Predictions

Predictions use AI to forecast visitor behavior. Create models that predict outcomes like purchase likelihood, churn risk, or content engagement.

What predictions do

Predictions analyze your visitor data and generate a score (0-100) indicating how likely each visitor is to take a specific action. Use these scores to:

• Personalize experiences for high-value visitors

• Target interventions for at-risk visitors

• Optimize marketing operations

Types of predictions

Prediction templates

Pre-built models for common use cases. Templates come configured with recommended settings.

Industry-based models

Models pre-trained on data from over 1.5 billion users. Deploy quickly for standard use cases.

Custom models

Build predictions from scratch for your specific needs.

Prediction list

The predictions page shows all your models:

• Prediction Target - The behavior you want to predict

• Status - Draft, Training, Training Completed, Published

• Model strength - How well the model performs

• Leading function - Key factors influencing predictions performance

Tabs

• Ongoing Models - Active predictions currently collecting data

• Archived Models - Inactive predictions saved for reference

Create a prediction from a template

1

Navigate to Predictions

Go to Core > Predictions and click Prediction Templates.

2

Choose a template

Browse available templates and select one that matches your goal.

3

Configure the prediction

Follow the wizard to:

• Name your prediction

• Set the target behavior

• Define any filters or conditions

4

Publish the model

Review your settings and click Publish. The model starts collecting data immediately.

Create a custom prediction

1

Start custom model creation

From Predictions, click Custom Model.

2

Define the prediction target

Choose what behavior you want to predict:

• Page visits

• Events

• Purchases

• Custom actions

3

Set the prediction window

Define how far ahead to predict:

• Within session

• Within 7 days

• Within 30 days

• Custom timeframe

4

Configure features

Select which data points the model should consider:

• Visitor attributes

• Behavior patterns

• Affinities

• Session data

5

Train and publish

The model trains on historical data. Once training completes, you can see the training characteristics and performance of the model. You can also change the threshold (useful for defining an audience later on) and eventually publish it to start scoring visitors.

Prediction scores

Each visitor receives a score between 0-100:

• 0-30 - Low likelihood

• 31-60 - Moderate likelihood

• 61-100 - High likelihood

Use these scores in audiences to segment visitors for different experiences.

Scores update in real-time as visitors interact with your site. A visitor's score can change within a session as they take actions.

Model strength metrics

AdaptiveCX uses different metrics to evaluate predictions:

• Accuracy - Overall correctness for balanced predictions

• Precision - How many predicted positives were correct

• Recall - How many actual positives were found

• ROC/AUC - Overall model performance across all thresholds

Choose metrics based on your use case:

• Use precision when false positives are costly

• Use recall when missing positives is costly

• Use accuracy for balanced outcomes

• Use ROC/AUC to compare overall model quality

Archive a prediction

To stop a prediction without deleting it:

1. Find the prediction in your list

2. Click the more options menu

3. Select Archive

Archived predictions stop scoring visitors but preserve historical data.

Use predictions in experiences

Once published, use prediction scores in:

• Audiences - Create segments based on score thresholds

• Adaptive Interactions - Show different content based on prediction

• Adaptive Search - Rank results based on predicted intent

• Adaptive Carousels - Personalize product recommendations