Cogblog

Companies are gearing up to make an absolute fortune in behavioral targeting by scraping Google search terms off of publisher referral data.  (Some already are!)

The theory is simple: Search has credibility with advertisers as a very effective behavior with high conversion rates.  The only people with active search behavior data today is Google.  When someone searches for something and then clicks, their browsers HTTP_Referrer is set to the last page they visited, possibly something like: http://www.google.com/q=Searching+For+Cars (Where “Searching For Cars” was their query on Google).  So if a behavioral data gatherer like Magnetic is sitting on a publisher with a lot of search driven traffic (most web sites), they can gather a lot of referrer data.  Logging this allows for (As Magnetic’s web site describes it):

Magnetic™ is search re-targeting. The Magnetic data marketplace empowers advertisers and publishers to use search data as the key indicator of intent and re-target campaigns to the most relevant audience online. With more than 270 million search profiles, Magnetic significantly lifts the value of media and improves campaign performance.

A few years ago I had a similar, albeit less brilliant idea: It would be easy, in the same vein, to write a few lines of javascript that compared lists of sites to a browsers user history.  In that way, you could easily sell conquesting: Dell could show ads to everyone that had visited apple.com AT ANY TIME IN THE PAST.  All, without requiring Apple’s consent.  And I was able to get that data because I had a deal with Website X that allowed me to run tests against user history data in the browser without the user knowing.

Now, maybe a post like this will cause some start-ups to start coding away on this (no need, really.  It is only about 10 lines of javascript, then you have to ajax-ly forward it to your server.  Here is how.).  That is not my intention, obviously.  The real question is: Is that OK?  And if it isn’t, how is it different than the search retargeting we are seeing blossom in the market today.

What about other uses of that technology?  Could Microsoft offer special discounts to people that had visited Apple.com recently?   Is it OK to essentially steal that data from a user?  Should you know what competitors a potential customer is talking to?

I posit that this whole area is about as far into a potentially unethical gray area as one could go.  There is basically no difference between any of these examples and they all make me want to turn off referrer data in my browser.

What do you think?  Are you OK with every website knowing every web site you visit and query you type into a search engine?  That seems like where we are going.

Just wanted to do a quick shout-out to the first sponsor of my blog, AppNexus.  AppNexus is clearly poised for transformational brand awareness the likes of which few companies in the space have ever seen.

Thanks, Brian O’Kelley!

Reading Eric Porres post about transparency in behavioral targeting on AdExchanger reminded me that I have been meaning to crank out a post about vector targets.

If we want to really simplify behavioral targeting, a behavior can be distilled to three attributes:

1. The event: Is this a visit to Kelley Blue Book or a visit to Yahoo! Autos.  A visitor to one or the other could perform differently.
2. Recency: Was it a week ago or a month ago.  Was it an hour ago?  True story: Retargeting cookies you dropped in the last hour perform a zillion times better than anything else you are doing.  Are you watching those?  Are you segmenting them out?
3. Frequency: Did they visit twice?  Three times?  That performs differently.

Painting in extremely broad strokes here, there are a couple of kinds of behavioral bucketing strategies that are commonly used.  The simplest would be to simply describe the above behavioral attributes: “Our auto intender population is made up of people that visited AOL Autos at least 3 times in the last 60 days.”  This bucketing approach is nice in that it is quite transparent to an advertiser what constitutes a population member, however the population size is fixed, which can limit deliverability for a campaign.  We only have X number of people with those attributes.

Typically, the way a vendor might work around this is tracking populations with varying recency and frequency.  So if someone wanted to spend a bigger budget than could be delivered at 3/60, we have a 2/60 or a 3/90 that we could offer to advertisers and that becomes our auto intender population for this RFP.  Another way to flex the population is by mixing behaviors.  We might acquire auto intender data from 10 different publishers.  By squishing this together, it is possible to generate larger populations – “3 visits to any of these 20 sites in the last 60 days”, but it makes regressing performance more challenging, particularly because the data points around frequency, recency, and the sites used to generate the data are typically not shared with the advertiser.  But then, advertisers aren’t asking for it.

