orcaqubits/sf-einstein

Einstein AI for Salesforce Commerce

Before Writing Code

Fetch live docs before implementing Einstein AI features.

1. Web-search: "Salesforce Commerce Cloud Einstein Recommendations API 2026"
2. Web-search: "Salesforce Einstein Search dictionaries relevance tuning 2026"
3. Web-search: "Salesforce Data Cloud B2C Commerce personalization 2026"
4. Web-search: "Salesforce Commerce Cloud PWA Kit Einstein recommendations 2026"
5. Web-fetch the Einstein Recommendations API reference for current configuration parameters
6. Web-fetch Data Cloud connector setup and unified profile schema docs

Conceptual Architecture

Einstein Recommendations

Recommender Types:

| Type | Description | Typical Placement | |---|---|---| | Product-to-Product | Similar or complementary items (cross-sell) | PDP | | Recently Viewed | User's browsing history | Homepage, category | | Also Bought | Products frequently purchased together | Cart, PDP | | Trending | Popular items across all users | Homepage, category | | Top Sellers | Best-selling products by category/site | Homepage, category | | Personalized | ML-driven per-user recommendations | Homepage (returning users) |

Recommender Configuration (Business Manager):

• Create recommenders with specific types and filtering rules
• Configure zone placement (homepage, PDP, cart, category page)
• Set filtering: exclude out-of-stock, price range limits, category restrictions
• Map recommendation zones to recommenders

Activity Collection:

Einstein activity tracking uses a collect.js library loaded on storefront pages. It automatically captures product views, add-to-cart, purchases, and search events. Configured via Business Manager > Einstein > Activity Tracking.

Warning: The _etmc beacon pattern is for Marketing Cloud Einstein, not Commerce Cloud Einstein. Do not confuse the two.

Recommendation Zones:

• Define placement areas on storefront pages
• Map zones to recommenders in Business Manager
• Customize rendering per zone (carousel, grid, list)

Einstein Predictive Sort

Personalized category page sorting powered by ML.

| Aspect | Detail | |---|---| | Input | User behavior (clicks, purchases, browse history) | | Output | Per-user product ranking on category pages | | Fallback | Default sorting for new / anonymous users | | Config | Per-category toggle in Business Manager |

Einstein Search

Search Dictionaries:

| Dictionary Type | Purpose | Example | |---|---|---| | Synonyms | Map equivalent terms | sneakers -> running shoes | | Hypernyms | Broader category terms | iPhone -> smartphone | | Compound Words | Multi-word phrases | ice cream, swimming pool |

Search Relevance Tuning:

• Boost or bury specific products in search results
• Configured per site/locale in Business Manager
• Sorting rules: relevance, price, newest, custom

Typeahead Suggestions:

• Search-as-you-type with phrase suggestions and hit counts
• Configured via Business Manager search settings

Data Cloud Personalization

Integration Architecture:

B2C Commerce -> Data Cloud Connector -> Unified Profile
  -> Segmentation + ML Models (Einstein)
    -> Personalized Recommendations / Content
      -> Commerce Storefront (SFRA / PWA Kit)

Key Concepts:

• Aggregates data from Commerce Cloud, Service Cloud, Marketing Cloud into unified profiles
• Real-time profile updates via Data Cloud connector
• Segment membership drives personalized content and recommendations
• Cross-channel offer consistency based on unified customer view

Data Cloud vs Commerce Cloud Einstein:

| Aspect | Commerce Cloud Einstein | Data Cloud Personalization | |---|---|---| | Data source | Commerce activity only | Cross-cloud unified profile | | Setup | Business Manager config | Data Cloud connector + config | | Segments | Implicit (ML-driven) | Explicit (rule-based + ML) | | Best for | Product recommendations | Cross-channel personalization |

Zone Placement Strategy:

| Page | Recommended Zones | |---|---| | Homepage | Trending + personalized (returning users) | | PDP | Similar products + complementary items (cross-sell) | | Cart | Cross-sell + upsell opportunities | | Category | Predictive sort + trending in category | | Search Results | Einstein-ranked results |

Code Examples

// Pattern: SFRA recommendation zone
// Fetch live docs for Einstein Recommendations API
var recs = einsteinAPI.getRecommendations(zone, customer);
// Render recs in ISML template

// Pattern: PWA Kit recommendations hook
// Fetch live docs for commerce-sdk-react useRecommendations
const {data} = useRecommendations({recommenderName, products});

// Pattern: Fallback when Einstein unavailable
// Fetch live docs for CacheMgr and fallback strategies
// try Einstein -> catch -> return getTopSellers(zone)

Best Practices

Activity Collection

• Configure activity tracking before enabling recommendations (minimum 2-4 weeks of data)
• Track all key events: views, add-to-cart, purchases, search queries
• Validate tracking via Einstein Activity Dashboard

Performance

• Cache recommendations (5-15 min TTL) to reduce API calls
• Lazy-load recommendation zones below the fold
• Limit number of products per zone (8-12 typical)

Privacy and Consent

• Respect customer privacy preferences (GDPR, CCPA)
• Allow opt-out from personalized recommendations
• Implement clear data retention and right-to-be-forgotten policies

Rollout Strategy

• Start with one high-traffic zone (homepage); monitor 2-4 weeks
• A/B test Einstein vs. manual curation or top-sellers fallback
• Track CTR, conversion rate, revenue attribution per zone
• Gradually expand to PDP, category, cart after proven ROI

Fetch the Einstein Recommendations API reference and Data Cloud connector docs for exact configuration parameters and SDK versions before implementing.