So you are studying for the AI-102 and you keep getting tripped up on service selection questions.. If you searched for Azure AI service selection guide, you're in the right place Azure AI Vision or Document Intelligence? Language or Speech? Azure AI Search or just a database query? You are not alone. Service selection is the single most tested skill on the AI-102 exam, and getting it wrong means the difference between passing and failing. Let us fix that.
The Service Selection Problem
Microsoft offers a lot of AI services. Many of them overlap in capability. The AI-102 exam loves to present scenarios where two services could technically work, but only one is the best choice for the given constraints.
The key is to stop memorizing service descriptions and start thinking in terms of decision rules. Every service selection question can be solved by asking four questions:
- What is the input type? (text, image, audio, video, document, mixed)
- What is the output type? (extracted fields, classification, search results, generated text, translation, speech)
- What are the security and compliance requirements?
- Is customization needed, or is a prebuilt model sufficient?
Azure AI Service Decision Matrix
This table is your single most valuable study tool. Memorize the "use it for" column:
| Service | Use it for | Do not use it when |
|---|---|---|
| Azure AI Document Intelligence | Extracting structured fields from invoices, receipts, forms, IDs, tax documents | You only need OCR without field extraction |
| Azure AI Image Analysis / Vision | OCR, image captions, object detection, spatial analysis, image classification | You need document field extraction (use Document Intelligence) |
| Azure AI Content Understanding | Multimodal extraction across content types beyond simple forms | A prebuilt model from Document Intelligence suffices |
| Azure AI Content Safety | Detecting harmful text and images, content moderation | You need content understanding or classification |
| Azure AI Language | Sentiment analysis, key phrase extraction, NER, PII detection, summarization, conversational language | You need speech capabilities or document field extraction |
| Azure AI Speech | Speech-to-text, text-to-speech, speech translation, speaker identification | You need text-only translation (use Translator) |
| Azure AI Translator | Text translation, document translation | You need speech translation (use Speech) |
| Azure AI Search | Lexical search, semantic search, hybrid search, vector search, document enrichment pipelines | You only need simple database queries |
| Microsoft Foundry / Azure OpenAI | Generative AI, prompt engineering, RAG, model deployment, agentic solutions | Prebuilt models meet your needs |
| Azure AI Agent Service | Autonomous agents with tool use, grounding, guardrails | You only need a simple chatbot |
High-Yield Service Comparisons
These are the comparisons the AI-102 tests most frequently:
Document Intelligence vs Azure AI Vision
Use Document Intelligence when the requirement involves extracting specific fields from structured or semi-structured documents. Examples: invoice numbers from invoices, line items from receipts, dates from IDs, totals from tax forms.
Use Azure AI Vision when you need general OCR, image analysis, captions, object detection, or spatial understanding. Examples: reading text from a street sign in a photo, detecting objects in a video frame, generating alt-text for accessibility.
The exam frequently presents an invoice extraction scenario as a trap. Azure AI Vision can do OCR, but it does not provide the structured field extraction workflow that Document Intelligence does. Always choose Document Intelligence for form and document scenarios.
Azure AI Language vs Azure AI Speech
Use Language for text analysis: sentiment, key phrases, named entities, PII detection, summarization, language detection, conversational intent classification.
Use Speech when audio is involved: speech-to-text transcription, text-to-speech synthesis, real-time speech translation, speaker verification.
The confusion point: Speech includes speech-to-text, which outputs text. But if the input starts as text (not audio, always use Language for analysis).
Azure AI Translator vs Azure AI Speech Translation
Use Translator for translating text or documents between languages. It preserves document formatting and supports batch translation.
Use Speech for real-time speech translation: someone speaks in one language, the output is synthesized speech in another language. This is a live audio pipeline, not a text translation.
Azure AI Search vs Direct Database Queries
Use Azure AI Search when you need semantic search (understanding meaning, not just keywords), vector search (similarity matching over embeddings), hybrid search (combining lexical and semantic), or document enrichment pipelines with custom skills.
Use direct database queries when you are matching exact values, filtering on known columns, or aggregating data. Search is overkill for "find all orders where status equals Shipped."
RAG vs Fine-Tuning
Use RAG when the model needs access to enterprise data, when documents change frequently, when you want to cite sources, and when you want to control cost (fine-tuning is expensive).
Use fine-tuning when you need the model to adopt a specific tone, format, or domain language that prompting alone cannot achieve, and when you have a large, stable training dataset.
Security Decision Rules
Security questions follow predictable patterns:
| Requirement | Answer |
|---|---|
| Least privilege, avoid keys | Managed identity + RBAC |
| Private network only | Private endpoints, disable public access |
| Key storage | Azure Key Vault |
| Content moderation before Azure OpenAI | Azure AI Content Safety |
| Audit trail | Microsoft Purview, Azure Monitor logs |
| Data residency | Region-specific resource deployment |
Generative AI Decision Rules
The generative AI domain is the newest and most heavily tested:
| Scenario | Answer |
|---|---|
| Ground model on enterprise data | RAG with Azure AI Search |
| Generate images from text | DALL-E model via Azure OpenAI |
| Evaluate prompt quality | Azure AI Foundry evaluation tools |
| Deploy models with safety guardrails | Azure AI Foundry + Content Safety |
| Build autonomous agent with tools | Azure AI Agent Service / Foundry agents |
| Control response length and cost | Max tokens parameter |
| Reduce harmful output | Content Safety filters / system prompts |
FAQ
What service extracts data from invoices?
Azure AI Document Intelligence. It has prebuilt models for invoices, receipts, tax forms, and IDs. Azure AI Vision does OCR but not structured field extraction.
What service detects harmful content in user-generated text and images?
Azure AI Content Safety. It handles both text and image moderation across multiple categories including hate, self-harm, sexual, and violent content.
What service provides semantic search over documents?
Azure AI Search with semantic ranking configured. It uses language understanding to rank results by meaning, not just keyword overlap.
When should I use Azure AI Search instead of a database?
Use Azure AI Search for semantic search, vector similarity, hybrid retrieval, and document enrichment pipelines. Use a database for exact match queries and aggregations.
What is the difference between RAG and fine-tuning?
RAG retrieves relevant context at query time and includes it in the model prompt. Fine-tuning modifies the model weights using training data. RAG is more flexible and cost-effective for enterprise knowledge; fine-tuning is better for behavioral changes.
What service translates documents while preserving formatting?
Azure AI Translator (document translation API). It handles multiple document formats and preserves layout and structure.
Master these service selection rules and you will pass the AI-102. Practice with 35 free questions at cert-pass.com/exams/azure-ai-102-azure-ai-engineer-associate/take. Full prep with 1000+ questions and detailed explanations starts at EUR 29.
