AI-901 — Microsoft Azure AI Fundamentals Exam Blueprint

How to Use This Exam Blueprint

Use this page as a practical study map for the Microsoft Azure AI Fundamentals (AI-901) exam. The goal is not to memorize every Azure product name in isolation. The goal is to recognize AI workload types, choose appropriate Azure AI services, understand responsible AI principles, and interpret common machine learning, computer vision, natural language, and generative AI scenarios.

For each topic area:

1. Read the readiness target.
2. Check whether you can explain the concept without notes.
3. Practice service-selection scenarios.
4. Review weak areas until you can justify your answer choices.

You are ready when you can answer “Which Azure AI capability fits this scenario, and why?” across machine learning, vision, language, speech, search, and generative AI examples.

AI-901 Readiness Areas at a Glance

Core AI Concepts Checklist

AI Workload Recognition

Check that you can identify these workload patterns quickly.

Can You Do This?

• Explain the difference between artificial intelligence, machine learning, and deep learning.
• Identify whether a scenario is classification, regression, clustering, anomaly detection, computer vision, NLP, speech, or generative AI.
• Explain why AI systems can produce incorrect, biased, or unsafe results.
• Describe why training data quality matters.
• Recognize when a prebuilt AI service is enough versus when a custom model may be needed.
• Explain why validation and testing are separate from training.
• Identify responsible AI risks in common workplace scenarios.
• Explain why monitoring is needed after an AI model or AI application is deployed.

Responsible AI Readiness

Microsoft exam scenarios often test whether you understand the practical purpose of responsible AI principles, not just the vocabulary.

Responsible AI Scenario Checks

Machine Learning Fundamentals Checklist

Key Concepts to Know

ML Task Selection Table

Evaluation Metric Readiness

You do not need to turn AI-901 into an advanced statistics exam, but you should understand what common metrics are trying to measure.

Useful classification formulas:

\[ \text{Precision} = \frac{\text{True Positives}}{\text{True Positives} + \text{False Positives}} \]\[ \text{Recall} = \frac{\text{True Positives}}{\text{True Positives} + \text{False Negatives}} \]

Can You Do This?

• Given a scenario, identify whether it is classification, regression, clustering, or anomaly detection.
• Explain why labeled data is needed for supervised learning.
• Explain why clustering can work without labels.
• Describe the purpose of training, validation, and test datasets.
• Identify overfitting from a simple description.
• Choose an appropriate metric for a classification or regression problem.
• Explain why a model may need retraining when real-world data changes.
• Distinguish model training from model inferencing.

Azure Machine Learning Readiness

Azure Machine Learning is the primary Azure platform for building, training, evaluating, managing, and deploying custom machine learning solutions.

Azure Machine Learning Lifecycle Checklist

• Define the business problem.
• Collect and prepare data.
• Split data for training and evaluation.
• Train one or more models.
• Evaluate model performance.
• Register or manage the selected model.
• Deploy the model for inference.
• Monitor performance and data drift.
• Retrain or update when performance declines.

Azure Machine Learning Scenario Cues

Computer Vision Checklist

Vision Capability Selection

Azure Vision Services to Review

Vision Scenario Checks

Can You Do This?

• Distinguish OCR from object detection.
• Distinguish image classification from object detection.
• Identify when a custom vision model is needed.
• Choose Document Intelligence for form and document field extraction.
• Recognize responsible AI concerns around face-related workloads.
• Explain why confidence scores may matter in image predictions.

Natural Language Processing Checklist

Natural language processing, or NLP, focuses on understanding, analyzing, translating, or generating human language.

NLP Capability Selection

Azure Language and Text Services

NLP Scenario Checks

Can You Do This?

• Identify the difference between sentiment analysis and key phrase extraction.
• Explain what an entity is in NLP.
• Distinguish intent recognition from entity extraction.
• Choose translation when the main task is language conversion.
• Choose summarization when the main task is shortening content while preserving meaning.
• Recognize when a chatbot needs language understanding, knowledge retrieval, or generative AI.

Speech and Translation Checklist

Can You Do This?

• Distinguish speech-to-text from text-to-speech.
• Choose Translator for text translation.
• Choose Speech capabilities for spoken audio.
• Explain why transcription may be followed by NLP or summarization.
• Identify when a voice bot needs multiple services working together.

Generative AI Checklist

Generative AI is a major AI-901 readiness area. Focus on concepts, service selection, and safe use.

Generative AI Concepts

Azure Generative AI Capabilities to Recognize

Prompt and Grounding Readiness

Generative AI Scenario Checks

Can You Do This?

• Explain what a prompt is.
• Explain why generative AI may produce hallucinations.
• Distinguish generative AI from traditional classification or regression.
• Explain what grounding does.
• Recognize a retrieval-augmented generation scenario.
• Explain why content safety matters.
• Identify when human review should remain in the workflow.
• Choose Azure OpenAI Service or Azure AI Foundry for generative AI application scenarios.
• Choose Azure AI Search when the scenario emphasizes search, indexing, retrieval, or grounding over documents.

Azure AI Service Selection Matrix

Use this table for final review. Many AI-901 questions are essentially service-selection questions.

