Friday, May 23rd, 2008
On Stats
There’s been a lot of discussion over the last few days about how video sites count viewership. This is an extremely important and constructive conversation to have. In general, blip is one of the most conservative video sites on the Web in counting viewership. We only count one view per IP address per session and we have a number of very stringent controls in place to prevent gaming viewership numbers, whether that gaming is intentional or not. We believe that it’s in our interests — and in the interests of the overall Web video industry — that we be conservative in measuring viewership. Failing to be conservative invites a backlash from advertisers, investors and content creators as they realize that they can’t trust viewership metrics offered by major Web video platforms. We don’t want to invite such a backlash. We want to be conservative from the outset.
The conversation started as a result of a TubeMogul research report that named blip and Metacafe as the most stringent in counting views. The report also blasted YouTube and Yahoo for basically counting anything as a view — including multiple refreshes of the video view page. The report also incorrectly stated that blip does not count embedded video counts. We do.
There were two big stories (one on NewTeeVee, one on Silicon Alley Insider) that discussed the TubeMogul report. I posted comments in reply to both, which I include below. Before going there, though, one bit of context. The NewTeeVee story suggested that the reason we’re so stringent in our counting is because we compensate content creators based on views. As I point out in my comment in reply to the NewTeeVee story, this isn’t strictly true. We consider advertising (and the attendant revenue share) to be one of the services we offer to content creators, and for us revenue share numbers have little relationship to viewership numbers. Our revenue share is based directly on one number: how much money we can make for advertising impressions on videos. There’s obviously a relationship between this revenue number and viewership (in the sense that more viewership generally means more impressions which generally means more revenue) but the two are, in reality, decoupled, as I explain in my comment.
I posted a comment in reply to Report: YouTube, Yahoo relax standards to inflate views on NewTeeVee:
We count video views and ad impressions separately. We have an incentive to count video views accurately for the benefit of our content creators. We also have an incentive to count advertising impressions accurately for the benefit of our advertisers. These are two different metrics which often but do not always coincide.
For example, a video may be viewed one thousand times but have two thousand advertising impressions because it carries both an overlay advertisement and a postroll advertisement. Or perhaps a different video may be viewed one thousand times but have zero advertising impressions because its content is objectionable to advertisers (maybe it’s of a sexual nature).
In each and every case we try to be as accurate and transparent in our calculations as possible. We have no incentive whatsoever to deflate view counts even though we do believe it’s in our interest — and the interest of the industry at large — to be conservative and accurate in our viewership accounting.
I posted this comment in reply to this post on Silicon Alley Insider:
I believe that TubeMogul has corrected their research about blip counting embeds. Their CEO e-mailed me today to apologize for their error.
In terms of restrictions by IP address, I understand AOL Survivor’s comments. We’re probably undercounting to some degree because we do restrict view counting based on IPs.
We’re not kidding ourselves, though. We understand that IP Address != Person. We understand that AOL viewers often come from the same IP, and we likewise understand that entire offices often come from a single IP.
That said, we face a challenge which is shared by the entire Web video ecosystem. We have to count views realistically, and in fact conservatively. Failure to do so invites advertisers, investors and content producers to lose confidence in viewership numbers offered by Web video companies… which would be disastrous in the medium and long term, as I’m sure you understand.
For the moment, therefore, we use IP addresses because they allow us to be suitably conservative. In the future we may move to a Flash stored object method for counting uniques — similar to the cookie-based approach that many Web analytics companies use. The only reason we haven’t moved to this approach already is because we serve a pretty significant slice of the Web video ecosystem that doesn’t use Flash (i.e. iTunes, Apple TV, Sony Bravia televisions, et cetera). So rather than go the way of YouTube (over-count, over-count, over-count) we’ve decided to slightly undercount and thereby maintain the confidence of advertisers, investors and content creators.
We could probably double, triple or even quadruple our viewership numbers by going the way of YouTube — but would that really be good for our business, or anyone else’s?
on Friday, May 23rd, 2008 at 10:52 am:
Your comment wasn’t posted to the Silicon Alley Insider blog. Maybe you entered a bad captcha? I just posted another comment linking to this post.
Just a heads up!
on Friday, May 23rd, 2008 at 10:58 am:
Thanks for the heads up, Anthony! I re-posted at SAI.
on Friday, May 23rd, 2008 at 11:46 am:
Interesting info. While it *would* be interesting/useful to see how many times a video was accessed from a particular IP address, so that the content creator knows how many times his/her video was ACTUALLY played, I agree with the style of undercounting for the sake of believability & advertiser confidence.
I’ve always figured multiple plays into my stats, so I’ve been basically undercounting my own views, now that I understand the blip.tv system. I know there are people that watch my videos several times, and I was subtracting that hypothetical number from my total view counts.
on Tuesday, June 24th, 2008 at 10:56 am:
I agree with your conservative approach because IP address is a way to identify uniques. However, you need to address the extreme overcounting of TVTonic views as evidenced by our stats of window media player views
on Wednesday, July 30th, 2008 at 1:35 pm:
I applaud you for your conservatism in this matter.
On IP address counting. You probably already count them both ways so why not report both the lower conservative figure and the higher aggressive figure. This gives the various stakeholders choice.