Order matters. Activity Capture first (the data foundation), then Lead Scoring and Opportunity Scoring (predictions that need the activity data), then Deal Insights and Forecasting (signals that build on scoring), then Conversation Insights and Agentforce for Sales (the conversational layer). Out-of-order rollouts produce features that talk to thin data and quietly disappoint.
- Confirm Sales Cloud edition and feature coverage
Most built-in Einstein Sales features are bundled with Enterprise+ editions. Conversation Insights and Agentforce for Sales are separate SKUs. Setup, Company Information shows the edition.
- Enable Einstein Activity Capture first
Setup, Einstein Sales, Activity Capture. Connect Outlook or Gmail. Roll out to reps in waves and watch the activity history per Opportunity get denser over 30 days.
- Enable Lead Scoring and Opportunity Scoring
Setup, Einstein Sales, Lead Scoring (and Opportunity Scoring). Confirm minimum data volume (1,000 converted Leads, 200 closed-won Opportunities). Let the model train for a week.
- Add the score fields to list views and record pages
Without the score field on relevant list views, reps cannot sort by score. Add to Lead and Opportunity list views, record pages, and any relevant reports.
- Enable Deal Insights and Einstein Forecasting
Setup, Einstein Sales, Deal Insights (and Forecasting). These features benefit most from a populated Activity Capture; enable after 30 days of Activity Capture data.
- Build a weekly pipeline review on Deal Insights
The signals are most valuable when reviewed in a structured weekly meeting with the sales manager. Walk through opportunities with active warnings; the discussion catches risk before the gut-check meeting would.
- Layer Conversation Insights and Agentforce for Sales after
Once the underlying features are tuned, evaluate Conversation Insights (call transcript analysis) and Agentforce for Sales (SDR Agent, Sales Coach). Skipping the underlying tuning produces conversational features that talk to thin data.
Which Einstein Sales features are enabled. Roll out in waves; Activity Capture first.
Filters that scope which Leads or Opportunities the scoring model trains on. Useful when older data has different patterns than current.
Outlook or Gmail integration. Drives the activity data foundation for nearly every other Einstein Sales feature.
Per-signal sensitivity (single-threaded, stuck-in-stage, no-recent-activity). Tune per segment.
Threshold for what counts as a high-confidence predicted commit. Tunes how the rolled-up forecast displays to managers.
- Skipping Activity Capture and enabling Deal Insights or Forecasting first produces signals that train on sparse, biased activity data. The features look broken; the data foundation is the actual issue.
- Lead Scoring needs at least 1,000 converted Leads to train an org-specific model. Below that, scores fall back to cross-org patterns and accuracy drops.
- Conversation Insights and Agentforce for Sales are separate SKUs. Edition coverage matters; check explicitly before assuming they ship with the Sales Cloud license.
- Score fields not added to list views means reps cannot sort by score. The feature is on but invisible until the field is added to the relevant list views.
- Deal Insights signals only matter if reviewed. Without the weekly pipeline review structure, the signals fire and no one acts on them.