ML segment - Auto Personalisation / Auto Target

[YT_Tsutsui]

Auto-target / auto-personalisation experts! Can you please help!! 

We recently started running an AP on a brand website, where our visitors are not mostly not authenticated (anonymous). After 15 days of ML training period, we got a list of ML segments, but majority of segments were based on browser, user OS, screen size, etc. We do not believe those would have had impact on the way visitors responded or not responded to half a dozen offers we put in mix. 

Can you give me suggestions on how we can improve this. We are about to reset and restart this. 

• Are there audiences that can be added in AAM or Target that would be effective in anonymous user segmentation?
• Are there effective tactics on designing blacklist for ML? 

Also, on the blacklisting process, if you have experience using it, can you share your learnings?

Thanks everyone for your help!

[SundeepKatepally]

Depending on how many products you have in the ecosystem you can define segments

1) you can define in AT(adobe target)

2) you can define in AA(adobe analytics)

3) you can define in AL (audience library)

4) AAM

[Ryan_Roberts_ · Accepted solution]

Hi @YT_Tsutsui,

This is a great question. Using Automated-Personalization/Auto-Target is powerful, but can certainly take some tuning and adjusting to get right. I like to frame the setup of these ML/AI activities like so:

1. Consider the different persona's or segments that would be coming to your site to convert. Often products or categories is a useful way to think about the different personas. For example, for a banking site:
   1. Customers interested in checking
   2. Customers interested in home loans
   3. Customers interested in credit cards
   4. Customers interested in auto loans
2. Now we need to create content relevant to those personas. Showing the right message to the right group can reduce friction to ultimate conversion and that's what we'd like Target's ML/AI to help with.
3. We need to help Target detect people that would fall into one of our personas. If we don't purposefully go through this exercise we force Target to rely solely on OOTB attributes like browser, geolocation, timing, etc. A little effort here can greatly increase the value of the attributes available to model off of. Some examples could be:
   1. Passing a referring campaign parameter
   2. Passing an internal search term (e.g., they search for 'auto loan rates')
   3. Passing any visitor preferences that could be helpful (e.g., unsubscribed from newsletter)
   4. Pass site behavioral (e.g., setup a profile script to count views of auto loans pages)
   5. AAM segments
   6. Customer attributes data

If you have AAM and have A4T and Shared Audiences setup then AAM segments should automatically get evaluated by Targets ML/AI models. Also, to your point you do at times need to black list certain items that you may know shouldn't be helpful for modeling. That can be requested via a client care ticket currently. Send them the attribute you wish to blacklist and your Target account client code. They can do it at an account or activity level. I usually find the account level most helpful as they tend to apply to other activities too.

Lastly, in some cases Target's ML/AI may not work very well. This might be true for a page that has a large percentage of first-time visitors hitting it. Since those visitors have really shallow profiles, Target models consequently have much less to go off of to determine what experience/offer they might best respond too.

Hope that helps,

Ryan

[YT_Tsutsui]

Hi @Ryan_Roberts_   

Thank you for your detailed response. Much appreciated. I understand the points you listed, and that is exactly what I expected the best practice to be.

In a way, my question is also thrown towards users and consultants of customers using this solution, to see if there is any comment on what else they might have found to be useful, when faced with the situation of the landing page x anonymous users combination.