging from lightest (5) to heaviest (1), on a scatter plot with average daily minutes of streaming and average daily minutes of TV as x and y axes.   I’ve put 4Q 2010 and 1Q 2011 on the same scatter plot and connected the dots for each time period.

The first thing to note from this reconfigured chart:  TV viewing for the first quarter is on a much higher plane than the fourth quarter.   The two periods seem so different, as far as media behavior is concerned, that we can’t really trend between the two; all we can do it see if there’s consistency to the story.

There seems a great deal of consistency in the behavior patterns reflected by the streaming quintiles:

  • Quintile 5 in 1Q 2011: very heavy TV with virtually no online video.  I bet if we dug into this group we’d find an older demographic profile.
  • Quintile 5 in 4Q 2010 and quintile 4 across both time periods:  relatively light TV viewers with little online streaming.  This light/light pattern may reflect an upscale group that skews toward print rather than video and/or young males who skew low across all media.
  • Quintiles 2 and 3 across both time periods:  heavy TV viewers who are also moderate online streamers.   This may represent the mainstream condition.   At this point in the story, online streaming and TV viewing are rising in tandem…
  • And then there’s quintile 1, the heaviest online streamers who, as the Nielsen report points out, show somewhat lower TV consumption than the other groups.

A couple of points about the heavy online streamers:

  • Streaming behavior is highly concentrated into this top quintile.   The graphic shows how sharply the heaviest users pull away from the other four groups in terms of time spent with streamed content.  My back-of-the-envelope from these numbers shows that the top quintile accounts for 80% of the total time spent with streamed content; in contrast the top TV quintile accounts for just 45% of total TV time spent.
  • The data begs for deeper drill-down into this quintile; demographic and behavioral characteristics.  Why are they consuming so much streaming content?  Are they sending online video over-the-top to their TV’s?
  • Yes, their TV consumption is lower than other groups but not by a whole heck of a lot.   The graphic shows, as their online streaming behavior shoots away from the pack, their TV behavior declines a bit, but nowhere near in proportion to the way their streaming behavior accelerates.

Another way to look at this is to compare TV viewing for the top streaming quintile against average TV viewing across all the quintiles.   We can see the shortfall is slight; the under-consumption of TV is far smaller than the streamed content that these consumers have added to their lives:

  •  In 4Q 2010, among P2+, the heaviest streaming quintile consumed 14.5 minutes of online streaming a day but only 4.3 minutes less TV than average
  • In 1Q 2011, among P2+, the heaviest streaming quintile consumed 18.8 minutes of online streaming a day but only 8.0 minutes less TV than average
  • There’s a much sharper equivalence between minutes of streaming and dampened TV viewing among 18-34’s specifically.   Among this demo, for 1Q 2011, the heaviest streaming quintile consumed 27.0 minutes of online streaming a day and 21.5 minutes less TV than average.   It may be, among this demo, that the sharper trade-off reflects cord cutting; I cited a J.D. Powers and Associates study in previous post that suggests perhaps 6% of younger consumers have cut the cord.

I think this analysis shows a truism about people and media; the more options we throw at them the more media they consume.   Yes, the heaviest online streamers may consume a bit less TV than other groups but, as streaming becomes a bigger part of their lives, they consume more video overall.  They don’t trade-off one medium for another to the degree that they layer them all together.    The question then becomes:  how do they make these integrations and what role does each medium play?

 

Is Social Media Driven Traffic Less Engaged?

Outbrain released a study in April on the referral sources driving traffic to content pages, in terms of percent of referrals and the quality of the referred traffic.   They describe the dataset as 100 million sessions from over 100 premium publishers using their services.

When traffic quality is assessed by page views per session, search is the strongest traffic source followed closely by content sites with portals and social networks trailing.   Looking at an inverse measure, bounce rate, social networks and portals show the most one-page-and-out behavior, followed by content sites and then search…the least bouncy.    Looking at hyper-engaged sessions, defined as 5 pages or more, content sites lead, followed by search with portals and social networks trailing.

So…it would seem that social-media driven traffic is less engaged than traffic from other sources.   But I think there’s more going on here than meets the eye, as I’ll note on the other side of these charts from the Outbrain report.

I don’t think this finding is really about loyalty.   It’s about how specifically each of these traffic sources drives users to the exact content within the site that’s of interest to them.

When people enter content pages through search they may be searching for the site per se, using search as a proxy for typing in the URL.   They’re likely to come through the front door of the site,  the home page.   Multiple pages per session through search may not indicate loyalty but rather a fumbling about trying to find the specific content on the site that’s of interest.   In contrast, when they come through social networks they are being directed to a very specific piece of content referred by a friend…they go directly there, read it and bounce away.

