Gaming The Attribution System

By
  • Facebook
  • Google Plus
  • Twitter
  • LinkedIn

sephzdarko“Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by Seph Zdarko, head of attribution initiatives and partner strategy at Quantcast.

Attribution models were designed with the best of intentions: to help advertisers understand which ads lead their users to convert and to inform which tactics effectively grow their business. While advertisers work hard to use attribution properly, they often unknowingly incentivize the wrong behavior from vendors, leading to ineffective spending of their budgets.

It’s a dirty secret of today’s digital ecosystem. I call it “attribution gaming.”

The main reason gaming is so prevalent in display advertising today is the widespread use of last-touch attribution, a model in which 100% of the conversion credit goes to the last ad served before a conversion. With today’s cluttered ad ecosystem, the problem with the simplistic last-touch model is that it is easily gamed and lends itself to manipulation.

One of the most common ways to game last-touch attribution is to allocate the majority of an advertiser’s budget to retargeting consumers who are already close to a conversion. Computer algorithms use predictive modeling on nuanced signals to know when someone is close to converting. These signals include visiting an advertiser’s site multiple times or placing items in a checkout cart. Here, attribution gaming isn’t about influencing a consumer further down the marketing funnel. It’s about capturing the last-touch credit for a conversion that is likely going to happen anyway.

Other dirty secrets about gaming: The ad on the bottom of a page, which is typically the cheapest, has the highest value from an attributional perspective. That’s because the last ad on the page is the last to load and will get the last view credit.

An ad also doesn’t have to be seen to count. An ad can be served in a hidden iframe, in an interstitial or hidden below the fold, 15 ads deep. Quality isn’t nearly as important as quantity when it comes to attribution gaming.

Finally, when a consumer is in the checkout process and searches for a promotional offer or coupon, or checks social media before clicking “purchase,” the ad served at that moment gets all the credit for the conversion.

Persistent Tactics

Even with the addition of more sophisticated multitouch attribution models, many of the same gaming tactics still work. Since most solutions today still do not distinguish between a retargeted impression and a prospected impression, it continues to be all about serving more retargeting ads than the competition. In this way, multitouch attribution models heavily incentivize retargeting, since those impressions continue to make up the majority of ads served.

You might think that a first touch is an upper-funnel prospected ad, but what if the first touch is after a consumer has already been to your website? By definition, this is a retargeting impression, so even a first touch can be gamed. Unique reach is another easily gamed metric. Gamers just target specific browser configurations or cookies with no data on them, giving the appearance of a large unique reach while keeping costs low.

Gaming 101

Here’s how a simple gaming strategy might work:

Buy a small amount of cheap ads for new cookies across exchanges to capture new and unique reach rates. Then don’t do any further prospecting. Buy as much retargeting as possible to maximize conversion credit.

Next buy the first ad immediately after a user visits the advertiser’s site for the first time to capture more of the first-touch credit.

Also buy lots of cheap retargeting ads – even below-the-fold or out-of-view ads are fine – but bid more and increase frequency as consumers revisit the site and move closer to a conversion.

Once users start the conversion process, capture the last-touch credit by aggressively targeting them on every inventory source, figuring that many will check email, search for a coupon or check social media right before clicking “purchase.”

Now relax and wait for your client to give you more budget.

Try It Yourself

If your attribution solution only cares about last touch and the lower funnel, then that is where your targeting will be focused.

To see an example of unsophisticated retargeting, open a new browser session in privacy or incognito mode. Then visit an advertiser’s site. Take a step toward a conversion, but don’t convert. Then go to an exchange-traded site with multiple ad slots per page and see how many retargeting ads appear. Click the “ad choices” button on each to uncover who is serving them.

Stop Getting Gamed

While we may never be able to fully eliminate gaming tactics, we can try to expose cheaters and increase transparency for everyone. You can start by splitting marketing funnels into two separate phases – prospecting and retargeting – by adding the “first site visit” as an additional point of measurement. This simple action cannot be gamed and is a major step toward preventing unwanted behavior.

Use MRC-accredited viewability standards to significantly reduce the false signaling noise in all attribution models. And get a second opinion by bringing in another solution or implementing an A/B test to look at the marketing funnel from more than one perspective to help eliminate hidden gaming.

Not everyone games attribution, but, unfortunately, many do. Through increased communication and transparency, we can help eliminate this shady underbelly of online advertising and improve advertising efficiencies in measurement and performance for everyone.

Follow Quantcast (@Quantcast) and AdExchanger (@adexchanger) on Twitter.

  • Facebook
  • Google Plus
  • Twitter
  • LinkedIn

Email This Post Email This Post

By on at

8 Responses to “Gaming The Attribution System”


  1. Rob Leathern says:

    Unfortunately the fundamental problem with almost everyone in the sector is that they're compensated based on some percentage of the amount of media spent. And of course almost all attempts to compensate based on "results" runs into the same attribution issues you correctly point out, Seph. It's in everyone's interests unfortunately to get the client to spend more money, and until that changes (e.g. enterprise platform fees regardless of spend level) this will remain a problem.

