Salesforce Data Cloud Consultant Mock Exam: Key Concepts by Domain

What Data Cloud Is: Data Cloud (Data 360) is Salesforce's customer data platform: it connects sources, harmonizes them to the Customer 360 Data Model, resolves identities into unified profiles, and activates those profiles to marketing, sales, service, analytics, and AI. It is not a records archive, a bulk email sender, an encryption service, or a Salesforce-to-Salesforce sync tool.

Know the Layers: A data stream ingests a source into a data lake object (DLO, raw storage), which is mapped to a data model object (DMO, harmonized schema); an identity resolution ruleset produces Unified Individual profiles; calculated insights, segments, activations, data actions, and data graphs consume the result. Many questions hinge on telling these layers apart.

Rename, Same Product: Data Cloud is now officially Data 360 and the credential is the Data 360 Consultant. Exam wording may use either name; the architecture, objects, APIs, and permission sets are unchanged.

Grounding Layer for AI: Unified, harmonized profiles are the grounding data for Einstein predictions and Agentforce generative experiences, so churn scores, next-best actions, and agent answers reflect the full customer picture. Data Cloud does not validate ingested rows in place of stream configuration and does not author email templates.

Predictive vs Generative: Einstein Studio hosts predictive models, out of the box or Bring Your Own Model from SageMaker, Vertex AI, or Databricks, that score Data Cloud records in place. Generative use cases rely on the vector database, search indexes, and retrieval augmented generation (RAG) to pull relevant passages into Agentforce and Prompt Builder prompts.

Start With Use Cases: The first phase of an engagement defines the business use cases, then identifies the data sources they need and assesses the quality of that data. Data spaces, migrations, and insights are implementation steps that only make sense once use cases and sources are known.

Recognize the Representative Scenario: The classic fit is many touchpoints (web, service, field sales, marketing, commerce) harmonized into one view that supports reporting, segmentation, and activation; industry flavors include the Unified Health Score in Data Cloud for Health. Payroll, point of sale, versioned document archiving, and a standalone consent repository are not Data Cloud use cases.

Trust-Based First-Party Data: A trust-based first-party data asset is collected transparently from people who consent to its use, can control it, and receive clear value in return. It is not a dataset to sell on a marketplace, a legal opt-in minimum, or competitive intelligence gathered through surveys.

Minimization & Purpose Limitation: Challenge whether sensitive attributes (age, gender, ethnicity) are needed for the agreed use cases before ingesting them; do not collect everything, grant blanket executive access, or pool sensitive data for convenience. Consent and right-to-be-forgotten requests are honored through the Consent API.

Match the Permission Set to the Role: Data Cloud Architect (formerly Data Cloud Admin) has full access including Data Cloud Setup; Activation Manager adds full control of activation targets; Activation Specialist (formerly Marketing Specialist) builds, publishes, and activates segments while targets and data streams stay view-only; Data Cloud User is read-only and cannot create segments.

The Integration User Matters: The Salesforce CRM connector reads through an integration user and respects its object and field-level security, so a new custom field that user cannot read never appears in the data stream editor. Grant field access before looking anywhere else.

Data Spaces Partition One Org: Data spaces logically separate data, metadata, and processes (data streams, DMOs, segments, insights, activations) per brand, region, or business unit inside a single Data Cloud org, with access granted per space. Permission sets alone do not partition data, and companion connections relate orgs rather than segregate units.

Right to Be Forgotten via the Consent API: Deletion requests are submitted through the Consent API for source Individual IDs; Data Cloud removes the Individual and the related DLO and DMO records, then reprocesses the request at 30, 60, and 90 days to catch reingested data. Processing takes hours, requests are scoped to Data Cloud (each connected cloud needs its own, Marketing Cloud Engagement about 48 hours later), Profile Explorer is read-only, and deleting in the source does not purge what Data Cloud already holds.

Sandbox Refresh Copies Metadata, Not Data: A refresh copies data streams, calculated insights, segments, and identity resolution rulesets from production and overwrites the sandbox configuration, but no records: streams must run again, external connectors must be re-authorized, and the first identity resolution run is started manually. The instance is not deleted and mappings do not revert to draft.