The site experience may, in fact, be superior for social networks even though the loyalty metric looks weak.   And it may be a desired outcome for the publisher if that one page is deeper in the site and commands higher premiums than more generic content near the front.    Not all page views are created equal.    The value of social networks may not be sheer tonnage of page views generated but rather more targeted page views…driving visitors more directly to the pages that the publisher most wants visitors to consume.

Though loyalty (or lack thereof) looks the same for social networks and for portals, I would bet there’s a critical difference.   When visitors go one-page-and-out after being referred by a portal they are likely bouncing off the home page.    When they go one-page-and-out after being referred by a social network they’re likely bouncing off a specific post or article, some of the site’s more premium content.

We’d have to dig deeper into the data to prove my case….

But that, I think, is the story behind the story of these charts.

Is Over-The-Top Driving Long-Form Growth?

I’ve made the assertion that people will always prefer to watch long form video on the 10-foot screen as a human truism that will drive how the market evolves.   But I’ve felt a little pang about being so dogmatic since I don’t have data to support it.   And what about the iPad; might it not become a major outlet for watching movies and TV shows?

So it was interesting for me to find this CNET item from November 2010 citing a comment from Netflix CEO Reed Hastings:  the tablet craze affects us very little:

“People prefer large screens,” Hastings said. “So the impact of Xbox, PS3, the Wii phenomenon–huge impact. The impact of the iPad–it’s a great system, but the Mac laptops outstrip the iPad for Netflix viewing by a huge factor.” Long-form video viewing does not translate that well to mobile platforms, he asserted.

So this downplays the impact of the iPad, from a company whose strategy is driving the trends; they would be in a position to know. Though it suggests that the 4-foot screen, in the form of Mac laptops, is a major outlet for Netflix product.

I’d love to see data on number of streams and time spent with online video broken down by screen type – current state and recent growth. My hypothesis: the 4-foot screen dominates current online video consumption by far, but dominates less as length of stream increases, with the 10-foot screen being relatively stronger for longer form and the driver of long-form growth.

Though another paragraph in the CNET article suggests that this hypothesis may be U.S. centric, perhaps true only for a culture in which the TV is the center of life:

Former News Corp. executive Peter Chernin, who joined Hastings on the panel, said he agrees, with regard to the U.S. market, but that the story will be very different in developing markets, where big-screen TVs are less commonplace and cheap tablet devices will soon be readily available.

I’ve also argued that over-the-top will penetrate the home through the path of least resistance, assuming that Internet-connected TV, the simplest connection from a consumer perspective, will be the key driver. But both the CNET article and a June 8, 2011 MediaPost piece citing new data from CBS research chief Dave Poltrack suggest that video game systems are currently behind the growth in over-the-top consumption:

Poltrack contrasted Netflix’s remarkable growth with rather tepid adoption rates of other so-called “over-the-top” TV streaming platforms, such as GoogleTV, Boxee and AppleTV, but said that video game platforms such as Microsoft’s Xbox, Nintendo’s Wii, and Sony’s PS3 have become a major means of streaming TV programming, and that many of those platform users are actually doing so via Netflix.

So…it’s a box that already exists in the household, that’s already connected to the TV for some other purpose that is serving to-date as the main over-the-top conduit. And that makes some basic human sense.

Busting Data Silos

One of the things I’ve noted across…let’s just say years of analyzing different types of data is the degree to which data silos can obscure the ability to see what’s really going on in the world.   By data silos, I mean the set categories of information that people who analyze and report information tend to organize around…and the categories that those who make use of the information expect to see.

There’s an example in my post: Is Your Video Traffic Upside Down?   In general, site dashboards are set up to show traffic trends (what’s happening with page views) and video trends (here are the video starts, completes, etc.).   But unless you put one in context of the other (as per the post) you won’t see the interesting patterns that reflect how people consume video relative to text…by day of week, time of day, etc.

Data silos run rampant when the results of survey research are reported.  They’re often reported by question, the answers to question 1 followed by the answers to question 2.   When the real insight is sometimes only apparent when you put the answers to question 1 next to the answers to question 17, or 21…or some other context that tells the story.   In general, I think the best practice is to isolate the themes, the learning, what do the results of the survey say?   And then pull in data to support those themes in whatever order makes conceptual sense rather than the order of the questionnaire.

Different sources of information can become data silos.   In the digital world, editorial people look at traffic data and ad sales people look at revenue, impressions and CPM data.  But the reality of the site and its business can only be truly understood by looking at the two sources in tandem.  In the world of TV networks, ratings, revenue data and brand strength data from survey research may seem like three disparate pieces of a puzzle.  But when you put the three together there are stories to be told…

I’m not recommending that we throw everything into one pile and make a great big data soup.  Hypotheses about consumer behavior and market dynamics have to be used as a lens to focus how and where we slice the information.

But I think, to large degree, the process of analysis is a process of breaking the silos of how the data was obtained and reclassifying it along the lines of what it says about the consumer and the marketplace.