  2. John Were says:

    Agree with that Rob. We need time or project based fees with little or no exposure to media budgets on the service side as well. Overnight, incentives are realigned to work in the advertiser's best interests and demonstrate real value in order to continue the relationship.

  3. Ronald says:

    As someone on the client side, re-building a $1M/month digital advertising program from the bottom-up this article hits home. After working with 5 or 6 major names in the industry, this is what I've found. It seems like if you just leave DSPs to their own devices, they are going to ask for 100% attribution on all retargeting conversions at a 365-day window. They always try to compare retargeting CPAs to prospecting CPAs. Typical conversation: "yeah that prospecting CPA is high and we're working on it, but look, retargeting is working great! look at that ROAS!". You ask them to do a PSA lift test- and they push back. You ask for funnel analysis- and they push back. It's like you have to wring them by their necks and threaten to cancel I/Os before they try to help you figure out what the campaigns are actually doing for your business. They want to keep the wool pulled over your eyes. All will try to sell a song and dance about how they are different during the sales process but after you get the first I/O signed, these things inevitably start happen. This is my perception after working with 5 or 6 BIG names in the industry.

    The question I have is, what "stack" is the best chance for success. A central ad-server (so reporting is centralized), and then an attribution vendor (Visiual IQ, Adometry, etc).

    • seph Zdarko says:

      Well put Ronald.
      In an attempt to answer your question... Yes, a central ad server, or DMP, with the addition of a 3rd party attribution vendor (or built in) is great if you can afford it. Sprinkling in some occasional A/B tests and cross channel incremental lift analysis is also great. The key is to actually use it and keep changing it up so that you aren't misled by an unknown blind spot. Unfortunately attribution isn't a set it and forget it proposition. The ecosystem is constantly evolving and changing and thus so does your marketing tactics and attribution measurement.
      The real difficulty in answering your question is that there is no one size fits all here either. Just like individual prescriptions for medicine, exercise and food is different for different people, so are digital advertising prescriptions for businesses and products. Everything is an experiment of one. But that’s why we do what we do right? Constant experimentation is what keeps us on our toes and it’s also what makes our jobs and life so much fun!

  4. Ricardo Mascota says:

    Actually, isn't most of the gaming due to post-view attribution. It seems like ignoring the risky strategy of post-view attribution would clean up much of the problem. At least a post-click system relies on the fact that the ad was visible (you can't click what you can't see) and that the user engaged with the ad. While it's not a perfect solution, the agencies and vendors with post-click offerings would seem to be working in alignment with clients.

    • seph Zdarko says:

      Hi Ricardo. You are correct that many of the gaming tactics occur in view based attribution, but it applies to clicks as well. Don’t think you are safe by just looking at clicks. While click based attribution is easier to track and more tangible to understand than a multi-channel regression model, that doesn't mean that easier is better. Going back to the way things used to be done isn't the answer. 10 years ago last touch, last click, worked because online targeting tactics were fairly simple, but today with the advent of big data and RTB, audience targeting can now slice and dice individual impressions that dance circles around this outdated attribution model. The reality is that clicks are very poor measure for display. Clicks work in search, as the medium is based on click navigation by design, but most people do not click on display ads without a strong call to action. Just ask yourself when you see a display ad if you click on it? I see display ads all the time, so I know they affect my perception, but I never click on them. I use search or direct navigation almost 99% of the time.
      Billboards, television, search, display, affiliate, email, etc… they all work differently and for different reasons. The goal of attribution is to do its best to determine the marginal and causal relationships to each other and ultimately to the influenced consumer. Evolution of our existing measurement is the solution, not reverting back to simple mono-channel tools.
      I apologize if I am being too direct, but this is the whole basis behind the open source movement around Split Funnel Attribution. By evolving the attributional foundation with the simple addition of another point of measurement it vastly improves the transparency and insights across all marketing channels.

      Transparency is like gravity, the industry will get there eventually, it just might take longer than some of us want it to.

      • Chris Smutny says:

        Ricardo, when you look at clicks, it's important to realize that some of those clicks are simply navigational clicks. There are still massive amounts of people that do a Google search for "Amazon" and click on the first result. That first result is a search ad and on a post-click basis, will generate a tremendous ROAS for the advertiser.

        Is that person really "engaged with the ad?" I think that is unlikely. They are really just using it as a navigational tool. A similar issue arises with display clicks, especially with very aggressive retargeting.

    • Julian Chu says:

      The other issue with relying on post-click only is that you effectively limit your traffic/revenue potential, given the extremely low click-through rates on display advertising (I'm talking from a retail industry perspective here).

      I think one of the key points in what Seph is advocating, is not that post-view should be dumped altogether, but that you have to have the right way of measuring the impact of an ad view so you can properly credit that campaign and allocate budgets accordingly.

Leave a Reply