CLient Login
Jan 08 2024

Scientific A/B Testing for Products Feeds and PLAs

By Allie Bates

How often do A/B tests run only to yield inconclusive data? Especially in the data feed and PLA world, the problem is prevalent. VersaFeed's new Dynamic Experiment solution solves this problem once and for all.

Our dynamic system will rotate experiments on/off randomly, through the entire test period, thus ensuring that there's no outside influences.

Here's the process:

  1. You choose ONE set of products that will contain both your test group AND your control group (e.g., all shoes with ROAS < 5). Ideally, this singular group should be a few thousand products and not lower than one hundred products.

  2. You create your test hypothesis data for the entire set of products. E.g., if you have a new title you want to test, you'd make that new title for every single item in your product set from (1) above.

  3. You (or your VersaFeed rep) configure your account with a new Dynamic Experiment in the VersaFeed dashboard. Don't worry, it's painless.

  4. Set it and run. Throughout the experiment, VersaFeed will dynamically turn on/off the experiment at the product level – in statistically sound combinations that ensure every product had some combination or both test and control. Custom Label assignments (e.g., "Test" or "Control") can be applied to corresponding/applicable products in tandem with each update to help track results.

Why is this magical?

Dynamic Experiments are essential for accurately measuring KPI shifts, such as ROAS, resulting from subtle feed optimizations—especially when incorporating AI. Reach out to your VersaFeed account manager to activate this industry-first feature!