Predictions Overview

mParticle Predictions help you understand which customers are likely to act, which offer is most relevant to each customer, and which users resemble an audience you already trust. Use predictions to build higher-performing audiences and activate campaigns with more confidence.

mParticle offers three types of predictions:

• Future Behavior Predictions: these predictions estimate the likelihood that a user will perform a specific action
• Next Best Actions (NBAs): these predictions identify which action or offer is most likely to result in a conversion
• Similar Customer Predictions: these predictions identify users who resemble a reference audience or segment you define

How do predictions work?

All prediction types generate predictive attributes that are stored on user profiles. You can use these attributes like other user attributes when building targeted audiences.

The key difference is that predictive attributes are created according to your prediction definition, rather than being captured automatically from user activity in your app or site. Once generated, predictive attributes behave the same as any other user attribute.

Why use predictions?

Marketers have traditionally relied on rules of thumb such as website visits, page views, and journey completions to create customer segments. While manual segmentation can be effective, it’s limited by the scope and speed of human decision making and the length of experimentation cycles.

Predictions remove these limitations by automating the process. After you define your goal, mParticle analyzes customer behavior and profile data to identify the users, offers, or audience expansions most likely to support your campaign. This lets you make data-driven decisions faster and launch campaigns with greater confidence.

Predictions setup requirements

Before you can create predictions, your workspace must have:

• Predictions product access enabled.
• The required Predictions event output and feed input configured.
• Prediction schema synchronization and setup verification completed.
• At least 45 days of usable historical data, collected through the live integration or supplied by a completed historical replay.

If setup is still in progress, prediction creation may be unavailable until setup completes and enough usable historical data is available. Contact your mParticle account representative to confirm that setup is complete and your data meets the requirement.

Historical data requirement

All mParticle prediction types require at least 45 days of usable historical data before you can create a prediction. For more reliable results, 90 or more days is recommended when available.

This data must be available to Predictions. Events visible elsewhere in mParticle aren’t necessarily ready for Predictions. The historical data requirement is separate from prediction configuration, Strength, pipeline limits, and calculation status.

If your workspace doesn’t yet have 45 days of usable historical data, prediction creation isn’t available yet. You can continue collecting data through the live integration or contact your mParticle account representative to discuss whether historical replay or setup support is available for your account. A replay must supply enough usable historical data to satisfy the requirement before you can create predictions.

Prediction types

Predictions fall into three main categories based on the business outcomes they enable:

Future behavior predictions

Future behavior predictions tell you how likely each user is to take a specific action, like purchasing a new product or upgrading a subscription.

Learn more about future behavior predictions in the Future Behavior Predictions overview.

Next best action

Next best action predictions, shown as Best Offering in the product, tell you which option from 2 to 10 defined offers or actions is most likely to result in a customer taking a defined action.

Learn more about next best action predictions in the Next Best Action overview.

Similar customer predictions

Similar customer predictions help you find users who resemble members of a reference segment, even when many user profiles are missing the attributes you’d normally use to identify them.

You can create a similar customer prediction directly from the Predictive Attributes page, or use the Expand with AI shortcut from an existing audience in the audience builder. Audience Expansion is the workflow; Similar Customer Predictions are one prediction type that workflow can create. For audiences defined using only user attributes, Expand with AI creates a similar customer prediction using that audience as the reference segment. For audiences that include event-based criteria, the same button routes to a Future Behavior Prediction instead.

Learn more about similar customer predictions in the Similar Customer Predictions overview.

When to use predictions

The most fruitful use cases for predictions tend to have the following characteristics:

• The data includes behavioral events. Behavioral events are an important factor in generating a prediction. User attributes alone (such as customer lifetime value or subscriber status) are not often sufficient to predict user behavior.
• The action occurs frequently. The more frequently an action occurs, the more behavioral data mParticle will have on the user base to make stronger predictions. Purchases and membership upgrades are good examples.
• The action is intentional and meaningful. Conversion actions like purchasing an item, subscribing, and upselling reflect meaningful consumer decisions, whereas product views, page visits, and other browsing behavior can be less intentional.

Predictions, pipelines, and predictive attributes

A prediction is the configuration you create and manage in mParticle. A pipeline is the machine-learning execution unit used to calculate the prediction. A predictive attribute is the result written to user profiles after successful calculation. You can use predictive attributes to build audiences and activate campaigns.

Pipeline usage and limits

Pipeline usage depends on prediction type:

• Future Behavior uses 1 pipeline per prediction.
• Similar Customers uses 1 pipeline per prediction.
• Next Best Action uses 1 pipeline for each configured offer.

For example, a Next Best Action prediction with 3 offers uses 3 pipelines even though it appears as 1 prediction in the UI.

Your account’s pipeline limit depends on your account plan and organization settings. The product displays current usage or warnings when the limit affects creation.

Each Next Best Action prediction supports 2 to 10 offers or actions. This offer-count limit is separate from your account’s pipeline limit. For a larger set, narrow the offers or contact your mParticle account representative to discuss the use case. Next Best Action isn’t intended to act as a recommendation system for a large catalog. For more information, see Next Best Action offer count.

How predictions generate attributes

Different prediction types produce different attributes on user profiles:

• Future Behavior Predictions generate two attributes: a score (0-100% likelihood) and a percentile (relative ranking). Both are available on user profiles, though percentile is recommended for audience building.
• Next Best Actions generate two attributes: a percentile score and the recommended action for each user based on a set of options you provide.
• Similar Customer Predictions generate a single percentile attribute that represents how similar each user is to your reference segment.

Using predictive attributes in audiences

New predictions can take up to 24 hours to calculate. Wait until calculation completes successfully and results and target ranges are available before using the prediction’s interactive chart to create an audience.

Open the prediction details page, review its results, select the target range shown in the product, and click Save as New Audience. Follow the steps for your prediction type:

• Create an audience with Future Behavior Predictions
• Create an audience with Next Best Actions
• Build an audience with Similar Customer Predictions

Predictions that haven’t been refreshed in over 30 days become inactive, but their attributes remain on user profiles and can still be referenced in audiences. Using inactive predictions may affect campaign performance since the predictions were generated using old data.