Predictive Audiences FAQ

Frequently asked questions about Predictive Audiences.

 

When should I use Predictive Audiences as opposed to Look-alike modeling?

Predictive Audiences and Look-alike modeling serve different use cases. The main differences between the two algorithms are these:

  1. Look-alike modeling takes a small audience as input and expands it. Predictive Audiences takes a large audience as input, and divides it into smaller distinct audiences, defined by your personas.
  2. The number of base segments is different for each algorithm. Predictive Audiences requires at least two baselines, while Look-alike modeling uses one baseline at most.
  3. Predictive Audiences performs real-time segment evaluation, while Look-alike modeling does not.

Based on your use case, you should decide which model will be more relevant to you.

You can think of building a Predictive Audiences model with a number of baselines as being the equivalent of building the same number of look-alike models, only without the real-time evaluation, and with a very high likelihood to have visitors belonging to multiple different personas, instead of one distinct persona.

 

How many personas/models am I allowed to create?

You can create up to 10 Predictive Audiences models. For each model, you can define up to 50 baseline traits or segments.

 

How can I build new segments from a Predictive Audiences segment?

Go to Audience Data > Segments, and click the Predictive Audiences folder. Find the desired segment, duplicate it, and edit it according to your needs.

 

When can I see the first results produced by my model?

Predictive Audiences model results are available within 24 hours from model creation, if the model runs successfully.

In case the model does not produce results within 24 hours, please reach out to your Adobe representative.

 

Why is my model not producing results or showing the Warning status?

Predictive Audiences models can fail to produce results due to a number of reasons:

  1. None of the selected persona traits / segments have enough user profiles. We recommend choosing your traits or segments so that each persona has at least a few hundred user profiles.
  2. None of the selected persona traits / segments have enough data in their user profiles (not enough traits to analyze).
  3. The target audience trait / segment does not have any active or onboarded users.
  4. The target audience users that were active or onboarded within the past 30 days do not have enough data in their user profiles (not enough traits to analyze).
  5. The target audience segment uses a different Profile Merge Rule from the one you chose for the model.
  6. The data source of your target audience traits may not be included in the Profile Merge Rule that you chose for the model.

For optimal results, follow the suggested guidelines from Selection Criteria for Personas and Selection Criteria for Target Audience.

 

Why is my model showing the Error status?

The model failed to run. In such cases, please reach out to your Adobe representative.

 

How can I change the Profile Merge Rule for a Predictive Audiences segment?

Create a new model by selecting the same personas and target audience as your previous model. During model creation, assign a different Profile Merge Rule.

WARNING

Alternatively, you can use Segment Builder to manually create a segment with an existing predictive trait and assign it a Profile Merge Rule of your choice.

However, we do not recommend this practice, since predictive traits automatically inherit the Profile Merge Rule of the model they belong to, and they are built from influential traits that comply with the Profile Merge Rule of the model.

 

What Profile Merge Rule should I choose?

When choosing the Profile Merge Rule for your model, analyze your use case closely.

Let’s say your target audience segment uses a Profile Merge Rule based on authenticated profiles + Device Graph profiles, and you select the same Profile Merge Rule for the predictive segments. In this case, both device level and cross-device level traits will be used in training the model and in the placement of the user into a predictive segment.

If, however, you select a Profile Merge Rule based only on device profiles, none of your cross-device traits will become influential and will not contribute to the placement of users into a predictive segment. This may adversely affect the model accuracy and reach.

Analyze your use case carefully and decide which trait types you want the model to learn from and what type of data you want the model to use for classification.

Could a user from the target audience who isn’t part of any persona trait / segment not be classified?

Yes, in case the user does not have any traits in his profile. In that case, the user will get a match score of 0 to all of the persona traits / segments, and therefore will not be classified into any of the predictive segments.

 

Can a user who was classified into one of the predictive segments be reclassified into a different Predictive Audiences segment?

Yes. Since the algorithm is trained on a daily basis, it applies the changes for each of the personas in terms of trait scoring. If a user who is part of a Predictive Audiences segment is active, the changes in their trait score can change the classification based on the past 30 days activity.

 

Can I see the traits by which audience classification is done?

Yes, you can see all influential traits for all baselines in the model reporting page. See Influential Traits.

 

What happens to the model if I edit one of its baseline traits or segments?

The model evaluates the traits or segments once a day. You should see the updated classification the next day after your update.

 

Can I select the data sources from which the model will learn?

No, selection of data sources is not supported. The Predictive Audiences algorithm learns from all your first party traits.

On this page