Overview
Data 360 Fundamentals has only two objectives but carries 20% of the exam, so each one is tested in depth. Data 360 is the current name for Salesforce Data Cloud, and this section focuses on the part of it that powers retrieval-augmented generation (RAG) for Agentforce.
The first objective covers the Agentforce Data Library, the setup shortcut that turns Salesforce Knowledge articles or uploaded files into searchable content for an agent. Creating a library builds the Data 360 pieces behind the scenes: a data stream, unstructured data objects, a search index and a retriever, which the standard question-answering action then uses. The second objective covers the foundations underneath: how text is split into chunks, how chunks are turned into vector embeddings and indexed, the difference between keyword, vector and hybrid search, and how retrievers, including individual and ensemble retrievers, select and return passages with filters and chosen fields.
How it's tested
Questions describe a knowledge source, an agent or prompt that answers from it, and a problem with the answers. You decide whether the Agentforce Data Library is enough, which search type suits the content, how chunking or filters should change, and whether an individual or ensemble retriever fits.
-
Library or custom setup
A team wants an agent to answer from Knowledge or files, and you decide between the Data Library and a custom Data 360 index and retriever.
-
Symptom to setting
Answers are vague, incomplete or wrong, and you trace the cause to chunk size, search type, filters or stale indexing.
-
Search type choice
Content contains exact identifiers, natural language or both, and you choose keyword, vector or hybrid search.
A manufacturer's service agent answers warranty questions from uploaded PDF manuals but often returns content for the wrong model when customers type part numbers such as AX-4410. Vector search matches meaning, not exact codes, so similar manuals score alike. Switching the index to hybrid search adds keyword matching on the part number while keeping semantic recall, and a retriever filter on product line narrows results further.
Objectives 2
Official objectives from the Salesforce outline, in outline order.
2.1 Explain the considerations of Agentforce Data Library and its concepts.Data Library sources, setup and limits.
- What it means
- The Agentforce Data Library (ADL) lets an admin point an agent or prompt at a knowledge source without building Data 360 components by hand. A library can use Salesforce Knowledge articles or uploaded files. When it is created, Salesforce provisions the data stream, unstructured data objects, search index and retriever needed for retrieval.
- What to understand
- Know what the library automates and what it does not. It requires Data 360 to be provisioned, it is linked to the agent through the standard answer-questions-with-knowledge action or used in prompt templates through its retriever, and changes to source content need the index to refresh before answers improve. For Knowledge, you choose which article fields are indexed. Custom chunking strategies, complex filters, multiple sources combined in one ensemble or non-supported sources call for building indexes and retrievers directly in Data 360.
- How it shows up
- A team wants an agent to answer from help articles or PDFs quickly, and you decide whether the Data Library meets the need and what it creates.
- Common confusion
- Thinking the Data Library stores content outside Data 360, when it actually builds standard Data 360 components.
2.2 Explain foundational concepts of Data 360 such as chunking, indexing, and retrievers.How content becomes searchable for RAG.
- What it means
- Unstructured content is ingested into Data 360 as unstructured data lake and model objects. A search index splits each document into chunks, converts chunks into vector embeddings with an embedding model, and stores them for search. A retriever is the configured query layer that agents and prompt templates call to fetch the most relevant chunks.
- What to understand
- Chunk size involves a trade-off: small chunks are precise but can lose context, while large chunks keep context but dilute relevance. Vector search finds meaning-similar text; keyword search finds exact terms; hybrid search combines both and suits content with codes, names and natural language. Individual retrievers query one index, while ensemble retrievers combine results from several indexes. Retriever settings include prefilters on fields, the number of results and the fields returned to the prompt.
- How it shows up
- Retrieval returns irrelevant or partial passages, and you choose the chunking, search type or retriever change that fixes it.
- Common confusion
- Assuming larger chunks always improve answers, when they often reduce precision and waste prompt space.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Data 360
Salesforce's data platform, formerly Data Cloud.