Data Kits Move Configuration: Data kits package Data Cloud metadata (data streams, DMOs, insights, segments) for deployment between orgs. They move configuration, not data, so they are the wrong answer for reshaping records or keeping live segment criteria in sync.

Order the Schedules: A segment can only publish what was true at its last refresh, so ingestion must finish before the segment refreshes and any calculated insight it filters on must run before the segment. The dependable order is data stream refresh, identity resolution, calculated insight, segment, activation; a mismatch adds one full cycle of staleness.

Automate on Data Stream Records: Data stream records live in the Salesforce org, so a record-triggered flow on the Data Stream object can email, post to Slack, and open a Case the moment a refresh status moves to failed. Data actions fire on DMO events, not on stream metadata, and reports or saved queries are passive.

Diagnose a Failed Refresh: A NO FILE FOUND error on a previously working Amazon S3 stream means no file matches the configured directory and name pattern, or the IAM user behind the connection lost ListBucket and GetObject rights; scheduled refreshes never run as the person who built the stream. Use Data Explorer to inspect the records of a DLO, DMO, or calculated insight within a data space.

Pick the Right Category: Profile is for any entity you segment or unify (individuals, accounts, employees) and is the only category that can be the Segment On entity; Engagement is for time-series events and needs an immutable Event Time Field such as Created Date, never Last Modified Date; Other holds reference data that is never the primary segmentation entity.

Data Types Are Literal: Phone, Email, and URL are stored as text and never validated or reformatted, so E.164 standardization needs a formula field mapped to Formatted E164 Phone Number. Keep leading zeros in identifiers such as purchase order numbers with Text, not Number, and never use a phone number as a primary key.

Refresh Modes & File Patterns: For cloud storage streams, a wildcard file name (transactions_*.csv) picks up each date-stamped file and Upsert keeps the rows already ingested; Full Refresh replaces the whole DLO with the latest file and loses history. Data Cloud never deletes source files; a bucket lifecycle rule does.

Salesforce CRM Connector: Batch streams start with a full refresh and then run incremental refreshes about every 10 minutes (interval not configurable, Refresh Now on demand, periodic full refresh off by default). Streaming mode reflects supported object and field changes with minimal latency, but adding a formula field moves the whole stream back to batch.

Ingestion API: The bulk endpoint is built for large historical loads and the streaming endpoint for small continuous updates processed as micro-batches in near real time, which is what changing inventory or app events need. Best practice is a bulk backfill followed by streaming.

Product Connectors: The B2C Commerce Order bundle is the only way to create B2C Commerce streams and backfills about 30 days, so older history needs a custom extract to cloud storage mapped to the same DMOs. The Marketing Cloud Engagement connector creates scheduled streams from data extensions with no code. The Web and Mobile Application Connector generates a Tenant Specific Endpoint for the SDKs, describes events through an uploaded schema whose updates are additive, and its streams must be deleted before the connector.

Formula Fields at Ingestion: Data stream formulas transform values row by row as they land: cast a datetime to a date, standardize a format, or build a composite primary key with CONCAT (COMBINE is not a function; COALESCE returns the first non-null value and CAST converts types). They do not deduplicate rows, sequence files, or edit rulesets.

Batch vs Streaming Data Transforms: Batch data transforms read existing DLOs and write new ones: union several product streams into one cleaned object with proper case and null defaults, or split a combined loyalty record into separate hotel and airline DLOs without touching the source. Streaming data transforms reshape events in flight; neither kind evaluates a condition or triggers an action.

File Federation vs Query Federation: File federation lets Data Cloud compute read Iceberg or Parquet files directly in object storage (Amazon S3, Azure Blob, Snowflake-referenced tables) using a manifest that lists file paths and metadata; it is the recommended fit for large open-table history. Query federation pushes each query to the external warehouse (Snowflake, Redshift, BigQuery, Databricks) and bills its compute.

Acceleration: Enable cached acceleration on a query-federated object when the same analytical queries repeat and there is no real-time requirement: Data Cloud stores and incrementally refreshes a copy, so repeat queries stop paying for external compute.

Clean Rooms: Data Cloud clean rooms are built on Zero Copy: each party's data stays in its own environment and only approved, aggregated query results are shared, never raw rows or decrypted identifiers. No shared bucket or broad permission grant is involved.

