Salesforce Data Cloud is a Customer Data Platform — not a warehouse, not a data lake, not a CRM table. Its job is to take every source-system record that mentions a customer, unify them into one harmonised profile, resolve identities across sources, and expose the result as queryable, indexable, RAG-ready data. It is the load-bearing piece behind AgentForce: without Data Cloud, AgentForce agents only see what's inside the local Salesforce org. With Data Cloud, they reason over a single picture assembled from every system that's been connected.
Salesforce Data Cloud is Salesforce's Customer Data Platform. Three properties matter:
"Customer Data Platform" is the analyst category. Treasure Data, Tealium, Segment, and Adobe Real-Time CDP are peers. Data Cloud's positioning is that it's the CDP that already lives inside Salesforce metadata — sharing rules, fields, and object relationships are inherited rather than reconstructed.
Data Cloud is also the substrate for AgentForce's retrieval-augmented generation. The Atlas Reasoning Engine grounds prompts in Data Cloud's harmonised profile, which is why a Service Agent can answer "what's this customer's recent order history?" without the SI team building a single integration. The integration is upstream — once SAP-or-equivalent is connected to Data Cloud, every AgentForce agent inherits it.
Salesforce's product naming has moved through several phases. Customer Data Platform → Genie → Data Cloud → "Data 360" (in some 2026 materials). The technical product is the same; the marketing emphasis shifts between "data" and "AI grounding". Don't be confused by the rebrands — if the conversation is about real-time customer profile, unified data, AgentForce grounding, you're talking about Data Cloud.
Every CDP runs roughly the same four stages. Data Cloud's particular shape is "real-time at every stage and tied to Salesforce metadata at the output side".
Connectors pull records from source systems in batch or stream. Native connectors for major SaaS, JDBC for databases, MuleSoft for the rest. Sub-second ingestion latency for streaming sources.
Each source schema is mapped onto Data Cloud's standard Data Model (Individual, Account, Engagement, Order). Field-level transforms run during ingest. This is the model-mapping work that determines how good the unified profile becomes.
Deterministic rules (email match, phone match) + probabilistic rules (name + DOB + address proximity) collapse records that point to the same person into one profile. The matching ruleset is configurable; the output is the Unified Individual ID.
The unified profile feeds downstream — AgentForce (RAG), Marketing Cloud (segmentation), Tableau (analytics), external systems via reverse ETL. The same profile is queryable, searchable, and pushable.
Data Cloud's semantic-search index handles unstructured data — knowledge articles, support transcripts, contract PDFs — alongside structured records. AgentForce's Atlas Reasoning Engine can retrieve from both in the same query. This is the AgentForce capability claim that comes directly from a Data Cloud feature: the agent doesn't know whether the answer is in a database row or paragraph 4 of a knowledge article. The semantic-search layer abstracts both.
Data Cloud lives in the same neighbourhood as warehouses, lakes, and other CDPs, but solves a different problem. Quick framing:
| Layer | Job | Examples |
|---|---|---|
| Warehouse / Lake | Analytical storage; "the source of truth for the analysts" | Snowflake, BigQuery, Databricks, Postgres |
| CDP | Unified, real-time customer profile across systems | Data Cloud, Segment, Tealium, Treasure Data |
| Reverse ETL | Push warehouse data back into operational tools | Hightouch, Census |
| CRM table | The system-of-record for sales / service workflow | Salesforce SObjects, HubSpot CRM |
You can build a unified customer profile in Snowflake or BigQuery — teams do, and it works. The Data Cloud arguments are: (1) real-time ingestion and identity resolution are not warehouse strengths; (2) the activation side — pushing profile changes into AgentForce, Marketing Cloud, Service Cloud — is native; (3) the security model inherits Salesforce sharing rules, so a sales rep only sees the unified profile for accounts they own. For a Salesforce-heavy enterprise, Data Cloud is structurally aligned. For everyone else, a warehouse + reverse-ETL stack is often the cheaper path.
POPIA Section 19 evidence. Data Cloud's role in a unified customer profile is exactly the activity POPIA Section 19 ("appropriate, reasonable technical and organisational measures") asks you to control. The platform's sharing-rule inheritance means the unified profile respects who-can-see-what at the source; the audit log captures access; identity-resolution rules are configurable and reviewable. The compliance argument is well-supported — provided the residency question (next) is also answered.
Hyperforce residency. Data Cloud runs on Hyperforce. No SA datacenter as of mid-2026; SA orgs typically land in Frankfurt or Dublin. For Section 72 cross-border-transfer compliance, the Salesforce-provided SCC equivalents need to be in the procurement package. If POPIA-strict residency is a hard requirement, Data Cloud is not the right answer until Hyperforce SA lands — a self-hosted CDP in Cape Town would be the alternative.
The "single view of customer" reality. Many SA businesses run customer data across SAP-or-Oracle (ERP), Salesforce (CRM), an SMS aggregator (campaigns), and a legacy support tool. Data Cloud's unification claim is exactly what those businesses need — but it's a multi-month integration project, not a configuration toggle. The cost-benefit conversation needs to be honest about the SI hours, not just the licence.
Cost model. Data Cloud is priced on credits consumed during ingestion, processing, and activation — separate from AgentForce's Flex Credits. For SA businesses budgeting in ZAR against USD-priced credits, the same FX-sensitivity exercise applies as it does for AgentForce. Pre-purchase the credit envelope, monitor consumption, treat it like cloud spend.