AI-901 AI workloads practice questions

12 sample questions with full explanations. Try each one before you open the answer. New to this area? Read the AI workloads study guide first.

Question 1

A travel company wants an assistant that reads a customer's request, checks seat availability in the company's booking system, and then books the flight. Which kind of AI solution fits this best?
- A A chatbot that answers frequently asked questions from a fixed script
- B An AI agent with tools
- C A text analysis service that extracts key phrases from the request
- D A generative AI model that writes a reply describing the available flights
Show answer and explanation
Answer: B

Agents are built on generative AI but can also use tools to take actions, such as querying and updating a booking system. Checking availability and then completing a booking needs those actions, not just a text reply.
- A: A scripted FAQ chatbot handles predictable questions. It can't check live availability or make a booking.
- C: Key phrase extraction would pull out words like the destination, but it can't check the system or book anything.
- D: A model on its own can describe flights, but without tools it can't look up real availability or complete the booking.
Source: Generative AI and agents (Introduction to AI concepts module) (Microsoft Learn)
Question 2

An accounting firm is listing AI projects for next year. Which two are information extraction workloads? Choose two.

Choose two.
- A Pull the invoice number, date and total from scanned supplier invoices
- B Decide whether each client survey comment is positive or negative
- C Turn thousands of archived paper tax forms into searchable database records
- D Generate an illustration for the firm's annual report
- E Read a monthly summary aloud to clients who call the help line
Show answer and explanation
Answer: A and C

Information extraction pulls structured data out of unstructured sources such as scanned documents and forms. Reading fields from invoices and digitizing paper forms into records are both extraction.
- B: Labeling comments as positive or negative is sentiment analysis, a text analysis workload.
- D: Creating an illustration is image generation.
- E: Reading text aloud is speech synthesis.
Source: Information extraction (Introduction to AI concepts module) (Microsoft Learn)

Question 3

A bank designs a loan-processing solution in which one agent gathers documents, a second agent checks them against policy, and a third agent sends the decision. For each statement, choose Yes if it is true or No if it is false.

Each agent in this solution can have its own specialty.
In a multi-agent solution, one agent must do all of the work while the other agents only review it.
A multi-agent solution requires every agent to use the same set of tools.

Show answer and explanation

Each agent in this solution can have its own specialty.	Yes
In a multi-agent solution, one agent must do all of the work while the other agents only review it.	No
A multi-agent solution requires every agent to use the same set of tools.	No

In a multi-agent solution, several agents collaborate, each with its own specialty and its own part of the work, rather than one agent doing everything. Because each agent has its own job, each can have different tools.

Source: AI agents (Introduction to generative AI and agents module) (Microsoft Learn)

Question 4

A hospital wants to share thousands of patient feedback comments with a research partner. Before sharing, it must remove patients' names, phone numbers and home addresses from the text. Which text analysis technique should it use?
- A Key-term extraction
- B Personally identifiable information (PII) detection and redaction
- C Sentiment analysis
- D Language detection
Show answer and explanation
Answer: B

PII detection is a specialized form of entity detection that finds private details such as names, addresses and phone numbers so they can be redacted before data is shared or analyzed.
- A: Key-term extraction finds the important words and phrases in a comment. It doesn't look specifically for private details or remove them.
- C: Sentiment analysis rates comments as positive, negative or neutral. It leaves personal details in place.
- D: Language detection only identifies the language of each comment.
Source: Text and natural language (Introduction to AI concepts module) (Microsoft Learn)

Question 5

A retailer is choosing between specialist text analysis tools and a generative AI model for analyzing product reviews. For each statement, choose Yes if it is true or No if it is false.

Specialist text analysis tools can be used where the retailer needs predictable results or custom rules.
Sentiment analysis is a type of text classification.
Language detection is usually the final step in a multi-stage text processing workflow.

Show answer and explanation

Specialist text analysis tools can be used where the retailer needs predictable results or custom rules.	Yes
Sentiment analysis is a type of text classification.	Yes
Language detection is usually the final step in a multi-stage text processing workflow.	No

Many language tasks can be handled by generative AI, but specialist text analysis tools are used where you need predictable results or custom rules. Sentiment analysis assigns text to a category (positive, negative or neutral), so it's a form of text classification. Language detection is usually the first step, not the last.

Source: Text and natural language (Introduction to AI concepts module) (Microsoft Learn)

Question 6

All of a company's meetings are held in English. Its meeting app must produce a written English transcript of each meeting, and then read a short summary aloud to people who join late. Which two speech capabilities does it need? Choose two.

