Adobe Target

Greetings @PASTROM ! Welcome to today's Target Community Q&A Coffee Break By posting in the thread, you are participating! This is a text-only thread discussion where one participates by posting  in this thread to today's Coffee Break experts,  @bkostak, @Ryan_Roberts_, and @msamaniego. Please feel free to share your Questions, Best Practices, or Hot Tips related to the webinar topics and our experts will weigh in in response!

Greetings @trahulc ! Welcome to today's Target Community Q&A Coffee Break By posting in the thread, you are participating! This is a text-only thread discussion where one participates by posting  in this thread to today's Coffee Break experts,  @bkostak, @Ryan_Roberts_, and @msamaniego. Please feel free to share your Questions, Best Practices, or Hot Tips related to the webinar topics and our experts will weigh in in response!

What is the link to join this coffee break event?

@pramodalapati How Do I Upload Custom Criteria in Recommendations? | Adobe Target

Hi @pramodalapati, the custom criteria uploads are meant to just be uploading the results of the criteria. So a key and its corresponding recommended items. So a simple CSV type layout is the most simple way to pass this. There is no benefit to being able to nest or group objects in that upload. Hope that helps.

Hi @wozinga, there are couple of options. Sometimes it might be a good option to switch it to an Auto Allocate test which is optimized for speeding up the results of an AB test.

You could also use an AP activity with reporting groups - but that only helps if the offers are getting replace by newer versions.

Offer Reporting Groups in Automated Personalization would be an option for creating personalization models for each group based on the data. Reporting Groups helps solve by grouping similar offers with each model getting more data to train on. 

ExL link for more details!

@ande2 Great question!  This was covered in a separate thread on the Community - it's setting an exclusion from the AA side: Any ways to automatically remove outlier data from... - Adobe Experience League Community - 265170

Thanks @Drburns27. We were trying to use the same approach suggested in the other thread. My question is - how do we determine the reasonable range to create the segment? I know it will differ for different companies and different success metrics. Is there any best practice we should follow there? It'll be great to hear your recommendations 

@ande2 Great question!  As you mentioned, there is no established standard as it will depend on the data set.  But 2 standard deviations from the mean should probably rule out outliers, especially in a smaller data set.  But again, looking at the data in a graph format usually points to the appropriate deviation from the mean and represents outliers that do not match a majority of the data set.

Thanks @Drburns27 - What's the date range we should check the data for? last 3months/6months/1year - any standards there ?

Hey @ande2, there isn't one right answer to that question, other than "it depends". Generally speaking more data is better. The way Target's built in extreme order filter works is on the data within that activity only. But for your own custom outlier filter, adjusting the threshold you use for extremes on more data would usually be preferable. If the data is reasonably consistent during those different time frames then I'd lean to using more rather than less. If there is seasonality that may naturally affect order sizes you would want to consider that and adjust the window you use to define an extreme order.

@trahulc Auto-Allocate is an excellent option when there are low traffic requirements, as the multi-armed bandit will divert traffic over the course of the test to the winning variant.  This activity is built for multiple experiences as well.  What is an Auto-Allocate Activity? | Adobe Target

Auto-allocate would ideally divert traffic based on the primary success metric which we specify in the test. What if we conclude/analyze a test results based on performance of more than one success metrics? Is there any way Auto-allocate can divert traffic based on multiple success metrics ?

Agree but the problem with that approach is we look at multiple metrics to determine a winner and auto-allocate looks at only 1 primary metric to determine the winning variant. Isn't it? In such a scenario, what do you suggest?

@trahulc This can be done in 3 ways, from my perspective.  Auto-Allocate does have built-in capabilities to show 95% confidence, where you could compare the multiple metrics performance at that point, training the model on the primary metric.  You could do phased auto-allocate activities in succession or with a portion of the population to compare multiple metrics at significance point.  Or you could train the auto-allocate activity based on a compound metric from analytics, selecting that metric as your primary in the activity.

