Predictions: Use Amplitude's AI to help maximize lift

As part of Amplitude Activation, predictions help you optimize targeting workflows to generate lift.

Instead of segmenting users by past behavior, predictions segment users by their likelihood to perform a specific future action. Predictions are most useful for communication frequency, dynamic pricing, and content personalization workflows. Use predictions to:

• Specify which users to include or exclude in a campaign.
• Adjust messaging frequency based on a user’s likelihood to convert.
• Modify pricing, offers, and discounts relative to a user’s likelihood to convert.
• Fine-tune ad, email, or website content based on a user’s affinity for that content type.

Predictions construct a mathematical model to forecast the likelihood that a user takes a specific action in your product. The model groups users who have similar probabilities.

First, decide what predictions to build.

What question should your prediction answer? In most cases, the answer connects to the objectives that guide your company. Start by mapping the user journey, including KPIs from the user’s first product interaction to their last touch. Common user journey steps include signup, activation, retention, and churn.

After you identify the steps, fill in the milestones by specifying each major button interaction between those steps. Build a prediction for each step of this journey.

For example, an ecommerce product might have this user journey.

Who should use predictions, and when

Predictions work best in specific situations:

• When your target outcome lacks a clear funnel. These outcomes usually result from complex user journeys and can be difficult to frame as a clear binary event. Common examples include activation, retention, engagement, and long-term value. If these metrics matter most to your product, consider predictive cohorts.
• When you’re trying to drive incremental lift to these outcomes. A thoughtfully designed prediction can drive a 5% to 20% lift relative to a behavioral cohort.
• If your product has over 100,000 monthly average users. Smaller products may not generate sample sizes large enough for reliable statistical inferences.

Conversely, your company is less likely to benefit from predictions if you:

• Sell physical products.
• Are in the B2B space.
• Lack a marketing team.

Before you work with predictive cohorts, read Build a prediction and Use prediction-based cohorts in your campaigns. The next section describes how Amplitude Activation builds predictions.

How predictions work

Predictions use past behavior to predict future behavior. When you build a prediction, Amplitude Activation creates a deep learning model to distinguish between users who perform the action you specify and users who don't.

Amplitude Activation starts with users who were in the starting cohort two periods ago. Amplitude then identifies which users did and didn't perform the action one period ago. You can set a period to seven, 30, 60, or 90 days.

Next, Amplitude Activation uses an advanced transformer-based sequence model to compare those two user groups across four variable sets: events, event properties, user properties, and user activity sequences.

• Events: How often each user triggers the top 25 events related to the prediction target, every week for the last four periods.
• Event properties: How often each user triggers the most frequently queried event properties, every week for the last four periods.
• User properties: The initial value and most recent value of each user property in the last four periods.
• User activity sequences: Each user’s activity sequence for their most recent 128 events and time intervals.

The transformer encoder processes these variables and builds an alias cohort on top of the user’s activity sequence. Most behavioral cohorts rely on three to five manually designed signals. Amplitude's predictions use a transformer-based AI model with hundreds of behavioral signals.

The model calculates a probabilistic score for every user in the starting cohort. The score measures how likely each user is to perform the target action during the specified period. As the model learns and responds to seasonal data, Amplitude recalculates each user’s probability score daily or hourly, depending on your configuration.

To get started, read Build a prediction and Use prediction-based cohorts in your campaigns.