Optimizing Dynamic Creative for Targeted Percentage

June 22, 2018

Dynamic creative presents itself as a simple idea: If a user views a product page they will later see an ad with that same product and some similar products. Despite this, building and trafficking dynamic creative is much more complicated than it seems. The initial challenges are usually taken on by your vendor. Building solid logic to determine how the dynamic creative will function (if a user views a kitchen table product page, do they want to see similar kitchen tables, or rather the matching set of chairs?). Designing the creative to match the advertiser’s brand, and scraping the site for consistent, high quality products photos, descriptions, and prices can also lead to challenges. But even if the initial set up is flawless, trafficking the creative is a whole other beast. I took one of my campaigns and rebuilt it to optimize for targeted percentage — the percent of ads where a user actually saw dynamic creative (and not fallback or default, which we’ll review).

Despite constant problems with the logic and scraping of data, I was able to triple the targeted percentage from 9% to 31% and save $700 per month in fees by reducing the number of untargeted impressions. Those numbers are still poor, but better than they were previously.

Default: Default is effectively static creative that would be served in a non-dynamic campaign. It signifies the dynamic creative was served to a device that had no data in our dynamic creative vendor’s system (the vendor’s pixel is separate from our exchange’s pixels and systems).

Fallback: Fallback creative is identical to default. The difference is the device the ad was served on did have data in the vendor’s system, but not enough to serve dynamic creative. Usually this would happen if the user visited the homepage of the site, but no product pages. So the system recognized the user, but there was no targeting data.

Targeted: Targeted impressions occur when a user actually sees dynamic creative. For this to occur the creative must serve to a device that matched a device in our vendor’s system, and the device had enough data in that system to target (A list of product pages viewed, products added to cart, etc.)

Optimizations

The goal of the optimization in mid-May was to align the audience I served to with the audience our vendor had. Since our retargeting pixel that gathers the audience is separate from the vendor’s pixel, I created a new audience and a new vendor pixel to get a fresh start. I placed them on the client’s site only on product pages. This way every user in my audience should have enough attachable data to send an actual dynamic ad.

In the trading platform I built a new ad group specifically for this audience. Critically, I turned off cross-device targeting and household extension targeting. Both of these massively reduce the targeted percentage. These settings are built to extend an audience beyond the user’s initial device into the other devices they own and the other members of their household. Unfortunately, our dynamic creative vendor does not has a device graph and can’t see user’s other devices, or other members of their household. So if those two options were enabled I would be served dynamic creative to devices that had never visited the website, and therefore would not have any data in our vendor’s systems and be served default ads.

Results

This strategy was obviously successful — the default rate plummeted from 55% to 5%. The targeted percentage rose from 9% to 31%. Unfortunately the fallback rate increased massively. The cause is that we did increase the number of targeted impressions, reduced the number of defaults, but did not affect the fallback numbers at all. This means 90% devices are being identified in the vendor’s system, but 60% of the time there isn’t enough data to send them targeted ads. Unfortunately this is due to the initial campaign build and requirements from the client that resulted in flawed system from the start. If this were to be set up on my own website with better data, the targeted percentage would have been closer to 90%!

Ultimately the goal here is to align the dynamic creative vendor’s data with the audience data in the serving platform. Regardless of the systems and vendors you use, working towards that goal will result in more dynamic creative being served, less wasted fees, and better performance.

Edit: Commenters have pointed out that you lose a significant portion of your audience when turning off cross-device and household extension. Depending on the platform you’re using, you might be able to create a second ad group that uses static creative to targeted only these audiences, and potentially exclude the original audience.