Essentially, the sense we make of the world reflects the categorical structure we use to see it…

And the essential job of the data analyst is to break the artificial, ingoing structures and build the one that best approximates reality…out there in the world.

 

Atomic Imagery/Digital Vision/Getty Images

How Will Networks Co-Exist With Over-The-Top?

I was busy poking fun at the Nielsen position on cord cutting the other day.   But it’s hard to deny that the trend seems relatively minor in the scheme of things.

A J.D Power and Associates study released yesterday shows that just 3% of residential TV customers have cancelled their multichannel subscriptions, 6% among Generation Y customers.   I’d argue that 6% among the younger audience is not insignificant; it’s a factor.   But those levels won’t be turning the TV world upside down.

This may sound odd, but I think the most important aspect of cord cutters not that they’re cutting the cord with their multichannel providers.   The most important thing is they’re likely to be driving online video over-the-top to their 10-foot screens.   The cord cutting aspect of the cord cutters…may be misdirection with regard to future trends.

To clarify terms, when I say “over-the-top” I mean streaming online video to a TV set rather than a computer or mobile screen and watching in lean-back mode, the way people traditionally watch TV.

Because over-the-top streaming is initially focused among the young, the tech-savvy, lower incomes, light TV viewers…it is more likely to replace traditional TV for these audiences.     But, as over-the-top becomes mainstream, only a minority of people will actually cut the cord.   Thus the key scenario, the one that will eventually play out for the majority of households, is that both traditional and over-the-top TV will compete for time in the 10-foot, traditional TV viewing experience.

I think some basic, human factors will drive to this scenario:

  • People will always prefer to watch long form programming on the 10-foot screen.   Watching full-length TV shows and movies on 4-foot and 2-foot screens will increase; the iPad will help drive its growth.    But the average length of an online video stream is still under 5 minutes.   I’d argue that the average length of online streams won’t really take off till the mainstream consumer is sending online video over-the-top to his TV.
  • Over-the-top will penetrate the home through the path of least resistance.   That path is the Internet-connected TV; the least number of extra boxes and wires entailed.   Blu-Ray players and Internet-connected game consoles will be secondary; boxes people already own for other purposes.   This path mitigates against cord-cutting; why buy a spanking new TV and then cut down your viewing options?    It also ties the rate of behavior change to TV replacement cyles and the briskness (or not) of new TV sales.
  • Network programming will continue to alleviate the burden of choice.    Lighter TV viewers may be able to subsist entirely on a diet of on-demand viewing.   But for heavier TV viewers it would be onerous to deliberately select everything they watch.  Channel surfing is a key part of their diet that can be supplemented but not entirely replaced with on-demand options.  And so…as over-the-top migrates to heavier TV viewers, a pattern of coexistence rather than replacement will be seen.

Here’s an interesting data point from the Leichtman Research Group.  30%  of all households report having at least one TV connected to the Internet via a video game system, a Blu-Ray player and/or the TV itself.   But only 10% of all adults watch video from the Internet via one of these devices at least weekly.    So a substantial number of households already have the capability; most don’t use it regularly.    Of course the numbers are much higher among Netflix subscribers…

When people have both forms of TV, traditional and over-the-top, available to them in lean-back mode, what do they do, what trade-offs do they make, what gets cannibalized?   If I were a network executive, this future scenario would concern me the most.

Frank Renlie/Photodisc/Getty Images

Yahoo Data: “Online Primetime”

Yahoo/Interpret recently published results of an online video study updating a study conducted in 2009.   One of the findings is that online video, which previously peaked mid-day, now peaks in the evening hours; the Yahoo report refers to “online primetime”.   As per their key chart, cited below…

From "Phase 2 Of Video Evolution Revolution", Yahoo/Interpret, 6/2011

This startled me because of both the similarities and differences versus my post of a couple of days ago, Is Your Video Traffic Upside Down.   I argued, for major content sites with a mix of text and video, that people are relatively more prone to watching video on the weekends and in the evenings when total traffic for these sites is at low ebb.   The Yahoo data shows that people are consuming more video in the evenings on an absolute basis.

Of course I was talking about content sites with a mix of text and video while the trend cited above is driven by Hulu and Netflix, video-focused sites.  That may be the difference.   I also wonder, this being self-reported data, whether evening online viewing is more deeply engaged in and therefore better remembered, more likely to be reported in a survey.

But the key data point that that this chart seems to beg for:  how much of total “online” video content is being watched on a computer screen versus how much is being ported to a TV screen?  What’s the trend for this?

What’s of interest to me is how much online viewing is in typical TV lean-back manner, on a big screen 10-feet or so away…versus on a small screen 2-feet or so away.

Because I would think the “online primetime” trend and a “online 10-foot viewing trend” would be happening in tandem.