Data Cloud and Apex for AI

AI is only as good as its data, which is why Data Cloud sits under Agentforce and Prompt Builder. Developer and architect interviews now ask how data gets into Data Cloud, how it becomes a unified customer profile, how unstructured content is made searchable for agents, and how to write Apex that AI features can rely on.

1. What is Salesforce Data Cloud?

Salesforce's data platform that brings together data from Salesforce orgs, websites, apps, data warehouses and other systems, harmonises it into a common data model, resolves identities into unified profiles, and makes that data available in real time to Agentforce, Prompt Builder, analytics, segmentation and automation — often without copying data, through zero-copy integrations with warehouses such as Snowflake and Databricks.

2. Explain data streams, DLOs and DMOs.

Data streams ingest data from a source. The raw data lands in Data Lake Objects (DLOs). You then map DLO fields to Data Model Objects (DMOs) — the harmonised, standard model (Individual, Contact Point Email, Sales Order and so on) that every feature uses. Good mapping is what makes data from different systems usable together.

3. What is identity resolution?

The process of matching records from different sources that belong to the same person — using match rules (exact or fuzzy name, email, phone) and reconciliation rules (which source wins for each field) — to build a unified individual profile. Agents can then see a customer's complete history, not fragments.

4. What are calculated and streaming insights?

Calculated insights are metrics computed over Data Cloud data, such as lifetime value or orders in the last 90 days. Streaming insights compute near real-time metrics over event data, such as page views in the last ten minutes. Both can trigger actions and enrich prompts and agents.

5. How do agents use unstructured data like PDFs and knowledge articles?

Data Cloud can ingest unstructured content, split it into chunks, create vector embeddings and build a search index (vector or hybrid). Retrievers query that index and pass the most relevant passages into prompt templates or agent actions — Salesforce's built-in implementation of retrieval-augmented generation (RAG).

6. What Apex practices matter for AI features?

Use with sharing and WITH USER_MODE so AI never sees data the user cannot; bulkify invocable methods; describe every invocable method and variable clearly; handle errors and return friendly messages instead of exceptions; keep callouts in Named Credentials; mind governor limits and use asynchronous Apex for long work; and write unit tests that mock callouts.

@IsTest
private class OrderStatusActionTest {
    @IsTest
    static void returnsFriendlyMessageForUnknownOrder() {
        OrderStatusAction.Request req = new OrderStatusAction.Request();
        req.orderNumber = 'DOES-NOT-EXIST';
        Test.startTest();
        List<OrderStatusAction.Result> res = OrderStatusAction.getStatus(new List<OrderStatusAction.Request>{ req });
        Test.stopTest();
        Assert.isTrue(res[0].message.contains('could not find'), 'Should explain the order was not found');
    }
}

7. How would you design a support agent grounded in Data Cloud?

Ingest cases, orders and knowledge articles into Data Cloud; resolve identities to unified customer profiles; index knowledge articles for retrieval; build Agentforce topics for order status, returns and FAQs; use Apex or Flow actions for transactions and a retriever for policy answers; add escalation to a human; and evaluate with a test set before going live on chat.

8. How do you keep AI features compliant with data-protection rules?

Collect and use only necessary data, respect consent captured in Data Cloud, apply field-level security and masking, set data retention policies, audit AI interactions, and give customers a clear way to reach a human. Align with regulations that apply to your users, such as India's DPDP Act or GDPR.