Adobe Target Experts on Target Best Practices

Auto-Target Reporting Issue - Where are my Important Attributes and Automated Segment Reports

Hello,

Looking for some support/insight from anyone that has utilized the Auto-Target feature within an A/B test. My test has been running for quite some time and I've reached the necessarily visitors and conversions levels needed for my algorithms to begin learning, 1,000 visits and 50 conversions (7,000 overall visits and 350 overall conversions) as I understand them. However, my reporting is still stating that the models have not been built yet - as indicated by the clock icon still next to my experiences. I'm also not seeing the Important Attributes report nor the Automated segment reports that are supposed to come along with this tool....presumably because the models are still being indicated as not finalize yet? I am using a Conversion goal, 'clicked an element' for our goal metric. 

This test has been running for over 100 days and it seems to be collecting the right information, but no models and no reports yet, any ideas? Are the experiences just too similar for the AI to pick up anything and create a model from? Appreciate any thoughts or comments!

@madisondicken You may want to try setting up the activity, depending on the use case, leveraging Auto-Target which is built into the A/B testing workflow, and uses the same Random Forest Ensemble set of algorithms as Automated Personalization for 1:1 Personalization and does support A4T, which can report on bounce rate as well as any other metric you wish to track.  Keeping the variants low is optimal for these AI-powered personalization activities but multiple activities can be run concurrently for greater efficiency.

@officialyi Adobe Target uses a Students t-test method which is optimal for multiple variants, especially if those variants perform similarly within a test, but does require waiting until 95% confidence is reached to provide rigorous, accurate results while reducing the incidence of false positives.  You may want to run an A/A test utilizing Auto-Allocate which allows for "peeking" during the course of the activity for relative lift/confidence with accuracy at any point during the activity when testing 2 variants. 

@jlisic Great question!  Target, as a default, will split the test population by 50% based on the experiences you set-up in the first step of the workflow.  One of the benefits of our SPA implementation are extensions that route to a specific location within the SPA to deliver one of the experiences to the population within the test.  One other method of distinguishing or segmenting a population in distinct groups is leveraging profile scripts: How to Create Profile Scripts | Adobe Target

@jj223 Yes, this is possible to track in Target but would need to be captured as a persistent value within the Target profile, when not using A4T/ AA.  This would require a profile script: How to Create Profile Scripts | Adobe Target

Hi @billozinga Can you clarify your point on 'subtractively'? Referring to Target Model API for blocklisting certain attributes, this allows customers to fine-tune the AI models to optimize success metrics specifically on remaining attributes. Are you looking for additional thoughts on how to do this? or perhaps when this approach would work for certain use cases?

@edouardbt22 , both Auto-Target and Automated Personalization work with the audiences and profiles available on Target edges at the time of the delivery. However, in both activity cases models are constantly learning and any new audiences will be taken into consideration as soon as they are available in Target.