Personalization and Recommendations
Personalization building blocks
Personalization in Target combines audiences, profile attributes, and recommendation algorithms. Experience targeting delivers a fixed experience to a segment; Auto-Target personalizes per visitor using a model trained on behavior. Understand that personalization quality depends on data quality - profile attributes must be populated and audiences must be meaningful.
Recommendations architecture
Recommendations require a catalog (the set of items), criteria (the algorithm and its rules), and designs (the display template). Criteria include behavior-based (people who viewed this also viewed), popularity, and attribute-based algorithms. The exam expects you to match the algorithm to the data available: a brand-new catalog with no behavioral data should start with popularity-based criteria, not collaborative filtering.
Feeds and catalog hygiene
Catalogs are populated by feeds (CSV, Google Product Feed, Analytics-based) or mbox collection. Feed failures, stale items, and missing attributes degrade recommendations. Know that the environment (staging vs production) must match the feed target.
Exclusions and business rules
Recommendations support exclusions (don't recommend items already purchased) and static filters. These are business rules layered on the algorithm.
Worked rule of thumb
If a client has a new site with no traffic history, recommend popularity or attribute-based criteria first, then migrate to behavioral algorithms once data accumulates.
Common traps
Assuming collaborative filtering works without behavioral data; forgetting that designs control rendering; ignoring feed freshness and environment mismatches.