« December 2009 | Main | February 2010 »
This is a really fascinating blog post from dolores labs that I'd suggest you dig into and read. Specifically across the board the internet doesn't really capitalize on these natural human behaviors in a way that I think would benefit everyone. The internet changes everything and understanding data like this (and also stats data so folks aren't surprised bythe birthday paradox) will change the game.
Enjoy.
On the margin optimizing to make your experience (yes you, the person reading this now) better is not a good idea for most of the advertising supported websites you use (the one notable exception is Google). I say this because if you are reading this post you must be a web power user :) so view a lot of page views. Their time is better spent working on their least loyal users.
Here is why (using data from my sites).
The deeper a user is into a site the less likely they are to click. The following graph shows CTR by depth (in page views) into my site for a user on ads. For a CPC driven advertising marketplace getting someone with 5 page views to view 6 is half as valuable for me as getting someone with 2 page views to view 3.
This following graph shows the relative value (for me) of getting you from 0-1 impressions on my site, 1-2, 2-3, etc… As you can see the value added is rising to 4 page views and then starts to decline after 4 page views. A while ago I did another post on this showing the double impact of google adsense sell through rate http://www.alexschultz.co.uk/weblog/2007/07/google-adsense-.html as you increase page views per user.
Now don’t get me wrong you are probably a massively valuable user to all the content sites you regularly consume. Below is the revenue per user curve for my sites based on how many page views you have viewed. The last bucket is 6+ (not just 6) so you can see someone who views 6+ pages is 5 times as valuable to me as someone who views one.
That being said my biggest return is by getting all the folks who view 1 page to view 2 and that’s pretty much true for every content publisher I have ever seen (have done a little side consulting for a few companies).
I guess the big question I don’t answer here is how much investment does it take to get a power user to view one more page than getting a low use user to view 2 pages rather than 1 :) Perhaps my note on reducing bounce rate via recommendation engines gives some answer to that.
I hope you stay with me and read them all :)
One thing I am not very good at is search outside of Latin-1 character sets so recently I have been taking a little look at Japan. I found it really interesting to see the Google search page covered in ads for google’s other products and yet the Yahoo search page completely minimalistic.
No real thoughts, just interesting.
I just completed a panel on facebook advertising at Affiliate Summit West.
I always enjoy affiliate summit because it is full of people who really understand internet marketing. It’s always good to talk to people here because their entire livelihood is based on how good at internet marketing they are. I find that when someone is using their skills to put food on their table that a lot of the theoretical rubbish is thrown out and folks really really focus on doing what works. Often in a big company people get very cerebral and theoretical and engage in plumping their egos about how clever they are rather than focusing on what works. For all the well publicized negatives about the affiliate industry that is a huge positive for me. This industry really knows it’s stuff.
The single most interesting thing about the panel today for me was Markus’ point about how plenty of fish see huge branding impact from Facebook ads. He called out that when they run an ad campaign they see an up tick in total traffic well beyond the direct clicks and when they stop that campaign often a chunk of that uptick continues afterwards. This is I believe the crux to where internet marketing is going in the near term. More focus on the harder to measure branding values online… if TV can do it, so can the internet!
Anyway thank you to my fellow panelists for stepping up to talk about fb ads:
Here are some totally self interested positive twitter comments a friend sent to me which I wanted to share:
My latest hobby horse is focusing on the basic difference of median vs mean. My belief is that for content based sites who make their revenue from ad sales (like YouTube, Yahoo, Gawker, Newpapers etc…) a focus on optimizing the median page views per user is far more important than optimizing to increase the mean. It is worth noting that sites like Google (where the goal is to get you to leave the site by clicking on a demand fulfillment ad), eBay or Amazon this is not true.
I am going to try and post a number of articles with data from my own sites that show this but I am going to start with one simple post that quickly gives two examples of median vs mean page views per user and resulting revenue per user.
My AVERAGE (mean) page views per user on paperairplanes.co.uk is 4 (it’s gone down a bit since I’ve focussed a lot on videos) and it’s pretty stable.
My median page views per user is only 2, half the mean:
My AVERAGE (mean) page views per user on cocktailmaking.co.uk is 8 (it’s gone down a bit since I’ve focussed a lot on traffic generation in the “cocktail name” space where people come in, view the cocktail then leave) and it’s pretty stable.
