Practice Quiz Bank
40 scenario-based questions across 8 topics — modelled on the AB-731 exam style. Each question is new, not from the study sections.
Study Sections
Azure AI Foundry
What Foundry is, RBAC vs anonymous access, Marketplace billing, classifying feedback.
Responsible AI — FAITHP
Microsoft's 6 principles. Covers Q17–Q25 — 4–5 exam questions in one framework.
Pretrained vs Fine-Tuned vs RAG
Building analogy, pharmaceutical scenario, Copilot Agent vs Foundry.
AI Governance
AI Council vs AI Champions — differences, Nigerian examples, deployment design.
Azure AI Ecosystem
Three-layer model: Azure AI → Foundry → Copilot. Customer data rules.
Purview & Entra Security
Purview DLP, sensitivity labelling, Entra identity, Conditional Access.
Copilot Licensing
Per-user licences, pay-as-you-go vs prepaid capacity, what is NOT in M365 plans.
AI Challenges
Fabrication vs Hallucination vs Reliability vs Bias — exam discrimination rules.
40 Practice Questions
All questions are new and scenario-based — different from the study section examples. Each quiz has 5 questions testing a specific topic. Aim for 4/5 before sitting the exam.
Section 1 · SS1, SS2, SS3, SS4, SS15
Microsoft Azure AI Foundry
You struggled with what Foundry is, why it exists, and how its parts connect. Each original Microsoft Learn question is shown first, then your comment, then the full explanation.
SS1 · Q13
What is Microsoft Azure AI Foundry?
📋 Original Microsoft Learn Question
Question 13 of 50
You are evaluating Microsoft Foundry for enterprise generative AI initiatives. Which two benefits of Foundry should you highlight to stakeholders? Each correct answer presents part of the solution.
Select TWO correct answers:
Your comment (SS1): Explain to me what this Microsoft Foundry mean, why did Microsoft create it, what problem is it solving? Use layman language with analogy.
The Shopping Mall Analogy
Before Azure AI Foundry, building enterprise AI meant going to many different vendors. Azure AI Foundry is the shopping mall — Microsoft brought everything into ONE platform: models from OpenAI, Meta, Mistral; testing tools; security controls; and deployment systems.
KEY DISTINCTION: Azure AI Foundry = where organisations BUILD their own AI. Microsoft 365 Copilot = the ready-made AI assistant for daily work. Never confuse them on the exam.
EXAM TIP: "Building enterprise AI," "models from multiple providers," "centralised AI governance" → answer points to Foundry, not Copilot.
SS2 · Q14
The Key Benefit of Foundry
📋 Original Microsoft Learn Question
Question 14 of 50
You need to explain the value of Microsoft Foundry to stakeholders. Which statement accurately describes a key benefit of Foundry?
Select ONE correct answer:
Your comment (SS2): I believe the last answer is correct, but I don't know why — can you explain in very simple terms?
The biggest benefit of a shopping mall is ONE management office, ONE security team, ONE set of rules for everyone. Foundry gives your organisation ONE place to approve, monitor, and enforce AI policy.
✗Foundry absolutely requires AI models — option A is factually false.
✗Foundry supports many frameworks beyond open-source — option B is false.
✗M365 productivity apps = Copilot's job, not Foundry — option C is wrong tool.
✓Centralised management and governance — CORRECT. This is Foundry's core value.
EXAM TIP: Any Foundry "benefit" question → governance, centralisation, multi-provider model access. Answers about M365 apps belong to Copilot.
SS3 · Q15
Security in Foundry — RBAC vs Anonymous Access
📋 Original Microsoft Learn Question
Question 15 of 50
You are evaluating Microsoft Foundry for building enterprise AI solutions. Which statement accurately describes a security benefit of using Foundry?
Select ONE correct answer:
Your comment (SS3): The answer is between option 3 and 4 — how do I ensure I do not confuse them in the final exam?
MEMORY TRICK: Ask yourself — "Does this sound like enterprise security or public internet access?" RBAC via Entra ID = enterprise control. Anonymous = no identity, no control — a CBN violation for any Nigerian bank.
EXAM TIP: When you see "anonymous access" in an enterprise AI security question, eliminate it immediately. Enterprise always requires identity-based access.
SS4 · Q16
Foundry Capabilities — Classifying Customer Feedback
📋 Original Microsoft Learn Question
Question 16 of 50
You are using Microsoft Foundry to design a solution for a company. The company needs to automatically classify customer feedback as positive, neutral, or negative. Which Foundry capability should you select?
