Best practices for better conversion lift -

Most marketers want to answer one key question: Did my ads actually cause new-customer behavior, or would the people who saw my ads have converted anyway?

Even the most advanced attribution models don’t resolve this question. Sure, they can help you understand how many of the people you reached went on to make a conversion. This can lead to helpful insights about your audience, including where to find them and how they are interact with your brand. But at the end of the day, many marketers use last-touch attribution, which only shows a correlation between an ad exposure and a conversion; it doesn’t measure causation or incremental lift. In other words, it doesn’t tell you if seeing your ad actually caused a person to convert.

This is where Conversion Lift comes in. By using Conversion Lift experiments to measure campaigns on The Trade Desk, you can understand how your ads are driving incremental results. You can answer questions like:

What is the impact of my media spend on The Trade Desk in driving incremental conversions?
Which of these three Connected TV (CTV) creative assets drove the most incremental conversions?
What audience strategy drove the most incremental conversions?

How does Conversion Lift work?

Conversion Lift uses test and control groups to compare the behavior of people who were exposed to your ads to other people from your target audience who were not exposed to your ads.

We’re essentially running an experiment to see if the variable of serving an ad to someone causes them to take a specific action — in this case, engaging with your brand in some way, whether that’s going to your website, visiting a store, or buying a product.

With some advertising platforms, lift experiments require a certain level of investment in public service announcement (PSA) ads. This is not the case with our solution. We use a process sometimes known as ghost bidding. Once a user is eligible to be bid on by a client’s ad media, they are randomly assigned to either a test or control group. If a user is assigned to the test group, we place a bid to show the user the ad; otherwise, we mark that we would have bid and they’re put into the control group (aka holdout). This means you don’t have to set aside any of your media budget for PSA ads, as there’s no additional cost to run Conversion Lift.

So how do we run Conversion Lift experiments on The Trade Desk?

1. First, you need to define the conversion event you want to measure for the campaign. Ideally the tracking tags are set up and collect data for at least two weeks (ideally a full month) before the experiment begins.

2. You work with your rep at The Trade Desk to set up the campaign, define audience targeting, and adjust the relevant cross-device settings.

3. Our platform then creates randomized statistically sound test and control groups based on your target audience, using our cross-device graph to ensure that all devices for a single user are assigned to the same group. We bid on the test group only, and the control group comprises a similar set of users we would have bid on but don’t.

4. We track conversions and compare the differences in conversion behavior between the test and control groups (which are essentially the same) to analyze, understand, and report on incremental lift in conversion events driven by ad exposure.

The next lift phase: single-cell vs. multicell experiments

The standard Conversion Lift experiment is called single cell and it helps you understand if your campaign spend is working to drive incremental conversions.

But things really start getting interesting when you consider multicell experiments. These give you the ability to conduct true hypothesis testing in our platform, with experiments that aim to answer the question “What is the best way for my campaign to drive incremental conversions?”

With multicell experiments, you can test a variety of variables:

Creatives. Compare different creative variations (such as alternate copy or visuals).
Media channels. Compare different channels and media-mix strategies (such as video versus display versus both).
Frequency. Test multiple frequency caps to uncover optimal ad exposure.
Recency windows. Discover the best way to retarget by comparing recency windows

Conversion Lift best practices

While Conversion Lift may sound like the perfect solution, it does not always make sense for every advertiser or campaign. For example, if you’re running a mass-reach campaign outside of The Trade Desk, let’s say on linear TV, then it’s likely that the control group will be contaminated due to ad exposure from non-The Trade Desk media. In this case, Conversion Lift may not be the right fit.

Once you’ve determined that it is the right fit, there are some nuances to setting up experiments and interpreting results. Here are some best practices that we recommend:

Feasibility:

Channels: Are you running on channels that are supported? Currently we support display, video, CTV, audio, and native.
Conversion types: All online conversions and in-app events are eligible for Conversion Lift experiments. We also support several offline conversion events.

You can measure lift in foot traffic from several location partners, including PlaceIQ, Foursquare (including Factual), and Adsquare.
You can also measure other offline conversion events, such as in-store sales from a weekly data feed, from partners like LiveRamp.

Tracking-tag (pixel) placement: Tracking tags must be set up and collecting data for at least two weeks (ideally a full month) before the experiment begins. This enables us to confirm that test and control users had similar conversion behaviors before the test group was exposed to ads.
Conversion Lift experiments can only run on decisioned media. This is not a good solution for programmatic guaranteed campaigns, since we’re not able to withhold ads at a user level.

Single-cell experiments

If you’re trying to understand the overall impact of an advertiser’s media within our platform (on top of any external media), the experiment should be done at the advertiser level so that the control group remains consistent across all campaigns and is not exposed to any of the advertiser’s media from our platform. This is recommended, especially if you’re using the product for the first time, as it will lead to the highest levels of lift.
If you’re trying to understand the impact of a subset of campaigns/ad groups on top of all other media running on (and outside) our platform, the experiment should contain this subset such that the control group does not receive ads from the chosen campaign but does from all other campaigns running for the advertiser.

Multicell experiments

We recommend testing a maximum of four cells for a single experiment to get statistically significant results without impacting scale.
Holdouts: You can choose anywhere from 5 percent to 50 percent for your holdout group, but we typically recommend around 20 percent. The larger the holdout, the greater the probability you will detect lift (if it exists), but the more it can impact your campaign’s ability to scale.
Cross-device: You should enable the appropriate cross-device targeting settings for all audiences in your experiment to ensure that all devices from a person or household are assigned to the same group (test versus control).

Results:

Results can vary greatly depending on the the brand, campaign, target audience, and conversion events

being tested. This means we cannot provide benchmarks for Conversion Lift results.
Generally, we recommend taking an iterative test-and-learn approach, trying out different variables to gather actionable insights.

Whether you’re running single- or multicell experiments, we always recommend implementing a learning agenda by thinking through the hypotheses you want to test with Conversion Lift. Testing different variables can help you understand how your media investment is driving incremental conversions so you can try to achieve more efficient performance and make better investment decisions.

Measure