AI Workload Types

The five AI workload categories

The very first thing the AI-900 exam asks you to do is name what kind of AI problem you are looking at. Microsoft's Introduction to AI concepts^[1] module groups Azure AI capabilities into a small set of recurring workload categories, and identifying them is largely a classification drill: read a business scenario, decide which category it belongs to. There are five categories worth knowing cold, but be precise about scope: the objective "Identify features of common AI workloads" names exactly four (computer vision, NLP, document processing, generative AI); the fifth, machine learning, is the foundation underneath them all and is assessed in its own AI-900 domain.

The fastest way to tell the categories apart is one lens: ask what goes in and what comes out. Each category below is one input→output signature shown in the figure. Read every description as an instance of that lens.

1. Machine learning

Machine learning (ML) is the practice of training a model on historical data so it can make a prediction about new, unseen data. The output is typically a number or a class label, a score, not new content. Predicting a house price from its size and location, or predicting whether a transaction is fraudulent, are textbook ML scenarios. ML is also the foundation under the other four categories: vision, language, and generative models are themselves ML models (usually deep neural networks) trained on huge datasets.

2. Computer vision

Computer vision interprets the visual world, still images and video. Given an image, it returns labels ("this is a cat"), detects and locates objects with bounding boxes, recognizes faces, or extracts printed and handwritten text via optical character recognition (OCR). "Count the cars in a parking-lot photo," "detect defective parts on a conveyor belt," and "read the license plate" are all vision tasks.

3. Natural language processing (NLP)

NLP interprets written and spoken human language. It takes unstructured text or audio and returns structured insight: the sentiment of a review, the key phrases in a document, the named entities (people, places, dates) in a contract, the detected language, a translation, or a transcription of speech to text. "Flag negative customer reviews" and "transcribe a support call" are NLP scenarios.

4. Document intelligence and knowledge mining

This category turns piles of semi-structured documents (invoices, receipts, forms, contracts) into structured, queryable data. Document intelligence extracts specific fields, key-value pairs, and tables from forms. Knowledge mining ingests large volumes of mixed content and makes it searchable and explorable. It overlaps with vision (both can do OCR) but its goal is different: producing structured, searchable information at scale, e.g. "extract every line item from ten thousand scanned invoices."

5. Generative AI

Generative AI takes a natural-language prompt and produces new, original content (text, code, summaries, or images) rather than a fixed label or score. "Draft a marketing email," "summarize this report," and "generate an image from a description" are generative scenarios. The defining contrast: the other four categories analyze or predict on existing input; generative AI creates something that did not exist before.

These categories are not mutually exclusive (a real support chatbot can combine NLP (understanding the user) with generative AI (composing the reply)) but the exam still wants you to name each constituent workload. When in doubt, fall back on the input→output lens and the figure above: match the scenario's input and output to one row.

Pick a workload by matching a scenario's input and output: prompt to new content is generative AI; image, text/audio, forms, or historical data route to the other four.

How each workload maps to an Azure AI service

AI-900 rarely stops at "which workload?". It usually pairs the workload with which Azure service you would use. Each category has a primary service family, and knowing the one-to-one mapping answers most service-selection questions.

Workload category	Primary Azure service	What it does
Machine learning (custom)	Azure Machine Learning^[2]	Train, deploy, and manage your own predictive models (regression, classification, etc.) and run MLOps.
Computer vision	Azure AI Vision^[3]	Analyze images for objects, captions, and OCR text extraction.
Natural language processing	Azure AI Language^[4]	Sentiment analysis, key phrase extraction, named entity recognition, language detection, summarization.
Document intelligence	Azure AI Document Intelligence^[5]	Extract text, key-value pairs, and tables from invoices, receipts, and forms.
Knowledge mining	Azure AI Search^[6]	Index large content stores and apply AI enrichment so the content becomes searchable.
Generative AI	Azure OpenAI / Foundry models^[7]	Generate text, code, and images from prompts using GPT and other large models.

Custom model vs. prebuilt service

The most important distinction the exam tests here is build vs. consume, and the figure groups the same services by that split. Azure Machine Learning^[2] is the platform you use when you must train your own model on your own data. It is the answer when the scenario is a bespoke prediction problem ("predict equipment failure from sensor telemetry," "forecast demand per store"). The vision, language, and document services are prebuilt: they package already-trained models behind a simple API, so you consume AI without training anything. When a stem says "with the least development effort" or "without data science expertise," it is steering you toward a prebuilt Azure AI service, not Azure Machine Learning.