In an effort to recognize the performance variability in different inventory, as well as recency and frequency, most behavioral targeters have transitioned to a vector approach.  To illustrate from an anonymous vendor deck that most of you have probably seen (and frankly, an awesome example of modern BT):

So as people surf the behavioral data providers data sources, their scores against a variety of behavioral targets are incremented and decremented based on how well the behavior provider thinks they will perform against an offer.  This is nice in that it allows you to easily keep track of many variables.  Rather than tracking recency and frequency and sites visited by user, you track a single number and then you have things that modify the number all the time.  Data storage is much less sophisticated.  The downside is that you are forced to trust the publisher to accurately score users – A “strong performer” in theory could have become a strong performer in a number of ways and the advertiser can never unwind it to see if there are attributes other than “vector score of X” that correlate with performance.

Vectors are great for data exchanges because you can easily vary the vector score to deliver appropriate campaign volume.  If you only have a few people with a vector score of 1000, you can dial it down to 600 and have a 10x bigger population.  Or something like that.  Even in the above example, it is impossible to tell when they stop being a great target because it is all relative.

Vectors today are used to gauge all kinds of things, even things like gender – “This person visits sports sites every day, ipso facto, they are male.”

Unfortunately, all of this ease of use is hamstrung by the fact that vectors obfuscate, even for the person generating the data, how someone became what they are.  In an effort to create a compact data structure that can return real time bids in 250 milliseconds, we have decided to limit ourselves to looking at the forest instead of trees.

Now, today, there isn’t much to really complain about regarding this vector strategy or simple bucketing.  Few, if any, advertisers are prepared to consume every individual behavioral data point and regress on it to determine optimal converting behavioral attributes.  Because advertisers aren’t prepared, publishers are using vectors and not really storing the data in a way they could give to advertisers.

When I think about the future though, I expect someone, somewhere, is going to want this stuff.  Thar’s is gold in them thar hills!

http://www.androidjunkies.com/index.php/2010/03/29/google-deny-app-ad-revenue-sharing/

Recently, Google went public and announced what their revenue share was for people participating in Google Adsense for Content: 68%.

I have to tell you, I have rarely been so disappointed.  When my friends and I talked about the magic of Google’s black box (“They get publishers to sign up and they don’t even tell them how much money they will make!”), we always thought of hundreds of ways that you could micro-optimize.  Ways we assumed Google was doing, because why not tell everyone if it doesn’t matter.

We assumed that Google was tracking the quality of publisher inventory using some sort of aggregated back-end performance (CPA-ish) algorithm.  We assumed they identified clicks that performed poorly.  We assumed that they took all this information about what sites had great performing clicks and what sites had poor performing clicks and then fine-tuned rev share, giving higher payouts to better quality inventory while punishing lower quality inventory.  This would then theoretically have the effect of subsidizing good inventory, resulting in abnormally high payouts, keeping the best inventory in the network, while pushing bad inventory out of the network.

What does bad inventory look like in a CPC campaign?  There are no shortage of gaming sites where you play a mouse-based game that requires you click all over the screen quickly, then they wrap ads all the way around the game.  The result is a lot of bad clicks.  On the one hand, you generate a lot of revenue charging CPCs on this inventory.  (You can typically generate CTRs of 3%+) On the other, none of these clicks result in any time actually on site or resulting in backend conversions.  We always assumed that Google punished these people algorithmically.  We even knew how we would do it.  The algorithms are easy to figure out.  All you need is to not be locked into a rev share with a publisher, but instead have the flexibility to vary a publishers rev share at any time.  Alas, only one company had the clout to do that: Google.

To hear Google’s mea culpa, that they had a straight rev share that they have applied for the last several years without any of this payout optimization reminds you that, while it pays to be paranoid, they are rarely actually after you!

When I first joined Advertising.com, one of the calculations that led me to join was that it seemed like the market was poised for explosive growth, far above and beyond expectations.  Here was my off-the-cuff theory:

• Online was 5% of US media spend
• Online was 17% of US media consumption
• Until those two numbers are the same, people are getting better value advertising online than anywhere else (arbitrage), so those two numbers will gradually come into alignment.

That made me think that the market would grow 20%+ YoY for several years, because I didn’t think that 17% would get any smaller.

Lo and behold, I was right about that, at least.  Today the numbers are headed in the right direction:

• 12% of US media spend is online
• 30% of US media consumption is online
• So, far from coming together, these two numbers, while increasing rapidly, are actually diverging.

