[object Object]

The Architecture

Salesforce Data 360 (formerly Data Cloud) sits at the center as the unified profile, segmentation, and activation layer. Ingestion comes from three classes of source: data warehouses via zero-copy federation (Snowflake, Databricks, BigQuery), application sources via APIs and connectors (Marketing Cloud, Service Cloud, HubSpot, ServiceNow), and streaming sources (event collectors, mobile/web SDKs, third-party CDP webhooks). Identity resolution unifies profiles across sources using deterministic keys (email, phone, account ID) and probabilistic stitching where deterministic keys fail. Activation pushes computed segments and per-profile attributes to Marketing Cloud, paid media platforms (Meta, Google, TikTok via Audience Activation), Slack, and external systems via webhooks or reverse ETL. Governance — consent, lineage, retention, residency — wraps the entire stack and is enforced by Data 360’s policy engine plus the source systems’ own controls.

The composable framing matters: Data 360 is the integration and decisioning layer, not the storage system. The warehouse remains source of truth for historical analytical data; Data 360 indexes, identity-resolves, and activates.

Zero-Copy Patterns

Snowflake, Databricks, BigQuery, and Iceberg-format lakes federate via zero-copy. Data stays in its source system; Data 360 issues federated queries against it and materializes only the projections needed for activation. Reduces duplication, governance surface, and storage cost. Trade-offs to plan for:

  • Query latency: federated query against Snowflake is slower than native Data 360 storage. Cache hot segments.
  • Source-side cost: federated queries consume Snowflake credits. Budget accordingly.
  • Schema drift: source-side schema changes break activation flows; instrument the contract.
  • Authorization: zero-copy queries must respect source-side row/column security.

Use zero-copy for tall, slow-changing analytical data (transactions, behavioral history) and native Data 360 storage for high-velocity profile attributes and real-time decisioning state.

Activation Surfaces

Segments and per-profile attributes flow out to:

  • Marketing Cloud Engagement (email, SMS, push) — native, low-latency.
  • Marketing Cloud Account Engagement (Pardot) for B2B journeys.
  • Paid media: Meta CAPI, Google Customer Match, LinkedIn Matched Audiences, TikTok Events API.
  • Slack via Slack Sales Elevate and direct webhook.
  • External systems via Reverse ETL (Hightouch, Census) or direct webhook.
  • Agentforce as grounding context for customer-facing agents.

Activation is decentralized; Data 360 is the unified decisioning layer. Each activation surface owns its own delivery semantics, retry, and reporting.

Governance Patterns

Consent flows through from collection point to activation. Each profile carries channel-level consent flags (email_marketing_consent, sms_marketing_consent, profiling_consent for AI use). Data 360 evaluates consent per activation; segments respect opt-out without per-team coding. Lineage is traceable from source field to activation event — required for GDPR Article 30, CPRA right-to-know, and AI Act technical files.

Retention policies propagate. Setting a 24-month retention on a source attribute should cascade through derived segments and historical activation logs. Modern composable CDP isn’t “data lake plus marketing” — it’s governed, identity-resolved, activation-ready data with a unified policy plane.

Implementation Sequence

  1. Define identity strategy first — which keys, what stitching tolerance.
  2. Stand up zero-copy from the warehouse before native ingestion.
  3. Build one segment end-to-end (source → identity → segment → activation → reporting).
  4. Layer governance — consent and retention before broad rollout.
  5. Add agent grounding (Data 360 → Agentforce) once activation is stable.

Common Failure Modes

  • Ingesting everything; no activation discipline. Resulting in a graveyard data layer with high cost and no value.
  • Skipping identity-resolution validation — duplicate profiles silently fragment activation.
  • Activations without consent enforcement — fastest path to a regulator complaint.
  • Federated queries without caching — surprise Snowflake bill at month-end.

What to Do This Quarter

Pick one activation goal worth $100K+ in attributable lift. Build that one path end-to-end with full governance. Use the win to fund the broader buildout.

[object Object]
Share