Promi will optimize discounts within the campaign settings and guardrails you set. However, if guardrails are not set appropriately, or the test design is not appropriate, you may see lower campaign performance. See below a checklist of suggestions to make sure you're able to get the most out of your campaigns.

Test Design Best Practices

1. Provide adequate discount range for the AI model: We typically recommend providing a discount range of at least 10% (e.g. min discount of 5%, max discount of 15%) for Promi to optimize within. This can be set in the "Min Discount" and "Max Discount" fields during campaign creation. A lower range will constrict our model's ability to incentivize shoppers unlikely to purchase otherwise, while earning back margin from shoppers who plan to purchase anyway.

   
   

2. Set an appropriate fixed discount for the control group: Depending on your goals for the campaign, you may want to choose different control group discounts:

   • 0% discount: best to understand how adding Promi discounts to your current setup impacts your store. We typically recommend this to merchants who are using Promi to increase conversion rates, not necessarily to replace existing discounts.

   • Midpoint of your AI discount range: best for generating additional revenue from replacing existing discounts. Setting the control group discount in the middle of your AI discount range is our most common recommendation for merchants. This allows Promi to generate more margin from certain visitors, while generating more revenue by reallocating that margin in the form of larger discounts to on-the-fence shoppers. You will likely see more revenue from Promi with the same or lower discount spend.

   • Maximum of your AI discount range: best for merchants looking to increase profits or reduce overall discount exposure. This setup will allow Promi to generate more profit by charging certain customers more, but you will likely see similar, if not slightly lower overall revenue.

     
     

3. Activate the Promi campaign on appropriate items or collections: For an accurate A/B test, Promi should be active on products with a sufficient number of orders per month. As a rule of thumb, we typically suggest at least 100 cumulative orders per month across all eligible products or collections.

   
   

4. Set appropriate product and visitor exposure rates: Visitor and product exposure settings allow you to limit discounts to a certain share of traffic or items:

   • Visitor exposure: If you are using Promi to add a new discount and optimize your conversion rates, we typically recommend exposing the discount to 50-70% of users. This is the default setting when creating a new Promi campaign. If you are replacing an existing discount, we recommend setting the visitor exposure to 100% of visitors. Note that if you include 0% or $0 as the minimum discount, the number of visitors seeing a discount will in practice be lower than the visitor exposure setting.

   • Product Exposure: If you would like to ensure a discount is available on all products, you will need to make sure the product exposure is set to 100%. Reducing the product exposure will cause Promi to selectively omit products. Products with the highest conversion rate will be omitted from discounts first. The product and visitor exposure settings can be found under "Advanced Settings" in the "AI Discount Parameters" card of the campaign creation form.

     
     

5. Round discounts to the right value: You should round the discount value based on the range you provide to the AI model. If you are rounding to the nearest 5%, for example, your discount range should be multiples of 5 (e.g. 0% to 15%). This setting is available in the "Advanced Settings" section of the "AI Discount Parameters" card in the campaign creation form.

   
   

6. Preview the UX configuration for your campaign: Make sure you have configured appropriate ways to communicate the discount to your customers. Before launching the campaign, we recommend using Promi's preview link to view how the configuration will show up to your customers. This link is available at the top of the "UX" tab in the campaign creation form. Note that this preview link will block other discounts from showing up. It will also use an example 10% discount, be eligible on all items, and not appear at checkout.

Best Practices During an Experiment

1. Avoid making significant changes to the campaign: Changes to the list of eligible products, discount range, etc. all can have an impact on the performance of the campaign. Our performance dashboard looks at the impact over the full range of active experiment dates. If, for example, you are looking to understand the impact of a 0-10% discount and change it to 5-15% during the experiment, the A/B results will blend the impact of both discount ranges and you won't have as clear a picture of the performance of either individually.

   
   

2. Allow experiments time to run: We recommend running experiments for 3 weeks, or as long as it takes to see statistically significant results, whichever period is shorter. During the first few days of a campaign, you will likely see large variations in performance. This is due to limited sample size. After a few weeks, performance should become more stable. You can use the "Statistical Significance" metric in the Promi dashboard for a sense of the validity and reliability of a result. 95% and greater is typically considered significant.

   
   

3. Monitor Results: Promi's performance dashboard provides updated data daily. If something does not look right with the performance, check your campaign setup.

   
   

4. Do not pause experiments prematurely: Pausing a campaign will cause the experiment to end. To restart the experiment, you would need to create a new campaign. Resuming a campaign after it has been paused will scale up the treatment group to 100% of traffic.