Map to the Customer 360 Data Model: Harmonization maps DLO fields to standard DMOs: email addresses from any source, standard or custom, go to Contact Point Email, orders to Sales Order, and shared external identifiers (tax ID, loyalty number) to Party Identification with the same Identification Name and type across sources. Extend a standard DMO with a custom attribute rather than creating a custom DMO, which forfeits standard relationships, insights, and bundles.

Profile Category and Segment On: A custom DMO must be created in the Profile category to serve as the Segment On entity; Engagement and Other objects can only be reached as related attributes. Segments can be defined before rows arrive, and no relationship to Individual is required to segment on a custom Profile DMO.

Data Graphs: A data graph is a materialized, denormalized view assembled from DMOs and their relationships for low-latency lookups by Agentforce and Lightning components. DLOs must be mapped to DMOs first; no transform, data action, or activation step is involved.

Match Rules Widen, Attributes Narrow: A unified profile forms when any one match rule is satisfied, so adding rules (normalized email, name plus phone, Party Identification) raises the consolidation rate while adding attributes to an existing rule makes it stricter. Fuzzy matching is for names; exact shared identifiers belong in a Party Identification match rule.

Reconciliation Rules Pick the Winner: Last Updated, Most Frequent, or Source Priority (rank the contributing DLOs) decide which conflicting value lands on a Unified Individual attribute, and Ignore Empty Value skips blanks from the winning source. They apply to Unified Individual attributes only: contact points keep every distinct value and use source priority order at activation time.

Incremental Processing & Deletion: After the first full build, identity resolution reprocesses only new or changed source profiles and skips a run with no changes; there is no unification score threshold or data quality gate. Deleting a ruleset permanently removes its unified DMOs (Unified Individual, Unified Link Individual, unified contact points) and their dependencies, breaking dependent segments, insights, and activations, while source Individuals and DLOs stay intact.

Profile Explorer vs Query Editor: Profile Explorer shows one unified profile with its linked source records, contact points, and insights; Query Editor runs ad hoc SQL across Unified Individual, Unified Link Individual, and custom DMOs to validate joins and matching at scale. Data Explorer browses one object at a time without joins, and metadata search only locates objects.

Harmonization vs Identity Resolution: Harmonization is mapping sources to a common data model; identity resolution applies match and reconciliation rules to produce Unified Individual or Unified Account profiles. "Data consolidation" and "data cleansing" are distractors, not Data Cloud features.

Anatomy and Scheduling: Every calculated insight needs at least one measure and one dimension; joins, functions, and filters are optional. Insights run in batch on a configurable schedule, so Salesforce reports show the last processed result, and they must run after the data stream refresh and identity resolution and before the segments that filter on them.

Join Through Unified Link Individual: Lifetime value per unified profile joins Unified Individual to Unified Link Individual to Sales Order, because Sales Order carries the source Individual ID and only the link object maps it to the unified ID. An insight appears in segmentation only for a Segment On object that one of its dimensions relates to.

Metrics on Metrics & Long History: Calculated insights can reference other insights (channel revenue feeding total lifetime value), and a scheduled insight plus a segment is the pattern for multi-year criteria such as deposits over five years. The Segment Membership DMO records who belonged to which segment and when, for historical membership analysis.

Real-Time Detection: Streaming insights aggregate web, mobile, and transaction events over rolling time windows as they arrive, and a paired data action fires a flow, webhook, or platform event when a threshold is crossed (two careers-page views in 24 hours, abnormal daily transaction volume). Calculated insights only recalculate on a schedule, and transforms cannot evaluate conditions.

Insight vs Transform: Insights (calculated or streaming) produce metrics; data transforms (batch or streaming) reshape data. When a question asks for a metric, an alert, or a segment attribute, the answer is an insight; when it asks to clean, split, or union records, the answer is a transform.

Data Shares (Bring Your Own Lake): Data Shares publish DMOs, DLOs, and insights to Snowflake, Databricks, BigQuery, or Redshift as live zero-copy tables, with no extract-and-load pipeline. Data bundles are inbound ingestion packages from Salesforce clouds, the opposite direction.

