AI Models

AI Models in BPP are machine-learning and analytical models that run on your customer data and produce prediction scores or customer segments. These outputs are automatically available as filter conditions when building audiences and as value inputs when configuring signals.

How AI Models work

BPP runs each active AI model on a daily schedule. The model reads your customer and event data, computes scores or segments, and writes the results back to your dataset. From that point, you can use those results in the audience builder — for example, filtering for users with a purchase probability above 0.75, or targeting only "Champions" from your RFM segmentation.

Results are always linked to BPP's unified customer profiles (Bytek IDs), so the same person's score is consistent across audiences and signals.

Supported model types

Predictive models

Action Prediction Assigns each customer a probability score (0–1) that they will perform a specific action — for example, make a purchase, renew a subscription, or churn. You define what counts as a positive outcome and what counts as a negative one.

• Use in audiences: action_prediction_score >= 0.75

pcLTV — Predicted Customer Lifetime Value Estimates the future revenue each customer is expected to generate over a chosen time horizon (3, 6, or 12 months). Trained on your transactional history.

• Use in audiences: pcltv_6m >= 100
• Use in signals: send predicted LTV as the conversion value to power value-based bidding.

Descriptive models

RFM Clustering Segments customers based on their transaction behaviour:

• Recency — how recently they last bought.
• Frequency — how often they buy.
• Monetary — how much they spend.

Outputs a segment label per customer (e.g., Champions, Loyal Customers, At Risk, Hibernating).

• Use in audiences: rfm_cluster = Champions

Interest Classifies customers' interests based on their browsing and event behaviour. Three subtypes:

• Product Interest — based on product categories already present in your event data.
• IAB Interest — classifies content into standard IAB taxonomy categories (e.g., Technology, Sports, Finance).
• Custom Interest — uses a taxonomy you define.
• Use in audiences: interest_iab contains Technology

Model instance lifecycle

Status	What it means
Draft	Configuration saved but not yet activated. No runs scheduled.
Ready	Activated. The model runs on its daily schedule.
Running	A model run is currently in progress.
Complete	Last run completed. Prediction scores are available in the audience and signal builders.
Error	Last run failed. Check the configuration and your input data.
Suspended	Paused. No new runs until re-enabled.

In this section

• Configure a Model — how to set up a new AI model instance.
• Model Instances — understanding the model instance concept.
• Predictions — how model outputs are stored and used.