Speech sits inside NLP

Speech-to-text, text-to-speech, and speech translation are delivered by the Azure AI Speech service, but for workload classification purposes the exam treats speech as part of the natural language processing family: language is language whether typed or spoken. If a scenario says "transcribe a phone call" or "read this text aloud," classify it as NLP/language.

Why the families consolidated

Many of these capabilities are now grouped under the umbrella of Azure AI services (formerly Cognitive Services) and surfaced in Microsoft Foundry, but the AI-900 mapping of one workload → one primary service still holds and is what the exam rewards: vision→Vision, language→Language, documents→Document Intelligence, search/enrichment→AI Search, generation→Azure OpenAI, custom training→Azure Machine Learning.

The same services split build vs. consume: train your own with Azure Machine Learning, or consume prebuilt Azure AI services (Vision, Language, Document Intelligence, AI Search, Azure OpenAI).

Exam-pattern recognition: classifying the scenario

AI-900 workload questions follow a predictable shape. The stem describes a goal in plain business language and asks "Which type of AI workload is this?" or "Which Azure service should you use?" The trick is that the four wrong answers are always other valid workloads. The question is testing whether you can tell them apart, not whether you know what AI is.

Decode the stem by input and output

Apply the same input→output signature from The five AI workload categories above, but now keyed off the verbs and nouns in the scenario so you can spot it fast under exam pressure:

"predict / forecast / estimate a value" + tabular data → machine learning. (Example: predict next month's sales from past sales.)
"identify / detect / classify objects in a photo," "read text from an image" → computer vision. See the computer vision^[8] overview.
"analyze sentiment," "extract key phrases," "detect the language," "translate," "transcribe speech" → natural language processing.
"extract fields from an invoice / receipt / form," "make a document archive searchable" → document intelligence / knowledge mining. See information extraction^[9].
"generate / draft / create / summarize new content from a prompt" → generative AI. See generative AI^[10] and the fundamentals of generative AI^[11] module.

Common distractors and why they're wrong

Computer vision vs. document intelligence. Both can read text via OCR, so they're often paired as distractors. The discriminator is the goal: if you just need raw text out of an image (a photo, a sign), that's computer vision; if you need structured fields (the total, the invoice number, the line items) from a form, that's document intelligence.
NLP vs. generative AI. Both work on text. NLP analyzes existing text and returns structure (sentiment, entities). Generative AI produces new text from a prompt. "Determine if a review is positive" = NLP; "write a reply to the review" = generative AI.
Machine learning vs. a prebuilt service. When the scenario is a standard task (sentiment, OCR, object detection), the better answer is the prebuilt service, not "build a model in Azure Machine Learning." Choosing custom ML for an off-the-shelf task is the over-engineering trap. Azure Machine Learning is correct only when the prediction is bespoke and no prebuilt service covers it.
"Knowledge mining" wording. If the scenario emphasizes searching/exploring a large body of mixed documents, the answer is Azure AI Search (knowledge mining), even though Document Intelligence might also touch the files. Search and discoverability is the AI Search signal.

What is out of scope here

This subtopic is about recognizing the category, not configuring it. Questions about pricing tiers, throughput, the six responsible-AI principles, or the internal algorithms (regression vs. clustering, the Transformer architecture) belong to other AI-900 objectives. On a pure workload-identification question, resist answers that drift into those areas. Pick the category whose input and output match the scenario, and choose the prebuilt service unless the task is genuinely custom.

Workload category	Typical input	Typical output	Example scenario
Machine learning	Structured/tabular data	Predicted value or class (score)	Predict house price from size and location
Computer vision	Images or video	Labels, bounding boxes, extracted text	Detect and count vehicles in a parking-lot photo
Natural language processing	Text or speech (audio)	Entities, sentiment, key phrases, translation	Detect negative sentiment in product reviews
Document intelligence & knowledge mining	Forms, invoices, documents	Structured fields plus searchable index	Extract line items from scanned invoices
Generative AI	Natural-language prompt	New original content (text, code, image)	Draft an email or summarize a report

Decision tree

Cheat sheet

Sharp facts the exam loves — give these one last read before exam day.

Cheat sheet

Sharp facts the exam loves — scan these before test day.