My median page views per user is only 1, 1/10th the mean:
Paper airplanes has a revenue per visitor of 2.1x Cocktail Making and hopefully through the coming posts I will be able to prove to you that this is due to the fact cocktail making has a lower median page views per visitor than paper airplanes even though it’s mean page views per visitor is 2.5x that of cocktail making.
Perhaps this is all very obvious to you and if so I apologize… don’t read the series but if this is all so obvious why do none of the big analytics providers include median by default into all their metrics?
I hope you stay with me and read them all :)
Had to post this… so cool to see so many people updating facebook about the earthquake so fast
It’s actually pretty fun to search for “earthquake facebook” and see what people are thinking about regarding Facebook and the earthquake updates!
Part 3 Median vs Mean: Cume vs GRP in offline media and what it means for internet marketing
One thing has been true for as long as marketing has been around: It doesn’t matter how good your ad is if no one sees it.
Dave Goldberg (CEO Surveymonkey) gave a great talk at an event I was at recently and shared an anecdote about MTV’s audience on 30min shows vs 5min music videos that is worth repeating and set me thinking a lot more about reach and frequency in internet marketing.
GRPs vs Cumulative Audience Rating
In television measuring how many people see your ad is done using set top boxes and often a combination of “gross rating points” (crudely: what % of the possible TV audience sees a given show, detailed description linked below) and cumulative audience rating (similar but what % watched a show in 15minute segments of that show).
In the 1980s MTV did great on GRPs because people would tune in to catch a 5minute video. The problem for MTV came when cumulative audience became a big measure in the 1990s, people just weren’t sticking around for 15minutes. If you followed a Stone Roses track with Madonna the two audiences were not the same and so the stone roses people switched off and by the time you reached the break (where the ads which funded the show were) MTVs audience was a fraction of it’s GRPs because the cumulative audience (who watched for a full 15minutes up to the break) was small since you just don’t like every video in the hot 40.
The following slide (from this presentation: http://www.scribd.com/doc/3046688/ANALYSIS-Audience-Cross-Rating) shows even today how rapidly the audience figures drop off with time on a given channel (MTV, MTV2 and CBS).
With 30minute shows like Beavis and Butthead MTV managed to get an audience that stuck around through the commercial break giving the ads greater exposure and making marketers really happy. Not only that but their audience seemed to love them too and their total GRPs went up too which led in many ways to the Hills etc… etc… and a lot more revenue for MTV.
MTV optimized to drive up their median user by changing their show format. They did not optimize to drive up their mean user by making their 5minute videos so perfect that the folks watching 1hr a day would watch 2hrs. This was a great decision.
So how does this apply to Internet marketing
In display media for internet marketing people buy impressions but this is really nuanced. You should not be just buying “impressions” you should be focusing hard on what the reach of those impression are and ensure whoever you buy from gives you a frequency cap. Below is the distribution of visits on my paper airplanes site in the last 30days.
As you can see if you want to hit 160k people with my site you can BUT after the first impression you will only be reaching 90k people and by the 4th impression you will only be reaching 54k people. I could make total delivery of impressions huge while still only giving you a small reach by not giving you the first 4 impressions. Frequency caps matter a lot.
Conclusions
In order to maximize the chance of your ad doing well you have to maximize it’s audience.
Understanding the data you are using to determine reach is therefore very very important.
Anyone who can offer great reach and high frequency will make a killing in advertising sales and THAT is a challenge of optimizing for the median user (in paper airplanes case the person who see 2 page views) and not the mean (who in paper airplanes sees 6 page views).
Planned posts:
I hope you stay with me and read them all :)
References:
Cumulative Audience Rating: http://www.answers.com/topic/cumulative-audience-rating
Cumulative Audience: http://www.answers.com/topic/cumulative-audience
Gross Rating Point: http://www.answers.com/topic/gross-rating-point
Presentation on MTV time on channel: http://www.scribd.com/doc/3046688/ANALYSIS-Audience-Cross-Rating
January 27, 2010 in general comments, median vs mean | Permalink | Comments (3) | TrackBack (0)