My new theory as to why this is: Advertising online sucks.  It simply doesn’t work as well as advertising on TV.  So TV is getting a disproportionate amount of advertising dollars relative to its media consumption because it works better than online advertising.  And that is not too surprising.  728x90s suck.  They don’t tell a story.  They are like tiny magazine ads.

TV ads have a story arc.  They have punch lines.  They are visually stimulating.

Targeting and measurability and all the stuff that most people in online advertising do is great.  They create a situation where advertising theoretically could work better by being more personalized.  Flight prices to Vegas from your city are not what we are talking about, though.  Without a fundamental change in the kinds of creatives that big brands can put out there, online ads simply won’t work as well, at least in a subjective sense for advertisers.

Solving this problem is hard.

First, we must overcome prohibitive expense.  People already spend way more outside of buying media when prepping an online campaign then they ever did for TV.  You build some web site with some viral thing on it, and all that stuff.  That is great.  If a consumer wants to engage with your brand, you want that brand experience to be awesome.  But the first step in engagement is seeing an awesome ad.  If you asked the top X agencies today if they would prefer a consumers first interaction with a brand be via a 160×600 or a 30 second spot, I think you would struggle to get anyone to say the web banner.  Making these great ads is expensive though and someone has to do it.

Second, there must be a visionary industry breakthrough.  Even if we knew how to do it all affordably, I don’t know that we would crack the nut anyway.  What do awesome ads look like?  Punch the monkey captured the imagination.  I just named a 728×90 ad that people actually recognize (although I don’t know what they were selling).  Dancing Lower My Bills ads?  These caught the eye but in a way that left me angry and bitter.  I don’t think Teracent or Tumri is solving this problem.  Making the ad green or blue or blue-green may increase the odds that I click, but it doesn’t vault the ad into the pantheon of great advertising moments.  At least not mine.

Everything that I see in the market is evolutionary and incremental, but this is not the time for incremental.  We are in the first inning.  You are not standing on the shoulders of giants, you are standing on a speck of dust.  There is so much room for change that we need people to swing for the fences.

Let this be a call to entrepreneurs.  If you figure out how to make great ads online, you get a billion dollars.

I think it is safe to say that everyone who reads Cogblog is a believer in Commerce and/or a believer in Advertising.  Now you can represent for your beliefs by buying ad space on the mighty Cogblog.  Every weekly or monthly sponsor receives a Cogmug, the official coffee cup of Cogmap.

Let me know if you have any trouble.

A couple of things about RTB that everyone already knew, but I wanted to make sure got documented by someone:

1. RTB has dramatically skewed the balance of technological sophistication required by buyers and sellers of ad inventory.  It used to be that ad buyers only had to operate on impressions they bought.  Now they need to look at impressions and decide what they are worth prior to buying.  And that decision needs to be made in less than 150 milliseconds.  Ad buying technology is changing quickly.
2. The ad sellers world has not changed nearly as much.  Most ad sellers had a yield optimizer before that algorithmically guessed about how the daisy chain should be arranged.  Not they simply RTB the impression.  The technology used by the yield optimizer has actually gotten much simpler – there is no longer any optimization.  You simply take the impression and RTB it.  There is no calculation, no learning, no consideration of past performance.  Easy.
3. Ad Buyers need algorithms and technology to bid in this new world.  One of the interesting ways that I always characterized AdLearn, Advertising.com’s algorithm (and this is painting in somewhat inaccurate, extremely broad strokes, so don’t think you are finding anything in here or that I am giving you some interesting information.  This is not accurate.) was that it was an algorithm that worked well in situations with poor context.  Unlike Google, which scraped the page and attempted to show contextually relevant ads, AdLearn started with no assumptions and generated learnings.  What this means is that some inventory, from an algorithmic persepective, was more likely to yield good performance than other inventory for different networks. My personal prediction was that Ad.com’s predictions would tend to work better on MySpace than Google’s algorithm.  Your algorithm has an inventory sweet spot also.  Know what it is.
4. RTB could, more accurately, be called “Real-time cookie inspection”.  I don’t think people are actually changing their bids in real-time.  Formulating a good bid takes time.  Frankly, in our current environment, most of this formulation is called “Media Planning”.  A media planner deduces that we should pay $8 for people with this data attribute at a given frequency.  Blamo, bid calculation complete.  After this, we RTB to look for two things: The frequency cap value  and other data attributes we are tracking.  Bidding: nil.  Cookie inspection: Potentially billions of times per day.