Decision-Point Review

Service Choice Flow

    flowchart TD
	    A[What is the main workload?] --> B{Predict from structured data?}
	    B -->|Yes| C[Azure Machine Learning]
	    B -->|No| D{Image or document?}
	    D -->|Image analysis| E[Azure AI Vision / Custom Vision]
	    D -->|Forms or invoices| F[Azure AI Document Intelligence]
	    D -->|No| G{Text or language?}
	    G -->|Analyze text| H[Azure AI Language]
	    G -->|Translate text| I[Azure AI Translator]
	    G -->|No| J{Speech audio?}
	    J -->|Yes| K[Azure AI Speech]
	    J -->|No| L{Generate content or answer questions?}
	    L -->|Yes| M[Azure OpenAI Service / Azure AI Foundry]
	    L -->|Needs document retrieval| N[Azure AI Search + generative AI]

Scenario Decision Table

Configuration and Integration Basics

AI-901 is not a deep implementation exam, but you should recognize the basic building blocks of Azure AI solutions.

Secure-Use Checklist

• Do not expose keys in client-side code.
• Use managed identities or secure secret storage when appropriate.
• Limit who can create, modify, or call AI resources.
• Protect sensitive training, prompt, and response data.
• Monitor usage and errors.
• Review generated output for high-impact decisions.
• Apply content filtering or safety controls where user-generated input is involved.
• Understand that responsible AI is both a design and operations concern.

Common Weak Areas and Traps

High-Value “Can You Do This?” Checklist

Use this as a near-final readiness test.

AI Concepts

• Define AI, machine learning, and deep learning in simple terms.
• Identify common AI workload types from short scenarios.
• Explain the difference between predictive, analytical, and generative AI tasks.
• Recognize where human oversight is important.

Machine Learning

• Distinguish supervised and unsupervised learning.
• Identify classification, regression, clustering, and anomaly detection.
• Explain features, labels, training data, validation data, and test data.
• Interpret basic classification performance language.
• Explain overfitting and underfitting at a conceptual level.
• Describe the Azure Machine Learning lifecycle.

Vision

• Choose OCR for text extraction from images.
• Choose Document Intelligence for structured forms.
• Choose image classification for whole-image categories.
• Choose object detection when item location matters.
• Recognize when Custom Vision is appropriate.
• Identify responsible-use concerns for face-related scenarios.

Language, Speech, and Translation

• Choose sentiment analysis for opinion detection.
• Choose key phrase extraction for important terms.
• Choose named entity recognition for names, dates, places, and organizations.
• Choose conversational language understanding for user intent.
• Choose speech-to-text for transcription.
• Choose text-to-speech for spoken output.
• Choose Translator for text translation.
• Identify when a bot requires several AI services together.

Generative AI

• Explain prompts, completions, tokens, and context at a basic level.
• Explain why generated answers may be inaccurate.
• Choose grounding/RAG for answers based on private documents.
• Recognize embeddings and vector search as similarity-search concepts.
• Identify when content safety controls are needed.
• Explain how Azure OpenAI Service, Azure AI Foundry, and Azure AI Search can work together.

Azure Service Selection

• Map each common AI workload to a likely Azure service.
• Explain when a prebuilt AI service is preferable.
• Explain when Azure Machine Learning is more appropriate.
• Identify basic authentication and endpoint concepts.
• Recognize monitoring and governance needs for deployed AI.

Final-Week Review Plan

5 to 7 Days Out

• Review every topic area in this checklist once.
• Build a one-page service-selection map.
• Drill classification vs. regression vs. clustering scenarios.
• Review responsible AI principles with examples.
• Practice identifying Azure AI services from business requirements.

3 to 4 Days Out

• Focus on weak service-selection areas.
• Review Azure Machine Learning lifecycle terms.
• Review computer vision distinctions: OCR, image classification, object detection, Document Intelligence.
• Review NLP distinctions: sentiment, key phrases, entities, intent, summarization, Q&A.
• Review generative AI concepts: prompts, grounding, RAG, embeddings, hallucination, content safety.

1 to 2 Days Out

• Stop trying to memorize obscure product details.
• Practice mixed scenarios under time pressure.
• For each wrong answer, write the reason the correct service fits better.
• Review responsible AI one more time.
• Review basic Azure resource, endpoint, key, identity, and monitoring concepts.
• Sleep and keep the final review lightweight.

Quick Final Drill

Answer these without notes:

1. A retailer wants to predict next month’s revenue. What ML task is this?
2. A bank wants to detect unusual transactions. What workload type is this?
3. A manufacturer wants to find cracks in product photos. Which vision approach fits?
4. A company wants to extract totals from invoices. Which Azure capability fits?
5. A support team wants to determine whether comments are positive or negative. Which NLP capability fits?
6. A call center wants searchable call transcripts. Which capabilities are involved?
7. A chatbot must answer only from internal policy documents. What design pattern reduces unsupported answers?
8. A generated answer sounds confident but is false. What is this risk called?
9. A model performs well on training data but poorly on new data. What problem does that suggest?
10. A team needs to build and deploy a custom ML model. Which Azure platform is the best fit?

If any answer takes more than a few seconds, return to that section and practice similar scenarios.

Practical Next Step

Use this checklist to guide one focused practice session per topic area. After each practice set, tag every missed question as one of these: concept gap, service-selection mistake, scenario wording trap, or responsible AI judgment issue. Then review only the matching section before taking another mixed AI-901 practice set.