Configure engagement signals and metrics

Updated Nov 8, 2026 · Based on web research, 2026-11-08

Engagement signals are the raw behavioural events — views, clicks, add-to-carts, purchases — that tell a recommender what “interest” looks like. Metrics and compound metrics turn those signals into numbers a recommender can train on and rank by.

Define signals and metrics

1. From Personalization, review the available engagement signals (for example view, add-to-cart, purchase) captured from your connected sources.
2. Choose the signals that best represent the behaviour your recommender should learn — don’t select every available signal.
3. Build a metric from one or more signals (for example a “purchase count” metric built on the purchase signal).
4. Combine two or more metrics into a compound metric where a single score should reflect several behaviours together.
5. For a per-profile affinity score that updates as new events arrive, build a real-time calculated insight instead of a static metric.
6. Confirm the signals and metrics appear before you train or retrain an objective-based recommender.

1. 1 Engagement signal view, click, purchase
2. 2 Metric built from one or more signals
3. 3 Compound metric combines several metrics
4. 4 Recommender trains on the result