Data Cloud-powered NBA in Salesforce (L5)

The steps

1. 01

   Connect and Unify Data Sources

   The foundation of Einstein Next Best Action (NBA) is a unified view of the customer. You must connect Salesforce CRM (Sales Cloud), your Product database (e.g., Snowflake, BigQuery, or Amazon S3), and your Billing system (e.g., Stripe, NetSuite) to Salesforce Data Cloud. • Log into Salesforce, navigate to Setup, and search for 'Data Cloud Setup'. • Use the 'Get Started' wizard to connect these external sources. For product data, you’ll likely use the 'Amazon S3 Connector' or 'Google Cloud Storage' connector. • Once connected, you must perform Data Mapping. Map your disparate data points to the Data Cloud Standard Data Model (DMO). For example, ensure that the 'Account ID' in your billing system and the 'Org ID' in your product database both map to the 'Account' DMO. • Configure Data Streams to refresh hourly. Use the 'Identity Resolution' feature to create 'Unified Individuals.' • Owner: Data Engineer or RevOps Lead. • Time Estimate: 10-15 hours. • Pitfall: Mapping data to the wrong DMO category (Individual vs. Account), which breaks downstream logic. • Definition of Done: You can see a 'Unified Link' record that connects one CRM Account to a specific Billing ID and Product ID.

2. 02

   Build Calculated Data Insights

   With unified data in place, you need to create 'Calculated Insights',the metrics that Einstein will use to make recommendations. These are SQL-based aggregations that live in Data Cloud. • Navigate to the 'Calculated Insights' tab in Data Cloud. • Create a new insight using the SQL editor. You need specific signals like 'Product Usage Drop > 20% in 30 days' or 'High Billing Volume + No Recent Sales Activity.' Example SQL: SELECT AccountId, SUM(usage_value) as total_usage FROM Product_Usage_DMO GROUP BY AccountId. • Ensure you have at least 24 months of historical data synced to training models to account for seasonality and long sales cycles. • Once insights are built, use the 'Data Cloud Triggered Flow' to send these signals back to Sales Cloud as 'Data Cloud DMO records.' • Owner: RevOps or Data Analyst. • Time Estimate: 8-12 hours. • Pitfall: Building insights on dirty or duplicate records. Always verify your 'Unified Account' counts before writing SQL. • Definition of Done: A dashboard showing calculated health scores and usage trends for every account in the system.

3. 03

   Configure Recommendation Strategies

   Now you must build the logic that decides which 'Action' to suggest based on the data. This happens in the 'Strategy Builder' or the newer 'Einstein Flow Builder.' • Go to Setup > Einstein Next Best Action > Recommendation Strategy. • Create 'Recommendations' (these are special Salesforce records). Each should have a Title (e.g., 'Upsell Higher Tier'), a Description, and an Action (usually an Auto-launched Flow). • Use 'Load' elements to bring in your Data Cloud Insights. • Use 'Filter' elements to set the rules. Example: IF [Usage > 90%] AND [Contract End Date < 90 Days] THEN [Recommend 'Renew Early']. • Map these recommendations to specific flows that automate tasks, like 'Generate Quote' or 'Send Drift Email.' • Owner: Sales Enablement or RevOps. • Time Estimate: 6-8 hours. • Pitfall: Creating too many competing recommendations (noise), which causes AEs to ignore the component. • Definition of Done: A strategy that outputs exactly one high-priority recommendation when a specific data threshold is met.

4. 04

   Embed NBA into the Sales Workflow

   To ensure adoption, the NBA component must be placed exactly where AEs spend their time. • Navigate to any Opportunity or Account record. Click the 'Gear' icon > 'Edit Page.' • In the Lightning App Builder, search for the 'Einstein Next Best Action' component in the left-hand menu. • Drag and drop it to the top right of the page or in a prominent center tab. • In the component settings, select the 'Strategy' you built in the previous step. • Set the 'Max Recommendations' to 1 or 2 to keep the view clean. • Use 'Component Visibility' filters to ensure the NBA block only shows up for high-value leads or specific sales stages. • Owner: Salesforce Administrator. • Time Estimate: 2 hours. • Pitfall: Placing the component below the fold where AEs won't see it without scrolling. • Definition of Done: An AE opens a live Opportunity and sees a 'Suggested Action' card with a clear 'Accept' or 'Reject' button.

5. 05

   Audit, Monitor, and Refine Logic

   AI models and data strategies lose accuracy over time due to 'drift',changes in market conditions or product features. You must implement a feedback loop. • Set up a 'Custom Report Type' that joins 'Recommendations' with 'Opportunity Outcomes.' • Track the 'Acceptance Rate' (how many times AEs clicked 'Accept') and the 'Conversion Rate' (how many accepted recommendations led to closed-won deals). • Schedule a quarterly 'Model Audit' meeting. During this session, review the bottom 20% of recommendations (those frequently rejected) and update the filter logic in the Strategy Builder. • If AE feedback is 'This isn't relevant,' go back to Step 2 and refine your SQL Calculated Insights. • Owner: RevOps Lead. • Time Estimate: 4 hours per quarter. • Pitfall: Failing to update the strategy after a major product launch or pricing change. • Definition of Done: A monthly report sent to sales leadership showing the Win Rate Uplift for opportunities that followed NBA suggestions versus those that didn't.

Next playbooks

• Voice-of-Customer → Product & Enablement Loop (L2)
• Zapier Agents for sales ops chores (L2)
• Agentic Nightly CRM Hygiene + Ownership Sync (L3)
• AI-driven price-quote acceleration with CPQ (L3)