Identify a workload by classifying its scenario into an AI category

AI-900 leans on recognizing a handful of recurring Azure AI workload categories from a business scenario: computer vision, natural language processing (NLP), document intelligence / knowledge mining, generative AI, and machine learning underneath them all. The official "Identify features of common AI workloads" objective, however, names exactly four: computer vision, natural language processing, document processing, and generative AI, and is essentially a classification drill: read the scenario, name the category. Machine learning is NOT under that objective; it sits in its own skill area. Categories overlap in practice (a chatbot blends NLP and generative AI), but each is tested individually.

Trap Listing knowledge mining or machine learning among the workloads under the single "common AI workloads" objective. The official outline names only four, and says "document processing," not "knowledge mining."

Classify a workload by what goes in and what comes out

The fastest way to name a workload is the input→output test: tabular data in, a predicted score or class out = machine learning; image in, labels/boxes/extracted text out = computer vision; text or audio in, entities/sentiment/translation out = NLP; a form in, structured fields + a search index out = document intelligence; a prompt in, brand-new content out = generative AI. Most AI-900 scenario items are answerable on this signature alone.

Trap Calling a scenario that outputs structured fields from a form "computer vision" because an image went in. Image-in alone is vision, but field/key-value output is the document intelligence signature.

11 questions test this

Machine learning predicts a value or class. It doesn't create content

A machine learning workload trains a model on historical data to predict a numeric value (regression) or a category (classification) for new, unseen data; the output is a score or label, never new content. Predicting a house price from size and location is regression; flagging a transaction as fraudulent is classification. Numeric-looking labels (tier 1/2/3) are still classification, because the values are discrete classes, not a quantity on a continuous scale.

Trap Calling a tier-1/2/3 or category prediction "regression" because the labels look numeric. Discrete classes are classification; regression is only for a continuous value.

1 question tests this

You need to choose an AI workload type for an application that predicts next month's sales revenue from historical sales figures. Which…

Machine learning is the foundation under the other workloads

Machine learning underpins the other categories rather than competing with them: computer vision, NLP, and generative AI models are themselves ML models, typically deep neural networks trained on large datasets. That is why a bespoke prediction task with no off-the-shelf option points to Azure Machine Learning, while the prebuilt vision, language, and document services simply package already-trained ML models behind an API.

Trap Selecting Azure Machine Learning whenever "machine learning" is mentioned. A prebuilt vision or language service is still ML under the hood, so the prebuilt service is the answer unless the scenario demands training a custom model.

Computer vision interprets images and video

A computer vision workload extracts meaning from visual input: still images, video, or camera streams. Its core solution types are image classification (one label for the whole image), object detection (per-object labels with bounding-box coordinates, so it can locate and count), semantic segmentation (per-pixel masks for more precise location), facial detection and analysis, and OCR to read text. "Count the cars in a photo" and "read a license plate" are vision tasks.

Trap Choosing image classification when the scenario needs each item located or counted. A single whole-image label has no coordinates; locating multiple objects is object detection.

13 questions test this

NLP interprets written and spoken language

A natural language processing workload turns unstructured text or audio into structured insight: sentiment analysis, key phrase extraction, named entity recognition, language detection, translation, and speech-to-text transcription. "Flag negative reviews" and "transcribe a support call" are NLP scenarios. The output is structure about existing language (a sentiment, an entity, a transcript), not newly authored content.

Trap Tagging "write a reply to the customer" as NLP because it involves language. NLP returns structure about existing text; authoring new text from a prompt is generative AI.

5 questions test this

Speech and translation are classified under NLP

For AI-900 workload classification, speech (speech-to-text, text-to-speech, speech translation) and translation belong to the NLP exam domain. The official outline lists "speech recognition and synthesis" and "translation" under NLP and names both Azure AI Language and Azure AI Speech as its services. So "transcribe a phone call" or "read this text aloud" is correctly an NLP/language workload.

Trap Classifying a speech-to-text or translation scenario as its own separate workload category. On AI-900 these roll up under the NLP / language domain rather than standing apart from it.

2 questions test this

Knowledge mining makes a large content store searchable

Knowledge mining ingests large volumes of mixed content and applies AI enrichment so it becomes searchable and explorable at scale, built on Azure AI Search. When a scenario emphasizes searching or exploring a whole body of documents, rather than pulling specific named fields from a form, the signal is knowledge mining, not document intelligence.

Trap Reaching for document intelligence when the goal is full-text search across a large corpus. Extracting named fields from one form is IDP; making the whole corpus searchable is knowledge mining.

