Einstein Next Best Action
Einstein Next Best Action (NBA) is the Salesforce recommendation engine that surfaces personalized action suggestions to users based on rules, data, and AI-predicted scoring.
Definition
Einstein Next Best Action (NBA) is the Salesforce recommendation engine that surfaces personalized action suggestions to users based on rules, data, and AI-predicted scoring. A user looking at an Account in the Service Console might see suggestions like "Offer warranty extension", "Schedule follow-up call", "Send retention discount", each tied to the rules and the predicted likelihood that the customer will respond. The engine combines a Strategy (the rules and branching logic) with optional Einstein Discovery predictions to rank the recommendations.
NBA is built on three concepts: Recommendations (the individual items shown to the user), Strategies (the branching logic that selects which Recommendations are eligible), and Action Strategies (the runtime objects deployed to a Lightning page). The recommendations render in the Einstein Next Best Action Lightning component, dropped onto the record page in App Builder. Each recommendation includes a label, description, and an action (accept, reject, mark as completed). NBA is the bridge between the data Salesforce holds about a customer and the next concrete step a salesperson, agent, or marketer should take.
How Einstein Next Best Action recommends customer actions
Recommendations: the unit of action
A Recommendation is a record on the Recommendation sObject. It has a Name, Description, Image, and an Action (a Flow, a URL, or a record creation). When a user accepts a recommendation, the Action runs: the Flow launches, the URL opens, the record is created. When a user rejects it, the Strategy can react (suppress for 30 days, learn the rejection pattern). Recommendations are designed to be human-readable; the goal is the user understanding "what is this and what should I do with it" in two seconds.
Strategies: the branching logic
A Strategy is the rule set that picks which Recommendations to show for a given context. It is built in Strategy Builder, a visual canvas with nodes that load recommendations, filter them, branch on conditions, and merge results. A strategy might Load all eligible Loyalty Offers, Filter by customer Tier, Sort by Predicted Acceptance Rate, Limit to top 3. The output is the ranked list shown in the NBA Lightning component. Strategies can call sub-strategies, reference Apex, and integrate with Einstein Discovery models. The visual builder makes complex logic accessible without code.
Einstein Discovery integration
NBA can call Einstein Discovery predictions inside a Strategy. An Einstein Discovery model trained on historical acceptance data ("which customers accepted retention offers last quarter") generates a probability score for each candidate recommendation. The Strategy uses the score to rank or filter. This is what turns NBA from a rules engine into an AI-driven recommendation engine: the rules pick eligibility, the predictions pick order. Configuring this requires both NBA setup and an Einstein Discovery story trained on the right data; the integration point is a Predict node in Strategy Builder.
Strategy Builder: the visual canvas
Strategy Builder is the Salesforce-built UI for defining strategies. It is a canvas where each node represents a logical step: Load Recommendation (start with a list), Filter (narrow by criteria), Map (compute derived fields), Branch (split into multiple paths), Merge (combine results), Sort, Limit. Nodes connect with arrows. The flow runs at page render time, with the user's context (record ID, field values) passed in. Most strategies have 5 to 15 nodes. Complex strategies can reach 30 to 50 nodes and benefit from sub-strategy decomposition.
The Lightning component and where NBA appears
The Einstein Next Best Action Lightning component is what users see. It is dropped on a record page through Lightning App Builder, configured with a Strategy and optional layout settings (number of recommendations, accept/reject buttons, dismissal behavior). The component refreshes when the underlying record changes, so the recommendations stay current. Most NBA deployments place the component on the Service Console case page, the Account record page, or a custom Sales Path screen. Mobile uses the same component with a touch-optimized layout.
Accept and reject feedback loop
When the user accepts or rejects a recommendation, the platform records the event. Accepted recommendations trigger the configured Action (Flow, URL, record creation). Rejected recommendations are logged with the rejection reason. The feedback feeds two things. First, the strategy can suppress repeatedly rejected recommendations for that customer. Second, Einstein Discovery models retrain on the accept/reject signal, improving prediction quality over time. The feedback loop is what makes NBA learn from real user behavior rather than just static rules.
Use cases beyond sales: service, marketing, field
NBA started as a sales-focused product but now spans Service Cloud, Marketing Cloud, and Field Service. Service Cloud uses NBA to suggest case escalations, refund offers, retention scripts. Marketing Cloud uses it to recommend journey entry events. Field Service uses it to suggest on-site service add-ons. Any record-page context with a customer relationship and a set of possible actions can benefit. The strategy logic differs by use case, but the underlying engine is shared.
Setting up Einstein Next Best Action
Setting up NBA is a Setup workflow: enable the feature, create Recommendation records, build a Strategy in Strategy Builder, drop the Lightning component on a record page, and iterate based on accept/reject feedback.
- Enable Einstein Next Best Action
Setup, Einstein Next Best Action, Enable. The feature requires a Salesforce license that includes NBA (most enterprise editions).
- Create Recommendation records
App Launcher, Recommendations, New. For each possible action, create a Recommendation with Name, Description, Image, and Action (Flow, URL, or record creation). Build 5 to 10 to start.
- Open Strategy Builder
Setup, Strategy Builder. Click New Strategy. The visual canvas opens. Pick the context object (Case, Account, Opportunity).
- Build the strategy logic
Add Load nodes to pull eligible Recommendations. Add Filter nodes to narrow by criteria. Add Sort and Limit nodes to rank and cap. Optionally add Predict nodes to call Einstein Discovery.
- Activate the strategy
Click Activate on the strategy. The strategy becomes available for the Lightning component.
- Add the Einstein NBA component to a Lightning page
Lightning App Builder, edit the record page, drag Einstein Next Best Action from the Standard components. Configure the Strategy reference. Save.
- Train users and monitor accept/reject feedback
Brief the user team on what the recommendations mean. Monitor the AcceptedRecommendation and RejectedRecommendation reports to tune the strategy over time.
The individual action suggestion record with Name, Description, Image, and Action (Flow, URL, record creation).
The branching logic that selects eligible Recommendations for a context. Built in Strategy Builder.
The visual canvas where strategies are designed with Load, Filter, Branch, Sort, Limit nodes.
Optional Predict node that calls an Einstein Discovery model for AI-driven ranking.
The user-facing widget rendered on record pages through Lightning App Builder.
User action telemetry that feeds back into strategy suppression and Einstein Discovery model retraining.
- NBA requires specific licensing. Check the Salesforce edition before scoping the project. Not every edition includes the feature.
- Strategies activated in production cannot be edited; you have to clone and create a new version. Plan the workflow to avoid mid-quarter changes.
- Einstein Discovery predictions require a trained model on the right historical data. Without data, the predictions are not useful and a rules-only strategy is more honest.
- The Einstein NBA component refreshes on record changes. Strategies that perform expensive operations (multiple Apex calls, large SOQL queries) slow record page load.
- The accept/reject feedback loop only improves quality if recommendations are actually relevant to the user. Bad initial recommendations train the model on noise.
Trust & references
Straight from the source - Salesforce's reference material on Einstein Next Best Action.
- Einstein Next Best ActionSalesforce Help
- Strategy BuilderSalesforce Help
- RecommendationsSalesforce Help
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. What does Einstein Next Best Action do?
Q2. Where do recommendations appear?
Q3. What's a typical use case?
Discussion
Loading discussion…