Future behavior predicts how likely each customer is to take a specific action, such as purchasing or churning. Next best action recommends which action to take for each individual. Similar customer predictions rank how closely each customer resembles a reference segment you supply.

No, mParticles predictive capabilities are designed for marketers. You define the conversion goal and the time frame, and mParticle handles the modeling. There is no pipeline to build and no model to maintain, which is what lets a marketing team run predictions without a data science queue in front of them.

Predictive attributes are machine-learning scores that live on the customer profile alongside behavior and demographics. You define the outcome, and mParticle analyzes thousands of behavioral signals to score which customers are statistically most likely to get there. Once generated, they behave like any other user attribute.

A new prediction can take up to 24 hours to calculate, and shows as calculating until values exist for every relevant customer. After that, predictions refresh automatically on a weekly schedule, so scores keep pace with behavior instead of aging out.

As a score and a percentile. The score is the likelihood of the action, from 0 to 100 percent. The percentile ranks each customer against everyone else scored. Both sit on the profile, and percentile is generally the better choice for audience building because it stays stable as the underlying population shifts.

Historical event data, and enough real conversions inside your chosen time frame for the model to learn from. Predictions most often fall short when the conversion window is too narrow to include a meaningful number of converters, so a longer lookback usually produces a stronger model than a short one.

Anywhere you use a regular attribute. Add them as audience criteria to target customers by likelihood, query them through the Profile API to personalize an experience in the moment, or forward them to any connected destination. See Segmentation for how predictions combine with rules-based audience logic.