Next Best Action Strategy is the declarative "flowchart" that decides which Recommendations to surface for a given record — load recommendations, filter by context, optionally rank with Einstein scores, output a list. Built in Strategy Builder. The companion piece to Recommendation records and the NBA Lightning component.
- Open Setup → Next Best Action → Strategy Builder
Setup gear → Quick Find: Strategy → Strategy Builder.
- Click New Strategy
Top-right. Wizard prompts for Strategy Name, Description, and Object scope.
- Pick the Object the Strategy operates on
Account / Case / Custom Object. Strategy reads context from current record of this type.
- Drag elements onto the canvas
Element types: Load Recommendations / Filter / Generate / Map / Branch / Sort / Predict / Limit / Output. Connect with arrows.
- For Load Recommendations: pick the source object
Usually Recommendation but can be any object. Filters narrow which records load.
- For Filter: configure criteria
Drop Recommendations not relevant to this record.
- For Predict: bind an Einstein Discovery model
Optional — score recommendations using a predictive model and rank by score.
- For Output: set the final list
Output is what the NBA Lightning component shows. Up to N recommendations.
- Save and Activate
Strategies must be activated to fire. Inactive strategies return nothing.
- Bind to a Lightning Page via the Einstein Next Best Action component
Lightning App Builder → Account / Case page → drop NBA component → bind to this Strategy.
Load / Filter / Generate / Map / Branch / Sort / Predict / Limit / Output.
Rank by Einstein Discovery model output.
Conditional paths based on record context.
Cap the output list size.
- Strategies must be activated. Built-but-inactive strategies return zero recommendations — the most common debug issue.
- Predict element requires a deployed Einstein Discovery model bound to the Strategy. Without one, ranking is rules-only.
- Heavy Load Recommendations + Filter operations can be slow. Strategies running on every record page load add latency — keep them lean.