Picking the right Intelligent App starts with the business problem, not the technology. Walk the decision from outcome to platform.
- State the business outcome
Be specific. We want to reduce lead-to-close time by 30 percent. We want to deflect 40 percent of Tier 1 support calls. Vague outcomes do not map to specific Intelligent Apps.
- Identify the closest Salesforce data
Intelligent Apps need data to train on or to ground their prompts. Lead Scoring needs a lead history; Service AI needs case data with resolution patterns. Confirm the data exists before committing.
- Match the use case to a product
Predictive scoring (Lead Score, Opportunity Score) for likelihood predictions. Einstein Bots for chat automation. Agentforce for multi-step agent workflows. Prompt Builder + Copilot for in-context AI assistance.
- Confirm Trust Layer applies
Verify the Einstein Trust Layer is configured for grounding, masking, and audit. Any Intelligent App handling sensitive data requires this baseline.
- Pilot, measure, scale
Start with a small pilot team. Measure the outcome (lead-to-close, deflection rate). Iterate the prompts or model configuration. Expand once results are solid.
- Build the integration with existing workflows
Most Intelligent Apps integrate with flows, page layouts, and Apex. Make sure the new capability shows up where users already work; bolt-on AI without integration fails to drive adoption.
- Many original Einstein point solutions are being modernized into the Einstein 1 Platform. Plan migrations alongside new feature adoption.
- Intelligent Apps without good source data underperform dramatically. Confirm data quality and volume before licensing.
- Trust Layer configuration is mandatory for any production Intelligent App handling regulated data. Skipping it creates compliance exposure.
- Agentforce conversation credits can exceed user-license costs at scale. Model the unit economics carefully before rolling out broadly.