Choose two.
- A Speech recognition (speech to text)
- B Speech translation
- C Speech synthesis (text to speech)
- D Language detection
Show answer and explanation
Answer: A and C

The transcript needs speech recognition, which turns spoken audio into text. Reading the summary aloud needs speech synthesis, which turns text into spoken audio.
- B: Speech translation converts speech into another language. Everything here stays in English.
- D: Language detection identifies which language is used. The meetings are always in English, so it isn't needed.
Source: Speech (Introduction to AI concepts module) (Microsoft Learn)
Question 7

A grocery chain's shelf cameras must find where each product is in an image and mark each one with a box, so staff can spot empty spaces. Which type of computer vision model fits best?
- A Image classification
- B Object detection
- C Optical character recognition (OCR)
- D Image generation
Show answer and explanation
Answer: B

Object detection identifies the location of specific objects in an image, typically with a bounding box around each one. That's what's needed to find each product on the shelf.
- A: Image classification gives one label for the whole image, such as "cereal shelf". It doesn't say where each product is.
- C: OCR reads text in an image, such as price labels. It doesn't locate the products themselves.
- D: Image generation creates new images. It doesn't analyze camera images.
Source: Computer vision (Introduction to AI concepts module) (Microsoft Learn)
Question 8

A hands-free kitchen assistant lets a cook ask a question out loud and hear a spoken answer that an AI agent decides on. In which order do the stages of this speech-to-speech pipeline run?
- A Speech to text, then speech translation, then text to speech
- B Reasoning, then speech to text, then text to speech
- C Speech to text, then reasoning, then text to speech
- D Speech to text, then text to speech, then reasoning
Show answer and explanation
Answer: C

The cook's spoken question must first become text (speech to text). The agent then reasons over that text and decides what to say. Finally, text to speech turns the reply into spoken audio the cook can hear.
- A: Speech translation only changes the language of what was said. Nothing in this pipeline decides how to answer the cook, so a reasoning stage is still needed.
- B: The agent can't reason about the question until the spoken audio has been turned into text.
- D: This speaks a reply before the agent has decided what to say. Reasoning must come before text to speech.
Source: Creating a speech-capable agent (Get started with speech in Azure) (Microsoft Learn)
Question 9

A government agency has thousands of recorded support calls. It wants to find the most common problems callers report and how each was resolved, and store the results as structured data. Which type of AI workload is this?
- A Speech recognition only, producing raw transcripts
- B Information extraction from audio
- C Speech translation into other languages
- D Sentiment analysis of each call
Show answer and explanation
Answer: B

The agency wants to pull specific facts (problems and resolutions) out of unstructured recordings and turn them into structured data it can analyze. That is information extraction, applied to audio. Microsoft Learn uses this exact support-call example as an information extraction scenario.
- A: Transcripts are a useful first step, but raw text isn't structured data. The agency still needs the problems and resolutions pulled out of it.
- C: Speech translation converts speech into another language. Nothing here says the calls need a different language.
- D: Sentiment analysis says whether callers sounded positive or negative. It doesn't identify the problems or how they were resolved.
Source: Introduction (Get started with AI-powered information extraction in Azure) (Microsoft Learn)
Question 10

An online store receives thousands of product reviews a day. The team wants to automatically label each review as positive, negative, or neutral. Which text analysis technique should they use?
- A Entity detection
- B Keyword extraction
- C Sentiment analysis
- D Summarization
Show answer and explanation
Answer: C

Sentiment analysis evaluates text to decide whether the opinion it expresses is positive, negative, or neutral, which is exactly the label the team wants for each review.
- A: Entity detection finds named things in text, such as people, places, and products, but not the opinion about them.
- B: Keyword extraction returns the main topics of a review, not whether the review is positive or negative.
- D: Summarization shortens text into its key points; it doesn't classify the opinion.
Source: What is Azure Language in Foundry Tools? (Microsoft Learn)
Question 11

A pharmacy wants its phone system to read each customer's order status aloud in a natural, human-sounding voice. Which speech capability does this require?
- A Speech recognition (speech to text)
- B Speech synthesis (text to speech)
- C Speaker recognition
- D Keyword spotting
Show answer and explanation
Answer: B

Reading text aloud means turning written text into spoken audio. That's speech synthesis, also called text to speech, which produces humanlike synthesized speech from input text.
- A: Speech recognition works the other way, turning spoken audio into text.
- C: Speaker recognition identifies or verifies who is speaking; it doesn't produce speech.
- D: Keyword spotting listens for a specific word or phrase in audio; it doesn't speak.
Source: What is Azure Speech? (Microsoft Learn)
Question 12

A logistics company is planning several AI features for its operations app. Which two features are computer vision workloads? Choose two.

Choose two.
- A Detecting forklifts and pallets in warehouse camera images
- B Translating customer support emails into Spanish
- C Generating a short text description of what a photo shows
- D Turning recorded support calls into written transcripts
Show answer and explanation
Answer: A and C

Computer vision works on images and video. Finding objects in camera images and describing what a photo shows (image captioning) both analyze visual content.
- B: Translating emails works on text, so it's a language workload.
- D: Transcribing calls works on audio, so it's a speech workload.
Source: What is Azure Vision in Foundry Tools? (Microsoft Learn)

Want more practice on AI workloads?

Practice the full question bank for this area, in random order, with your progress saved in your browser.

Practice AI workloads