1 question tests this

A company wants to extract insights from thousands of PDF documents, images, and scanned forms stored in Azure Blob Storage. The extracted…

Generative AI creates new content from a prompt

A generative AI workload takes a natural-language prompt and produces new, original content (text, code, summaries, or images) instead of a fixed label or score. "Draft a marketing email," "summarize this report," and "generate an image from a description" are generative scenarios. The defining contrast: the other categories analyze or predict on existing input, while generative AI synthesizes something new.

Trap Labeling "summarize this report" as NLP because it processes text. Condensing existing text into new prose is a generative AI task; NLP returns structure like entities or sentiment, not authored output.

7 questions test this

Each workload maps to one primary Azure service

AI-900 pairs each workload with its primary Azure service: custom ML → Azure Machine Learning; computer vision → Azure AI Vision; NLP → Azure AI Language (plus Azure AI Speech for audio); document intelligence → Azure AI Document Intelligence; knowledge mining → Azure AI Search; generative AI → Azure OpenAI. This one-workload-to-one-service mapping answers most service-selection items. (Docs may now title these "in Foundry Tools," but the established names remain exam-current.)

Trap Picking Azure AI Vision for a read-the-fields-from-a-form scenario because both involve images. Text-only OCR is Vision, but extracting structured fields maps to Azure AI Document Intelligence.

2 questions test this

Build custom only when no prebuilt service fits

AI-900 tests build-vs-consume: Azure Machine Learning is the answer only when you must train your own model for a bespoke prediction with no off-the-shelf option. The vision, language, and document services are prebuilt: already-trained models behind a simple API, no data-science skill required. Stems saying "least development effort" or "without data science expertise" steer you to a prebuilt Azure AI service, not Azure Machine Learning.

Trap Picking Azure Machine Learning for a "least effort / no data scientists" scenario. That phrasing points to a prebuilt Azure AI service; AML means you train the model yourself.

Raw text out is vision OCR; named fields out is document intelligence

Computer vision and document intelligence both read text via OCR and are paired as distractors. The discriminator is the goal: if you just need the raw text out of an image (a photo, a sign, a poster) that is computer vision (Read OCR); if you need specific structured fields such as the total, invoice number, or line items from a form, that is document intelligence, which uses OCR as its foundation but adds field/table/key-value extraction.

Trap Choosing document intelligence just because the input is text in an image. Pulling raw text from a sign or poster is Vision Read OCR; document intelligence is only when you need named fields from a form.

13 questions test this

Analyzing text is NLP; writing new text is generative AI

NLP and generative AI both operate on text, making them common distractors. NLP analyzes existing text and returns structure: sentiment, entities, key phrases; generative AI produces new text from a prompt. "Determine if a review is positive" is NLP (sentiment analysis); "write a reply to the review" is generative AI.

Trap Calling "write a reply to the review" NLP because a review went in. Analyzing the review's sentiment is NLP, but composing a new reply is generative AI.

1 question tests this

Which scenario is a common use case for a generative AI workload rather than a traditional analytical AI workload?

A channel connects one bot to many communication apps

A channel is the connection between Azure AI Bot Service and a communication app such as Microsoft Teams, web chat, Slack, Facebook, Telegram, or SMS (via Twilio). Because the Bot Framework is channel-agnostic, you build one bot and deploy it to many channels without rewriting code per channel. Bot Service is the build/deploy framework: it delegates language understanding to Azure AI Language, it is not itself an NLP service.

Trap Treating Azure AI Bot Service as the service that does the NLP. It hosts and connects the bot but delegates intent/entity understanding and FAQ answers to Azure AI Language.

7 questions test this

The Bot Connector Service relays messages between a bot and its channels

The Bot Connector Service is the Azure Bot Service component that relays messages and events between a bot and its channels, translating the Bot Framework activity schema to and from each channel's native format. This normalization is what lets the bot stay channel-agnostic: the bot speaks one schema while the connector adapts it per channel.

Trap Assuming the Bot Connector Service interprets the user's intent. It only relays and translates messages between the bot and its channels; understanding the language is delegated to Azure AI Language.

4 questions test this

Indexer crawls and cracks, skillset enriches, knowledge store persists output

In Azure AI Search, an indexer is the crawler that connects to a data source, performs document cracking to extract text and images, and populates the index. A skillset is the collection of AI-enrichment skills (OCR, entity recognition, key-phrase extraction, translation) the indexer runs over that content. A knowledge store optionally saves the enriched output as tables or objects in Azure Storage for use outside search, for example analysis in Power BI.

