Guardrails, AI Models, and the future of Decisioning
Learning objective
By the end of this lesson, you will be able to:
- Recall the two guardrails most commonly encountered in practice
- Differentiate auto-optimization from personalized optimization AI models
- Explain how decisioning extends beyond the legacy ODE product
- Summarize how the eight building blocks fit together end to end
Lecture
The below video covers the two most common decisioning guardrails, how AI ranking models differ from manual ranking formulas, how decisioning extends beyond the legacy Offer Decisioning Engine, and a recap of how the eight building blocks connect end to end.
Key takeaways
- The two most commonly hit guardrails: 10,000 decision items per IMS org (not per sandbox), and 100 custom attributes per schema; check product documentation for current numbers, as these are subject to change
- AI models can be used within ranking formulas; auto-optimization is non-personalized and optimizes on global performance, while personalized optimization serves items toward specific business goals per profile
- Model scores computed outside AEP can be brought in as profile attributes and used in eligibility rules or ranking formulas
- Decisioning goes beyond the legacy Offer Decisioning Engine: it uses XDM for reusability, delivers JSON to headless applications, and separates the decision item from the treatment
- Decisioning can condition journey pathing and entry priority on a decisioning response
- End to end: decision item XDM defines attributes → decision item creation assigns values and eligibility → collections group items → ranking formulas adjust priority per profile → selection strategies rank and filter a collection → decision policies apply strategies to a channel → decision packages live on the hub or edge
recommendation-more-help
blueprints-learn-help-blueprints