Is it possible to use calculated metric from Analytics for Auto-allocate activities?

@ande2 calculated metrics have been enabled for A4T, such as revenue.

Hi @trahulc,

Another idea might be re-running the AB test for just the small audience (after the applying of the customer attributes). Of course this also sort of starts over your reports. Maybe (warning could be a crazy idea here) after reaching a conclusive result for the broader audience, but not your smaller audience, you could continue the test but exclude the non-smaller audience. In other words the majority of the audience could start seeing their new winner, but you continue the test on the smaller audience by simply adjusting the entry audience condition on the activity.

Correct @JoseD_Q, and welcome! By posting in this thread, you are participating Please feel free to share any of your webinar-related Questions, Best Practices, and Hot Tips in the thread for our experts to weigh in!

Hi @ben_mrm, 

There can be cases where your models build but are valid. This could be because of a few things.

1. The experiences perform very similar to each other in general, too close to statistically confirm one group is performance better on one than the other.
2. The optimizing metric needs 50 conversions per experience so if you have 350 in the activity you probably have 50 in each experience, but worth checking that.
3. Lack of meaningful features/attributes - if you are not passing any custom attributes into Target and it's only using out of the box items it may not be getting good enough "signals" to know when a person would prefer one or the other.

@JyotiSharmaV we have many companies who do this, and you can set how often these audience segments should be updated through customer attributes or bulk profile upload via api.

Hi @JyotiSharmaV,

Are you thinking efficiency from an implementation perspective or a post-implementation using the 3rd party segments perspective?

I am thinking of efficiency in terms of the audience pool and whether using 3rd party tool for the audience would reach correctly to the desired audience group or not.

Are you thinking you'd like to breakdown a report in AA by a visitor's category affinity preference?

In theory you could pass the value in a response token and then client side have Analytics pick it up in an evar.

Yes, this is exactly I am looking for. Great to know that we can pick it up. I will give it a try

@JyotiSharmaV The URL you’re asked to provide is the location of the data source for your feed. This could be a CSV file, Google product Search feed 

How Do I Use Feeds in Target Recommendations? | Adobe Target

That's what I am asking. If I have the csv feed ready with me on my local desktop. Then do I need to upload it somewhere or anything else to provide path?

Hey @JyotiSharmaV,

Yes, you will need to upload the CSV from your laptop to a hosted location with a URL. Target checks that URL for updates to the feed at a schedule of your choosing. That's why you need it to be in a hosted location.

Hi @JyotiSharmaV - great to see you here today in the thread! And as Drew mentioned - great timing for this question! We have two related  Recommendations-focused events upcoming: 
(1) 2/13/24 Webinar on Recommendations (Personalization Maturity Series Part 2) 

(2) 2/28/24 Coffee Break as an extended Q&A for the 2/13 Recommendations Webinar 

Please do mark you calendars, register for both events, and share the information out with colleagues who may be interested in diving deeper into Target Recommendations  

Please be sure to register for both events and share with any of your colleagues who may be interested as well!

@Amelia_Waliany et al., SEO is not impacted with differing experience variations within a test or activity with a limited life span, especially when using our most common client-side implementation, as it's based on the default content.  The experience variations are content that is delivered as a layer above the page.  The only way that SEO would be impacted is with an "always-on" personalization activity leveraging AI where there isn't a fixed horizon, especially if you're leveraging a server-side or hybrid deployment of Target.

@Amelia_Waliany  @Coops Our AI-powered activities, automated personalization and auto-target, will evaluate all existing profile data and segments defined or shared in Target.  It will score these and use them to determine the right exprience to deliver to maximize the conversion goal.  You can also select a subset of the profile data and audience segments that you want to evaluate.  In any activity with Target reporting, you can present audiences you wish to evaluate in reporting, but you can also select A4T for reporting and then any defined segment prior, during or after the activity can be used, even retroactively, for reporting and subsequent targeting opportunities.