AI Assistant & Audience Management in Adobe Experience Platform

[#1]

Not too long ago, I saw an ask that got me curious.  The ask was to reduce the number of streaming segments the group had created.  As you may know, I help with our own implementation of Adobe Experience Platform.  I wanted to dive in and see what was going on.

General architectural pattern

First, we must understand something that applies all over any system.  In any system that has a shared environment across groups, it presents a challenge.  The more an environment is shared, the more standardization, governance and shared resources you must incorporate into your environment.  The more environments are split up and isolated, the less need of these you have, but you introduce the possibility of each having done things vastly differently ways, re-inventing the wheel (e.g. 20 different definitions of revenue, 20 different definitions of an audience or an abandon cart journey) and have more difficulty in sharing across the groups (e.g. preventing overcommunication to a customer across all these groups, coordinating messaging across these groups).  I've seen my far share of customers come back trying to unwind this isolation in favor of sharing and the various structures each group had implemented were making things very hard.

Our design in how we use Adobe Experience Platform at Adobe is to enable more coordination and sharing across groups.

That is just context, but not the focus on this blog.

The focus on this blog is one of the challenges companies run into when having many groups in a shared environment.

What we ran into

At Adobe we are moving more of these groups into one area.  Each of the groups had gone and created many streaming audiences, and while each didn’t create enough to exceed the guardrail, when we added up all the streaming segments across all the groups, it had.

Problem: We had exceed the guardrail of total number of streaming segments threshold of 500 per sandbox (we had 730).
Ask: Everyone to go in, evaluate their streaming segments.  See if they needed them, delete if not needed anymore and/or change their segment to batch if they could.

Me: All audiences aren’t created equal or needed forever. How can we come up with a suggested list of low or no value audiences?

Theories on audiences that might be of little/no value:

1. Any audience of zero (since can’t activate to it)
2. Any audience not being activated or have anything using it

Theory #1 - Any audience of zero

While I could scroll through the UI looking for all audiences with no profiles in it or even make some API calls, I’m now trying to use AI Assistant for everything first.

Prompt: list any audience that has a population of zero, created date, the first date it went to zero, the last date it went to zero, count of days it has been greater than zero, count of days it has been zero

Answer: List of 134 audiences

Some of these audiences have always been zero.  Probably an easy choice to delete these. Others look like they might fluctuate.  We could look at each of those closer if we wanted. I’ll let the people who created them do that.

Here is the prompt they can use:

Prompt: Show the population by day for audience "[audience name]”

Theory #2 - Any audience not being activated or having anything using it

Prompt: list all the streaming audiences, created date, their population, count of used in other audiences, count of used in destinations and count of used in journeys, sum up all the counts in another column where the sum of all the counts is 0

Answer: List of 67 audiences

This list gave us a great place to start and in one shot almost fixed the problem (assuming we can just delete all of these)

Summary

I’m sure I could keep going in my analysis and I might at some point, but the combination of both lists allowed us to easily identify a high likelihood of 201 audiences to be deleted that had little to no business value.  I could come up with count thresholds or journeys that are no longer active.  A number of ways to keep looking at this.  At the very least, they gave us a list to start with our analysis.

What it also did was give me more ideas on how I can tap into the AI Assistant for operational questions.  I’m going to dive more into this.

How are you using AI Assistant?  Put your examples in Experience League Communities.  I’d love to see them.  What works, what doesn’t work? What would you like to see?

Related Articles

#1 Using Adobe Experience Platform AI Assistant