Dynamic Creative Optimization & Multivariate Ad Testing¶

Key Benefits¶

Using DCO, you can now test user-controlled features for your campaigns. That means, you can choose 3 specific attributes for DCO - Title, Description, and Image - and three variants for each of those 3 attributes (a maximum of 27 combinations in all [3x3x3]).

Note that you don’t have to use all 3 attributes included in DCO. For example, you can choose to set 3 options for the Title, 2 options for the Description, and 1 option for the Image. This will make DCO work with 6 ad combinations only (3x2x1).

Once you submit disaggregated creative elements that can be mixed together, DCO systems will test and conclude which creative element combinations produce the best results for you, without having to test every possible combination. Success is evaluated on the basis of CVR.

The DCO algorithm will optimize to the ad combinations with the higher CVR.

Case Study¶

This case study example illustrates how a digital bank was able to optimize their Native ad campaign results and increase their lower-funnel conversion (CVR) by as much as 54%.

Creative Elements¶

Creative elements for DCO optimization may include the following, with variants of all three attributes:

• Title

• Description

• Image

How It Works¶

On Native, DCO works as follows:

1. You can define up to 3 variants for 3 different ad assets: Title, Description, and Image. This leads to 27 (3x3x3) possible ad combinations that the DCO algorithm can use to experiment and optimize for campaign best results.

2. You use Bulk Sheet as your interface for defining those ad assets.

3. In the UI and with downloadable reporting, you can view overall ad-level performance, including all standard traffic and conversion metrics, as shown below.

Campaign Setup¶

For campaign setup, all your assets need to be uploaded through the bulksheet.

The following is a sample definition of DCO Ad Assets in the bulksheet input:

Reporting¶

Reporting is available at these levels:

• Standard ad-level performance reporting by the standard means in the UI.

• Performance reporting for each ad combination to every client without requiring assistance.

How to Add the DCO Level to the Ad Performance Report¶

The following illustrates a sample report file and a screenshot from Ad Platform UI:

Download Sample Ad Performance Reports¶

To download a sample .csv report file, click here.

The ad report created above:

Some Best Practices¶

The following are some best practices for optimal DCO ad definition when working with ad assets:

1. Make the assets diverse. This is your opportunity to test diverse images, titles, and descriptions. Too similar assets will not generate enough difference in the CVR and will not gain from DCO.

2. A higher budget is helpful to generate enough traffic for the model to identify the better-performing combinations. Too low of a budget will slow down the optimization process.

3. Once the ad assets are introduced into the ad, the corresponding ad-level properties are ignored. In the UI you’ll see one of the ad combinations.

4. The image ad asset requires URLs of existing images. If the image is available only locally, you can create a paused ad in your account with the image and reference it in the links in the bulksheet.

5. If you want to update one of the ad assets, you can do so via the bulksheet by first deleting the existing one and then adding the updated one. To do so, issue a DELETE operation in the bulksheet on the ad asset ID to be removed and then issue an ADD operation with the correct ad asset.

6. Splitting the budget between several creatives is quite common among advertisers. However, to expedite DCO exploration, it is best to concentrate all the budget on a single DCO ad.

7. Per Native Ad Platform image requirements, ensure that you use one regular, HQ, and Thumbnail images for each of the image asset versions.

8. DCO is valuable for ads that generate conversions (regardless of the campaign bid strategy). If the ad is not tracking or generating conversions, all combinations will stay with equal chances to be presented.

Caveats & Limitations¶

• DCO does not cover non-ad specific ad attributes that may be optimized behind the scenes by ad serving or the property. For example, the ad should be shown fully expanded or collapsed for a different section.

• Some ad formats don’t lend themselves to DCO, like video, cinemagraphs, possibly mobile moments, slideshow ads, touchpoint ads, and so on. Other ads may already have some other form of creative optimization that may need to be rationalized further with DCO – for example, Dynamic Product Ads (DPA), Hotel ads, and possibly carousel ads.

• The bulksheet doesn’t currently support an UPDATE of the ad asset. You can achieve that by removing the existing ad asset and adding the updated one. If you wish to update one of the adassets, you can do so via the bulksheet by first deleting the existing ad asset and then adding the updated one. To do so, issue a DELETE operation in the bulksheet on the ad asset ID to be removed and then issue an ADD operation with the correct ad-asset.

• When an ad asset is updated, the optimization process is restarted.

FAQs¶

Q: How are the ads optimized?

A: The algorithm optimizes to the ad combinations based on higher CVR.

Q: Will each of the 27 combinations be given an equal split of traffic in the beginning?

A: The 27 initial combinations will get initially an equal split across all traffic. That means, each gets the same chance of being served.

Q: What’s the difference between DCO and creative testing with even ad rotation?

A: DCO is the right way to achieve it. Creative testing by even ad rotation has a few flaws. It has usability issues because the advertiser needs to manage the various ad combinations. It has scalability issues because growing the number of ads in the campaign leads to a bottleneck of ads in the exploration. It also extends the time to mature from exploration because each ad gets a smaller exposure. And lastly, it has correctness issues since the ads in the campaign compete with each other. It means this approach can lead to favoring the ad with the first engagement while in DCO all variations get equal chance before a winner is selected.

Q: What’s the methodology of the optimization?

A: The optimization to the better performing ad combinations is done by a post-auction approach. The DCO ad participates in the standard auction model. In case it wins the auction, the serving system draws the presented combination. The probability to draw each combination reflects the user engagement level where high-CVR combinations are assigned with higher probabilities and vice versa. The process starts with a uniform distribution and as more information is gathered the probability is tuned by the system and concentrated over the “best” combinations. This optimization process is a continuous process. This optimization does not start before the ad combinations reach a minimum of impressions and clicks.