Predictive analytics overview
Amplitude's predictive analytics features help you anticipate user behavior and act on it. Use them to segment users by what they're likely to do next and to surface the content most likely to drive each user toward a goal.
Predictions
Predictions segment users by their likelihood to perform a future action, such as subscribing, churning, or completing a purchase. Amplitude scores each user's probability hourly based on past behavior and groups users with similar probabilities.
Use predictions to:
- Decide who to include or exclude from a campaign.
- Adjust messaging frequency, pricing, or discounts based on a user's likelihood to convert.
- Personalize content for users with the highest affinity for a category.
To get started, read Build a prediction and Use prediction-based cohorts in your campaigns.
Recommendations
Recommendations rank the items most likely to drive each user to a predictive goal. Amplitude's AutoML system clusters similar users and assigns each user a ranked list of items based on historical conversion data.
Recommendations work best for user-based digital commerce personalization:
- Assortment: Rank items on a homepage or category page to increase engagement.
- Next-best action: Surface a second item in checkout or post-purchase email to increase conversions.
- Cross-sell: Rank items that signal new lifecycle stages to increase LTV.
To get started, read Build a recommendation and Use recommendations in personalization campaigns.
How predictions and recommendations work together
Predictions identify which users are most likely to act. Recommendations identify what to put in front of them. Combine them to maximize lift:
- Build a prediction for the outcome you care about, such as a second purchase or a subscription upgrade.
- Save the predictive cohort and sync it to your campaign destination.
- Build a recommendation that ranks items for those users.
- Deliver the ranked items through Amplitude's Profile API or a sync.
Related features
- Computations transform an event or property into a computed user property you can segment by.
- Behavioral cohorts group users by past actions.
- Predictive cohorts group users by predicted probability ranges.
Was this helpful?