Experience Design and Activity Configuration
Know each activity type by its decision logic and statistical model. A/B test: two or more experiences, random split, frequentist or Bayesian statistics, one primary metric. Multivariate (MVT): tests combinations of offers across locations, reports on main effects and interactions; it needs far more traffic per location. Experience Targeting (XT): rules-based assignment to audiences, not a statistical test - no winner declaration. Automated Personalization (AP): machine learning assigns visitors to offers using a random-forest model, with a control experience. Auto-Target: ML selects the best experience (not offer) per visitor. Recommendations: algorithms (e.g., people who bought this, recently viewed, popularity) plus inclusion rules and design.
Traps: treating XT as a test (it is deterministic targeting); expecting AP/Auto-Target to declare a "winner" (they report lift vs control); forgetting that MVT's traffic requirement scales with combinations; confusing Auto-Allocate (bandit, picks a winner) with Auto-Target (personalizes per visitor). Also remember success metrics must be defined before launch and A4T requires Analytics linkage.
Rule of thumb: minimum sample size grows with the number of experiences/combinations and shrinks with larger effect size; always compute required sample before launch.
Worked example: four locations × three offers = 12 combinations. With modest traffic, MVT is infeasible; instead run sequential A/B tests per location or use Auto-Allocate to find the best single experience.