cross-cloud-data-deployment

# Cross-Cloud Data Deployment

Activate when an implementation spans multiple Salesforce clouds (Sales, Service, Marketing, Commerce, Data Cloud, Industries) and data must be consistent across them. Cross-cloud is less about "how do we integrate?" and more about "who owns which field, and how does the truth propagate?"

## Before Starting

- **Identify the system of record per entity.** Customer in Sales Cloud or Data Cloud? Product in Commerce or PIM? Case in Service or Industries? Write it down. Ambiguity here breaks everything downstream.
- **Draw the flows BEFORE picking tech.** A whiteboard diagram of data flowing between clouds is the artifact that drives integration selection, not the other way around.
- **Establish the shared key strategy early.** External IDs, Person IDs, Account IDs — whichever key ties clouds together must be chosen before data model finalization.

## Core Concepts

### Data Cloud as the hub

Data Cloud ingests from Sales / Service / Commerce / Marketing, performs identity resolution, builds a unified profile, and exposes back via Calculated Insights, Activations, and Zero Copy. It becomes the cross-cloud customer brain.

### Identity resolution

Matching rules (deterministic or probabilistic) collapse multiple Contact / Lead / Customer records across clouds into a Unified Individual. The rule set is per-org and evolves.

### Change Data Capture (CDC) vs Platform Events

CDC emits record-level field changes automatically; Platform Events are bespoke business events. For cross-cloud sync of CRUD, prefer CDC; for domain events (Order Shipped, Case Escalated), prefer Platform Events.

### Shared keys

External ID fields on each cloud point to a common reference. Pattern: every record carries the same `Global_Customer_Id__c` with External ID + Unique. Upserts become trivially safe.

## Common Patterns

### Pattern: Sales + Service co-resident; Marketing via Data Cloud

Sales and Service share one org (one Account, Contact). Data Cloud ingests from the org and from Marketing Cloud; Marketing activates segments back. Clean.

### Pattern: Sales + Commerce with Order propagation

Commerce creates Orders; OMS / Sales Cloud handles service. Use Platform Events for "Order Placed" rather than REST sync — decouples cadence.

### Pattern: Data Cloud Zero Copy for analytics, not operational

Zero Copy (Snowflake, BigQuery, Databricks) for analytics without duplication. NOT for operational sync — latency and semantics differ.

## Decision Guidance

| Situation | Recommended Approach | Reason |
|---|---|---|
| Shared customer identity across 3+ clouds | Data Cloud as hub | Native identity resolution |
| Sales ↔ Service single-org | Native relationships, no integration | Same org |
| Marketing engagement → Sales | Marketing Cloud Connect or Data Cloud Activation | Supported paths |
| Commerce order → Service case | Platform Events | Decoupled cadence |
| Analytics across clouds | Data Cloud + Zero Copy | No duplication |

## Recommended Workflow

1. Inventory the clouds in scope and the shared entities each owns.
2. Assign system-of-record per entity; document ownership explicitly.
3. Select shared keys; add External ID + Unique fields on each cloud.
4. Decide sync vs event per flow; pick CDC, Platform Events, Marketing Cloud Connect, or Data Cloud.
5. Design identity resolution rules in Data Cloud; validate with a two-cloud sample.
6. Plan the deployment order: data model first (shared keys), then integration, then activation.
7. Build an end-to-end smoke test from creation in cloud-A to arrival in cloud-B.

## Review Checklist

- [ ] System-of-record documented for every shared entity
- [ ] External ID + Unique present on every cross-cloud object
- [ ] Sync vs event decision made per flow
- [ ] Data Cloud identity resolution rules tested
- [ ] Deployment ordering respects key dependencies
- [ ] Error handling and replay designed for every flow
- [ ] Cross-cloud reporting strategy addressed

## Salesforce-Specific Gotchas

1. **Marketing Cloud Connect has its own identity model (Subscriber Key).** Mapping Subscriber Key to Data Cloud Unified Individual requires explicit rules.
2. **Data Cloud is not real-time for all flows.** Streaming ingestion + CI refresh can mean minutes of lag; do not architect operational flows around Data Cloud latency.
3. **CDC has order guarantees per record but not across records.** Cross-object consistency must come from a different mechanism.

## Output Artifacts

| Artifact | Description |
|---|---|
| Cross-cloud topology diagram | Clouds, flows, shared keys |
| System-of-record matrix | Entity × cloud with SoR marked |
| Shared key inventory | External IDs across clouds |
| Deployment ordering plan | Data model → integration → activation |

## Related Skills

- `architect/multi-cloud-architecture` — multi-cloud org strategy
- `integration/integration-pattern-selection` — mechanism choice
- `data/data-cloud-foundation` — Data Cloud specifics