Trap Confusing the knowledge store with the search index. The index is what search queries hit, while the knowledge store persists enriched output to Azure Storage for use outside search.

9 questions test this

Knowledge mining runs in three phases: ingest, enrich, explore

A knowledge-mining solution on Azure AI Search runs in three phases: ingest content from data sources via the indexer (document cracking), enrich it with AI skills to extract information and patterns, and explore the indexed result through search queries, apps, or visualizations. The enriched output always lands in a search index, and optionally in a knowledge store for downstream use.

2 questions test this

In CLU, the intent is the action; entities are the details that fulfill it

In Conversational Language Understanding, an intent represents the task or action a user wants to perform (e.g., BookFlight), while entities are the words in the utterance that supply the details needed to fulfill that intent (the destination, date, or product). The model predicts one overall intent per utterance and extracts its entities. An utterance is the user's short natural-language input that the model is trained on.

Trap Mixing up intent and entity. The intent is the single overall action the user wants, while entities are the detail words within the utterance that fill in that action.

8 questions test this

Match the NLP task to its Azure AI Language feature

Map the task to the feature: named entity recognition identifies entries in text and categorizes them into predefined types (people, organizations, locations); key phrase extraction returns the main concepts/topics as a list; sentiment analysis returns positive/negative/neutral with confidence; and PII detection identifies (and can redact) sensitive data such as SSNs and phone numbers. These are all Azure AI Language features, not the legacy brand names that appear in distractors.

Trap Selecting a retired brand like LUIS or QnA Maker for these tasks. On exam-current AI-900, the single answer for NLP text features is Azure AI Language.

9 questions test this

Document Intelligence offers prebuilt models and custom models

Azure AI Document Intelligence (formerly Form Recognizer) provides prebuilt models for common documents: invoice, receipt, ID, and layout, where layout extracts text, tables, selection marks, and structure, plus custom models for unique business forms, all returned as structured JSON. Choose a custom template model when your forms share a consistent, static visual layout, and a custom neural model for mixed-type or varied documents.

Trap Building a custom model for a standard invoice or receipt. Document Intelligence already ships prebuilt models for those; custom models are for unique business forms with no prebuilt match.

8 questions test this

Tagging labels content without location; object detection adds bounding boxes

In Azure AI Vision Image Analysis, tagging returns content tags (objects, living beings, scenery, actions, even indoor/outdoor) each with a confidence score but no location. Object detection goes further, adding bounding-box coordinates for each detected object. A caption returns a single natural-language sentence describing the whole image. So "what is in this image" can be tagging, but "where is each object" requires object detection.

Trap Expecting image tagging to tell you where each object is. Tags carry confidence scores but no coordinates; bounding boxes come from object detection.

12 questions test this

The Read OCR engine extracts both printed and handwritten text

Azure AI Vision's Read OCR engine uses optical character recognition to extract both printed and handwritten text from images and documents, including mixed-language and mixed-mode (print + handwritten) pages. It returns pages, text lines, and words with their location and a confidence score. "Read the text," "handwritten," or "OCR" fingerprints Azure AI Vision rather than Document Intelligence.

Trap Assuming Read OCR only handles printed text and reaching for another service for handwriting. The Read engine extracts both printed and handwritten text, including mixed print-plus-handwritten pages.

9 questions test this

Face detection locates a face; recognition identifies whose it is

Azure AI Face detection (the Detect API) locates each face and returns its rectangle coordinates without identifying anyone. It is required as the first step in every other scenario. Recognition goes further: verification is one-to-one ("do these two faces belong to the same person?", e.g. selfie vs. ID photo, with a confidence score), identification is one-to-many against a person group, and the Group operation clusters unknown faces by similarity. Face recognition is Limited Access, gated by Responsible-AI eligibility.

Trap Assuming face detection tells you who a person is. Detection only returns the face's location; matching it to an identity is verification or identification.

3 questions test this

The five AI workload categories

1. Machine learning

2. Computer vision

3. Natural language processing (NLP)

4. Document intelligence and knowledge mining

5. Generative AI

How each workload maps to an Azure AI service

Custom model vs. prebuilt service

Speech sits inside NLP

Why the families consolidated

Exam-pattern recognition: classifying the scenario

Decode the stem by input and output

Common distractors and why they're wrong

What is out of scope here

Decision tree

Cheat sheet

Also tested in

References