Profile API vs Query API: The Profile API returns a unified profile's attributes, related engagement (orders, cases), and calculated insights by key with low latency, which is what portals and Lightning web components need; the Query API runs analytical SQL with higher latency. The Ingestion API writes data and the Metadata API only describes structure.

Surface Data in CRM: Developers pair the Profile API (typically through Apex) with a Lightning web component on the record page; the standard Data Cloud related list and unified profile components are the no-code route. Data actions emit events rather than render UI, and data share targets expose tables to external platforms.

Report on Data Cloud Objects: Segments and activations are Salesforce records, so a report plus a dashboard lists each segment's last publish date and status; reports built on calculated insights show values as of the last processing run. The semantic data model (semantic layer) defines calculated fields evaluated at query time for Tableau Next, so users slice Net Profit by any dimension and always see current data.

Predictive & Generative AI: Einstein Studio Bring Your Own Model registers a SageMaker, Vertex AI, or Databricks model that scores Data Cloud records in place and stores results as insights for segments, flows, and activations. The vector database and search indexes power retrieval augmented generation that grounds Agentforce and Prompt Builder in unstructured content.

Segment On and Filter Types: The Segment On object defines what a member is: Profile DMOs such as Unified Individual, Individual, or Account, or an Engagement DMO such as Sales Order when one row per order is required. Filters come from direct attributes, related attributes, and calculated insights; unmapped data stream fields and streaming insights are not available.

Container Paths & Complexity: When a related object can be reached through several paths, the chosen path changes the population without an error; Data Cloud neither suggests alternatives nor substitutes the optimal path. "Segment is too complex" is resolved by moving stacked filters and nested operators into a calculated insight and combining text values into a single Is In operator.

Value Suggestion: Enabled per text attribute on a DMO (up to 500 per org), value suggestion offers up to 1,000 frequent values as a pick list in the segmentation canvas. Suggestions can take up to 24 hours to appear, work on direct and one-to-one or many-to-one related attributes, and need no special permission set.

Nested & Waterfall Segments: Nest an existing segment as an inclusion or exclusion filter so a compliance exclusion list maintained once flows to every brand segment on its next refresh, and combine containers with OR for unions such as Platinum members or spend above 5,000 USD. Waterfall segments carve mutually exclusive priority groups, not unions; there is no reusable container block.

Publish Schedules: Standard segments publish every 12 or 24 hours and rapid publish every 1 or 4 hours; neither is immediate, so event-driven reactions belong to streaming insights with data actions. Monitor last publish date and status with a Salesforce report on the segment object.

Contact Points & Related Attributes: An email activation to Marketing Cloud Engagement drops members without a Contact Point Email, which is why activated counts fall below the segment population. Personalization data travels as related attributes (Fund Name filtered on Fund Type equals Mutual Fund) or calculated insight attributes (RFM scores) that land in the activation data extension.

File Storage Targets: Amazon S3 targets authenticate with an IAM user's access key and secret key. Each run writes .csv (or Parquet) payload files, one row per activated individual with the selected attributes, then a .json metadata file that signals completion; files are named {segmentName}_{activationName}_{timestamp}, so a Campaign ID goes in the segment or activation name.

Data Cloud Activation Target: Publishes segment membership into the Audience DMO inside Data Cloud so another system can read it through the Query API, with no connection to an external platform. Marketing Cloud Engagement, Personalization, and external platform targets all push the audience out instead.

Data Actions: A data action watches a DMO for records that meet a condition and publishes a platform event or calls a flow within moments: update a CRM field, or inject a platinum customer into a Marketing Cloud Engagement journey the instant a streaming insight crosses its threshold. Data share targets, rapid publish, and streaming transforms are the standard distractors.

Real-Time in CRM: Data actions plus Lightning web components (or the standard unified profile and related list components) bring Data Cloud into Salesforce record pages; the Sales and Service starter bundles move CRM data into Data Cloud on a schedule, the opposite direction. "Data model triggers" is not a Data Cloud feature.

Grounding Agentforce: Retrieval augmented generation over Data Cloud search indexes lets Agentforce answer from internal policies, documentation, and case history instead of generic responses. Disabling embeddings or forcing every document into structured records removes the grounding rather than improving it.