Forum Discussion
Wanting to test the new AI Intelligent Selection
Our team is wanting to perform an A/B test using the AI Intelligent Selection. We are going to select a basic one step push to test on where we want to run a copy test.
In terms of reporting, i'm just curious to understand how the setup works to know how this will appear in our reporting and wondering if there is any type of braze reporting that comes with using this testing option that we could leverage when reporting out on performance - such as how frequently the audience is re-assessed, how it decides what percentage of audience to change, etc? Any help on this would be appreciated to educate our strategy team. Thank you!
1 Reply
Hey c_hixon
I'm not sure you're going to see the exact data you want about the AI decisioning process but there is some information about the algorithm used in the Braze docs which links out to the 'Multi Armed Bandit Theory' wiki page.
https://www.braze.com/docs/user_guide/brazeai/intelligence/intelligent_selection#about-intelligent-selection
https://en.wikipedia.org/wiki/Multi-armed_bandit
BrazeAI TL;DR for others
Multi-armed bandit theory is the idea of being faced with multiple slot machines at a casino (your canvas paths) but with a limited number of coins to spend (your entered users) and figuring out how you will get the best return. The canvas won't initially know what the best path to send users down will be for optimal results and even when one path seems like it is winning, should all users be sent down this path or was that just luck? Smart algorithms try to reduce loss and maximize returns in this problem by balancing sending as many people as possible down winning paths while still exploring new options for better results. This is the basis of BrazeAI.