- Remember
- It stores the indexes, retrievers and audit data Agentforce uses.
- Tested as
- Required for Data Library, retrievers and audit trail.
- Trap
- Using the old Data Cloud name as if it were a separate product.
-
Agentforce Data Library
Setup feature that turns Knowledge or files into an agent-ready knowledge source.
- Remember
- Automatically provisions a data stream, index and retriever.
- Tested as
- Fast-start scenarios for knowledge answers.
- Trap
- Expecting full custom chunking control.
-
Retrieval-augmented generation
Fetching relevant passages and adding them to the prompt before the model answers.
- Remember
- Grounds answers in current company content.
- Tested as
- Underlies every retrieval question.
- Trap
- Fine-tuning a model instead.
-
Chunking
Splitting documents into smaller passages for indexing.
- Remember
- Chunk size balances precision and context.
- Tested as
- Answers that miss or blur content.
- Trap
- Assuming bigger chunks are always better.
-
Vector embedding
Numeric representation of a chunk's meaning.
- Remember
- Enables semantic similarity search.
- Tested as
- Natural language queries.
- Trap
- Expecting it to match exact codes reliably.
-
Search index
Data 360 structure holding chunks, embeddings and metadata.
- Remember
- Must refresh after source changes.
- Tested as
- Stale answer scenarios.
- Trap
- Blaming the model for outdated results.
-
Keyword search
Matches exact terms.
- Remember
- Strong for IDs, product codes and names.
- Tested as
- Content full of identifiers.
- Trap
- Using it for paraphrased questions.
-
Vector search
Matches by meaning.
- Remember
- Strong for paraphrased natural language questions.
- Tested as
- FAQ and policy content.
- Trap
- Using it alone for part numbers.
-
Hybrid search
Combines keyword and vector scoring.
- Remember
- Best general choice for mixed content.
- Tested as
- Content with both codes and prose.
- Trap
- Assuming it removes the need for filters.
-
Individual retriever
Queries a single search index.
- Remember
- Simplest option for one source.
- Tested as
- Single knowledge base scenarios.
- Trap
- Building an ensemble for one index.
-
Ensemble retriever
Combines and ranks results from multiple indexes.
- Remember
- Use when answers span several sources.
- Tested as
- Agents that must search manuals and Knowledge together.
- Trap
- Creating separate agents per source.
-
Retriever filters
Conditions on metadata fields applied to search results.
- Remember
- Narrow results by product, language or region.
- Tested as
- Answers mixing up products or audiences.
- Trap
- Trying to fix this with prompt instructions alone.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Search types
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Keyword | Exact term matching | Precise on identifiers | Misses paraphrased questions |
| Vector | Semantic similarity | Handles wording differences | Confuses near-identical documents |
| Hybrid | Keyword plus vector scoring | Balanced for mixed content | Treating it as a cure for poor chunking |
Retrieval setup options
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Agentforce Data Library | Guided setup over Knowledge or files | Auto-builds index and retriever | Expecting custom chunking options |
| Custom index and individual retriever | Built in Data 360 for one source | Full control of chunking, fields and filters | Overbuilding a simple Knowledge use case |
| Ensemble retriever | Combines several indexes | Answers across sources | Using it with only one source |
Chunk size trade-off
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Small chunks | Short passages | High precision, less context | Answers lose surrounding detail |
| Large chunks | Long passages | More context, lower precision | Irrelevant text fills the prompt |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Model blame
- What candidates pick
- Switch to a larger model when answers are wrong.
- Why it looks right
- Better models seem to answer better.
- Why it's wrong
- If the retriever returns the wrong passages, any model will answer from them; the fix lies in chunking, search type or filters.
Clue to watch for The response quotes the wrong document.
-
Vector for everything
- What candidates pick
- Use vector search for all knowledge sources.
- Why it looks right
- Semantic search sounds most advanced.
