7.1 Service Comparison and Selection Guide
Key Takeaways
- The AI-102 exam frequently tests your ability to select the correct Azure AI service for a given scenario — memorize the service boundaries.
- Azure AI Language handles pre-built NLP tasks (NER, sentiment, key phrases); Azure OpenAI handles generative text tasks (summarization, Q&A, content creation).
- Azure AI Vision Read API handles general text extraction from images; Azure AI Document Intelligence handles structured field extraction from forms and documents.
- Azure AI Content Safety is for content moderation; Azure OpenAI content filters are built into the OpenAI deployment for generative AI output filtering.
- CLU is for structured intent/entity extraction; Custom Question Answering is for FAQ-style Q&A; Azure OpenAI is for open-ended generative responses.
Last updated: March 2026
Service Comparison and Selection Guide
Quick Answer: The AI-102 heavily tests service selection. Key distinctions: Language (pre-built NLP) vs. OpenAI (generative AI), Vision Read (general OCR) vs. Document Intelligence (structured extraction), CLU (intents/entities) vs. Question Answering (FAQ) vs. OpenAI (open-ended).
Text Processing Decision Matrix
| Scenario | Correct Service | Why NOT the Alternative |
|---|---|---|
| Extract sentiment from customer reviews | Azure AI Language (sentiment analysis) | OpenAI could do it but Language is purpose-built, cheaper, and faster |
| Summarize a long legal document | Azure OpenAI (chat completions) | Language has summarization but OpenAI handles complex summarization better |
| Detect entities (people, places, orgs) | Azure AI Language (NER) | OpenAI could extract entities but Language NER is structured and deterministic |
| Extract custom entities (policy numbers, case IDs) | Azure AI Language (custom NER) | OpenAI could do it via prompting but custom NER is trained and consistent |
| Classify support tickets into categories | Azure AI Language (custom text classification) | OpenAI could classify but custom classification is trained on your data |
| Generate marketing copy from a brief | Azure OpenAI (chat completions) | Language cannot generate creative content |
| Translate text between languages | Azure AI Translator | Language has no translation; OpenAI translation is less reliable |
| Detect PII and redact it | Azure AI Language (PII detection) | OpenAI might miss PII; Language PII is comprehensive and deterministic |
| Build a chatbot that understands user intents | CLU (Conversational Language Understanding) | OpenAI is generative, not structured intent recognition |
| Answer FAQ-type questions from a knowledge base | Custom Question Answering | CLU doesn't have Q&A pairs; OpenAI needs RAG setup |
Image Processing Decision Matrix
| Scenario | Correct Service | Why NOT the Alternative |
|---|---|---|
| Extract text from a photograph of a sign | Azure AI Vision (Read API / OCR) | Document Intelligence is for structured documents, not general images |
| Extract fields from an invoice (vendor, amount) | Azure AI Document Intelligence (prebuilt-invoice) | Vision OCR extracts raw text without understanding invoice structure |
| Classify product images into categories | Azure AI Custom Vision | Vision Image Analysis provides general tags, not custom categories |
| Detect faces and verify identity | Azure AI Face | Vision detects people (bounding boxes) but does not verify identity |
| Count people in a retail store via camera | Azure AI Vision (Spatial Analysis) | Face API does recognition, not counting/tracking |
| Analyze video for transcripts and topics | Azure AI Video Indexer | Vision API processes images, not video |
| Generate an image from a text description | Azure OpenAI (DALL-E 3) | No other Azure AI service generates images |
Search and Knowledge Decision Matrix
| Scenario | Correct Service | Why NOT the Alternative |
|---|---|---|
| Full-text search over enterprise documents | Azure AI Search | Other services don't provide search |
| AI-enriched search (OCR + NER + key phrases during indexing) | Azure AI Search (with skillsets) | Individual services don't provide search indexing |
| Build a Q&A system over company FAQs | Custom Question Answering | Search provides documents, not direct answers |
| Build a conversational AI grounded in enterprise data | Azure OpenAI + Azure AI Search (RAG) | Search alone doesn't generate answers |
| Process invoices, receipts, and forms at scale | Azure AI Document Intelligence | Search extracts existing content; Doc Intel understands form structure |
Deprecated Services and Their Replacements
| Deprecated Service | Replacement | Migration Path |
|---|---|---|
| LUIS | CLU (Conversational Language Understanding) | Export LUIS app, import into CLU |
| QnA Maker | Custom Question Answering | Export KB, create new Q&A project |
| Form Recognizer | Azure AI Document Intelligence | Name change only — same SDK |
| Cognitive Search | Azure AI Search | Name change only — same API |
| Cognitive Services | Azure AI Services | Name change — multi-service resource |
| Emotion API (Face) | Retired — no replacement | Feature removed for Responsible AI |
| LUIS Authoring/Prediction | Azure AI Language REST API | New API endpoints |
On the Exam: Questions about deprecated services test whether you know the current service name and migration path. If you see LUIS, QnA Maker, or Form Recognizer in answer options, they are likely wrong (unless the question specifically asks about migration).
Test Your Knowledge
A company needs to extract vendor names, invoice numbers, and total amounts from thousands of scanned invoices. Which service combination should they use?
A
B
C
D
Test Your Knowledge
A travel company wants to build a chatbot that understands "Book a flight to Paris" and extracts the destination. Which service should they use?
A
B
C
D
Test Your Knowledge
Which deprecated service should be migrated to Conversational Language Understanding (CLU)?
A
B
C
D