God forbid I am mischaracterizing the world, if you are a company that is changing your bids more than once a minute per campaign and the change is more than simply random variation to test bid strength, let me know!  I would love to hear your story.

Every time I talk to people about this transaction, I have the same talking points.  I figured I should put them out there:

1. Invite Media is written virtually entirely in Python, the language of Google-ness.  I would expect to see a more seamless transition to Google-ey-ness than we did with JotSpot, Feedburner, or DoubleClick, where things disappeared (for, in some cases, years) while they were being re-written in Python – Google’s preferred mode of acquisition integration.  That probably allows Google to get a little more excited than they might otherwise.
2. If you are Invite Media and Google calls and tells you they are acquiring a DSP, that DSP has to be you, right?  Google’s acquisition of Invite has taken the biggest acquirer out of the market, dropping the valuation of all the DSPs.  You do not want to be the one that does not get bought.  This probably means that Invite was open to a lower price since it was the best possible acquirer.
3. I imagine a world where Invite technology is suddenly free.  Most people thought that Invite had the best UI in the market and most people agree that the UI for interacting directly with the exchanges sucks.  Given the 8-figure price tag (cheaper than many other deals they have done), it is easy to imagine that Google’s ROI is off the charts here.  They can probably make all that money back by driving more volume through the DoubleClick exchange front-running impressions and increasing volume via a great free UI.

The DSP landscape will be rocked if Google suddenly announces that they are giving away Invite technology to agencies.  People better be bringing their A-game.

Congrats to Nat and Zach.

True or False: Current usage of view-through conversions is not a success metric customers benefit from, but rather a pricing mechanism by which agencies are able to imply a correlation with performance while networks are able to use their reach to spend large budgets, pleasing all parties except the advertiser.

Discuss.

“Today, in response to feedback from many of you who run branding campaigns, we’re announcing a new filter that allows you to show your ads only on AdSense sites among the 1000 largest on the web, as defined by DoubleClick Ad Planner. This new feature will ensure that your ads reach a large number of users, but only on well-known sites best suited for branding goals.”

- Google Inside Adwords blog post this week

Yeah, Ad Networks sell the same way all the time.  comScore top 100, top 500, top 1000.  It is kind of amazing that Google has not released this feature previously, but I am sure that what held them up was that the Product Manager responsible kept de-prioritizing this because it is, if not “evil”, a kind of silly feature that generally serves to mislead people that are not paying particularly close attention.

Perpetually, this is positioned by the sales force as “branding inventory”.  It is implied that the largest web sites in the world are generating inventory ideally suited for large brands to advertise on.  That is just plain silly.

The amount of page views a web site generates has absolutely nothing to do with the relative brand safety or value of brand association that a web site has.  Let’s spend 30 seconds looking at this list of top 1000 web sites:

• Facebook
• Blogspot
• MySpace
• Photobucket
• Orkut
• WordPress
• Blogger
• Partypoker
• Megauploader
• Imageshack

Those are just some of the non-Chinese sites in the top 60 (so maybe 1/2 of the english language content available) and let me tell you:

• These sites are not just predominantly UGC content, they feature tons of NSFW content.  (Unlike, say, LinkedIn, which, while on the list and primarily UGC, is mostly brand-safe.)
• Further, the inventory that Google is getting from some of these guys (e.g. MySpace) will definitely tend toward the distinctly less brand-safe and NSFW.  MySpace is probably not sending their best impressions to Google.

Popularity of the web site is not even loosely correlated with brand safety.  The strategy for this stuff is simple.  For most ad networks (and this is probably still true for Google, although maybe slightly less so), most of their ad inventory came from these sites anyways.  After all, these sites account for the majority of impressions on the Internet.

This “Top X” assurance comforts media planners in some way while only cutting out 10% – 30% of the inventory the network had for delivery.  The result is plenty of room to optimize for the network and the deal is closed, so a victory all the way around.  Buying “Top X” inventory is a way to pay a premium for the inventory you were probably going to get anyway without significant brand protections for advertisers.