- Why it's wrong
- Exact identifiers such as SKUs need keyword matching, which hybrid search provides alongside semantic recall.
Clue to watch for Queries contain codes, model numbers or names.
-
Bigger chunks
- What candidates pick
- Increase chunk size to improve answers.
- Why it looks right
- More context sounds helpful.
- Why it's wrong
- Large chunks dilute relevance scores and consume prompt space, so precise questions can retrieve broad, unfocused passages.
Clue to watch for Answers are vague or include unrelated content.
-
One agent per source
- What candidates pick
- Build a separate agent for each document set.
- Why it looks right
- Separation seems tidy.
- Why it's wrong
- An ensemble retriever can search several indexes for one agent, keeping a single conversation and topic design.
Clue to watch for Answers need content from more than one source.
-
Library does everything
- What candidates pick
- Rely on the Data Library for a source it does not support or for custom chunk rules.
- Why it looks right
- It is the easiest setup path.
- Why it's wrong
- The library covers supported sources with defaults; custom requirements call for building the index and retriever in Data 360.
Clue to watch for The scenario lists special chunking, filtering or source needs.
Answer framework
Use these steps when a scenario involves grounding agents or prompts in unstructured content.
- Identify the content source and confirm Data 360 is available, since the Data Library, indexes and retrievers all depend on it.
- Decide whether the Agentforce Data Library covers the source and requirements, or whether a custom Data 360 index and retriever are needed.
- Match the search type to the content: keyword for identifiers, vector for natural language, hybrid when both appear in queries.
- Tune chunking and retriever settings, including filters, result count and returned fields, based on the symptom described in the scenario.
- Use an ensemble retriever when answers span several indexes, then confirm the index refreshes after source content changes.
Practice question
Want more like this? The AI-201 practice test has questions for every domain.
Checklist
0 of 11
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
With only two objectives, depth matters more than breadth. Start with the Trailhead modules on Data 360 unstructured data and the Agentforce Data Library, then build a library in a Data 360-enabled org from a handful of Knowledge articles or PDFs. After setup, open Data 360 and find each component the library created: the data stream, unstructured objects, search index and retriever. Seeing them makes the abstract terms concrete.
Next, create a custom search index on the same content with a different chunking setting and hybrid search, and a retriever with a filter. Test both retrievers from a prompt template preview and compare the passages returned. Then combine two indexes in an ensemble retriever.
High-priority ideas are what the library automates, the search type trade-offs and retriever filters. Commonly underestimated are index refresh timing and returned fields. For practice questions, write down the symptom in plain words before reading the options, then ask whether the cause lies in source, chunking, search type, filtering or the prompt.
FAQ
How much weight does Data 360 Fundamentals carry on the AI-201 exam?
Data 360 Fundamentals is 20% of the exam per third-party copies of the current guide, around twelve scored questions as a planning estimate, built on two broad objectives.
Is Data 360 the same as Data Cloud?
Yes. Salesforce renamed Data Cloud to Data 360. Older study material and the previous exam outline still use the Data Cloud name.
Do I need to know Data 360 ingestion and identity resolution?
Only lightly. The objectives focus on unstructured content for retrieval: the Data Library, chunking, indexing, search types and retrievers, not the full Data 360 Consultant scope.
What is an ensemble retriever?
It is a retriever that queries several search indexes and combines the ranked results, so one agent or prompt can draw on multiple knowledge sources.
When should I choose hybrid search?
Choose it when users ask in natural language but also include exact identifiers such as product codes, names or error numbers that semantic matching alone may miss.
Does the Data Library need Data 360?
Yes. The library is a guided way to create Data 360 components, so Data 360 must be provisioned in the org.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 2 knowledge
You have covered the Agentforce Data Library, chunking, indexing, search types and retrievers. See whether you can trace retrieval symptoms to the right fix under time pressure.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the AI-201 AI tutor Get any concept explained your way