Saturday, February 22, 2014

View Through Attribution and Display Measurement

Are view-through conversions bogus or are they the path to measuring the true value of your campaign?
When it comes to measuring the success of your display campaigns, and increasingly content or native campaigns, too often the conversation goes like this:
Agency: “Your campaign received 7,000 conversions last week, for an effective CPA of $20.”
Marketer: “Too good to be true. I’ve told you before, I don’t care about view-throughs, they’re bogus. Looks like the CPA is actually $119, which is too high, so we’re redeploying our cash.”
What’s wrong with this conversation? Agencies and publishers have an incentive to lay claim to as many conversions as possible in an effort to show the most value for their platform and inventory. Advertisers and marketers have to deal with stringent post-click attributions models used by their analytics teams. But somewhere in the middle is where the true number of conversions lies. That’s because simple post-click attribution models don’t account for the richness of the conversion path—and counting all your post-view conversions is far too generous an attribution claim. Getting to the accurate number of actions driven by a campaign is the science of view-through quantification, a poorly understood display attribution method that has big implications—not only for display, but also for native ads and the industry’s burgeoning investment in content marketing.
So how do you quantify a campaign’s view-through contribution? It starts with setting up a holdout group and measuring the lift between the exposed and unexposed group. This is a capability readily available in major ad serving systems, DPS’s, or even your AdWords account. You’ll also need some public service ads (or non-branded ads) in the same ad sizes as your campaign creative. Running this experiment will require you to setup two campaigns, one featuring your branded ads, and the other using public service announcements. After running your campaign, which may take days or even weeks depending on the size of your budget and volume of conversions, organize your data in a table like the one below:

Using the data above, you can calculate the true contribution of your campaign— here are some items you should notice:
  • You can see that about $5K, or 3% of the total campaign budget went to the control group, the segment unexposed to the branded advertisement. Consider this an investment in understanding the percentage of post-view conversions that are in fact real conversions that would not have been realized if this campaign were not to have run.
  • Your experimental group received the bulk of post-click conversions—this is exactly what you should expect. The reason the control group shows 5 post-click conversions is the simple fact that a click doesn’t necessarily mean the customer was influenced by the campaign. In this case some users clicked the public service announcement, went to the public service announcement landing page, and still actually registered as a conversion. This is most common with large advertisers and offers that have a wide uptake rate.
  • You can see that there were 6,303 post-view conversions for the experimental group. But, how many of these are actual conversions that you can attribute to the campaign? Answering this question is the purpose of our experiment. By subtracting the post-view conversion rate of the experimental group (0.013%) from the post-view conversion rate of the control group (0.010%) and dividing by the post-view conversion rate of the experimental group (0.013%), you can get to the percentage of post-view conversions attributable to the campaign. That’s (0.013% - 0.010%) / 0.013% = 21%, which means 21% of your post-view conversions are actually attributable to the campaign, an incremental 1,303 conversion.
  • After running you calculation, you can see that the total number of conversions the campaign drove was 2,564, bringing the campaign’s CPA down from $119 to its real value, $58.54.
So what does all this mean? If you’re only counting post-click activities, you’re underinvesting in the channel, so make sure to use this methodology to determine the true benefit of your campaign. But what’s even more important to remember is that this attribution method unlocks possibilities in content marketing and native ads, channels where measurement can be even less clear and difficult to capture from a purely post-click perspective.


Anonymous said...

Hi Will, how can you isolate audiences when doing a PSA test? In other words how can you ensure that PSA audience didn't see the regular ad?

Anonymous said...

Great read!

How do you get 1303 view-through attributed conversions from the 21% PV attribution percentage?