Podscribe’s AI Optimize is in beta, as of Summer 2026, but early results indicate a 2-4x average performance improvement over baseline.
You can enable Podscribe’s AI Optimize within your Targeting Profile:
Once enabled on a campaign, the AI optimize will serve ads when it predicts the listener & context are most likely to generate a conversion.
If the advertiser has no conversion events, then it will fallback, in order, to optimizing for leads, signups, installs, and then visitors. If you want to optimize towards a custom event, please email smartserve@podscribe.com and likely we can support it.
How it works
From an advertiser’s measurement data, Podscribe learn where a brand’s ads are likely to perform best. From past attribution, it builds a machine learning model to predict which impressions best drive conversion, and then which factor(s) are the most predictive. The following factors may be used for optimization:
- listener geo
- listener app & device
- listener frequency in this campaign and across all of the brand’s campaigns
- listener demographics
- content of ad
- ad placement, eg pre/mid/post
- prior listener engagement with the brand
- time of day & day of week
Scale vs. Performance
There is always a tradeoff of scale vs. performance. Based on the expected impressions inputted for a campaign and the timeframe (start and end date), AI optimize will dynamically adjust the performance threshold to hit the estimated scale.
This means that one should pay careful attention to estimated impressions, as well as start and end date inputted for SmartServe AI optimized campaigns.
Are these conversions incremental?
Like with any other SmartServe campaign, you can configure either a ghost holdout group, or a PSA ad to be served to measure how incremental these conversions are. Both these would be set up just as they are set up in other SmartServe campaigns.
Can we see the learnings, ie what factors the model is using?
Yes! Currently we can manually email over a recap report after the campaign, but soon we’ll also display in the UI the ‘most predictive factors’ the model used. Can it be given guard rails? What are “targeting bounds”?
Yes, you can enable Targeting Bounds to add some bounds to the AI targeting, such as frequency. Generally though we do not recommend using targeting bounds, because if the AI optimize works as we hope, it should automatically filter out any impressions that don’t perform well.
One use case of targeting bounds may be though to avoid targeting listeners reached in other campaigns, such as parallel control campaigns being compared to the AI optimize campaigns.
Is there a cost?
Yes, a $3.00 add-on CPM. This feature is in currently in beta, so please email partnerships@podscribe.com if interested in participating in a test.
When should I use AI Optimize vs Re-Targeting vs Lookalikes?
In theory, if Podscribe has done its job, AI Optimize should always perform at least as well as any other targeting option.
However, there are cases where it makes sense to retain more control over targeting, e.g. if you want to run a direct A/B test of different audience strategies, target a specific customer segment, or validate performance against an existing re-targeting or lookalike audience. In those situations, using Re-Targeting or Lookalikes can provide a useful baseline or satisfy specific campaign objectives, even if AI Optimize is expected to deliver the strongest overall performance.
Powered by