Select ONE correct answer:
Your comment (SS4): I don't know this answer — can you explain all 4 options and draw a simple Foundry architecture?
AAzure Language — pre-built general-purpose sentiment. Too generic for 3-category custom classification.
BAzure OpenAI GPT-4 — powerful but expensive for high-volume repetitive daily classification tasks.
C ✓Custom classification model in Foundry — CORRECT. SPECIFIC + REPETITIVE + AUTOMATED with YOUR labelled data.
DAzure AI Search — finds information in documents; does not classify messages.
Foundry Architecture — 3 Layers
1
Models Layer
GPT-4, Llama, Mistral, custom models — the AI brains.
2
Tools Layer
Azure Language, Azure AI Search, custom training pipelines.
3
Governance Layer
Microsoft Entra ID (RBAC), content filters, monitoring dashboards.
EXAM TIP: SPECIFIC + REPETITIVE + AUTOMATED with your own data = custom model in Foundry. GENERAL + one-off = prebuilt model.
SS15 · Q31
Copilot Studio + Partner Models — Azure Marketplace Fix
📋 Original Microsoft Learn Question
Question 31 of 50
Your company plans to pilot Microsoft Copilot Studio for customer support and deploy a protected partner model from Microsoft Foundry Models by using the same Azure subscription. During deployment, you receive a notification that the subscription is NOT authorized for partner model purchases. You need both initiatives to use pay-as-you-go billing with the existing Azure subscription without changing billing models. What should you do?
Select ONE correct answer:
Your comment (SS15): I do not even understand the question — breakdown the answer and draw a diagram showing the relationships.
✗Prepaid capacity — changes billing model. Question says NOT to do this.
✗New subscription — question says use the EXISTING subscription.
✓Enable Azure Marketplace purchases — partner models are sold through Marketplace. "Not authorised" = Marketplace disabled by IT security policy.
✗Copilot Studio message packs — only addresses Copilot Studio billing, not the partner model problem.
EXAM TIP: "Not authorised for partner model purchases" error → always fix by enabling Azure Marketplace purchases on the existing subscription.
Section 2 · SS11
Microsoft 365 Copilot Licensing
You did not know the answer and asked for resources. This section clears it up completely.
SS11 · Q27
M365 Copilot Licensing Options
📋 Original Microsoft Learn Question
Question 27 of 50
Select the answer that correctly completes the sentence. Microsoft 365 Copilot licensing options include [Target_A].
Select ONE correct answer:
Your comment (SS11): I do not know the answer. I suspect it is per-user subscription. Please find YouTube or Microsoft Learn resources.
M365 Copilot is licensed per individual user — each person needs their own Copilot licence added to an eligible M365 subscription. Same model as Office 365 — per person, not per device.
| Product | Billing Model | Key Rule |
|---|---|---|
| M365 Copilot | Per-user add-on licence | NOT included in any standard M365 plan — must purchase separately |
| Azure OpenAI | Pay-as-you-go (tokens) | Charged per 1,000 tokens consumed |
| Foundry partner models | Via Azure Marketplace | Subscription must have Marketplace purchases enabled |
| Azure AI production workloads | Prepaid capacity commitment | Best for stable, predictable monthly usage |
EXAM TIP: M365 Copilot = per-user add-on. Azure OpenAI = pay-as-you-go. Copilot is NEVER free or automatically included in any M365 plan.
Section 3 · SS5, SS6, SS7, SS9, SS10
Responsible AI — The FAITHP Framework
These questions appear repeatedly. Master FAITHP and 4–5 exam questions become straightforward.
SS7
Microsoft's 6 Responsible AI Principles — FAITHP
Your comment (SS7): Please list what constitutes Microsoft responsible AI principles, create an Acronym if possible, sketch a schematic.
Use FAITHP — write this on your scratch paper the moment the exam starts. P is highlighted in red because it is the most changed letter from FATHIR:
F
Fairness
No bias based on race, gender, age, or location.
A
Accountability
People must be answerable for how AI behaves.
I
Inclusiveness
AI benefits everyone — no one excluded.
T
Transparency
People should know when they interact with AI.
H
Human Oversight
Humans must be able to intervene and correct AI. (Reliability & Safety)
P
Privacy & Security
Protect personal data. P = Privacy — unambiguous.
EXAM TIP: Every Responsible AI question maps to one or two FAITHP letters. Find the keyword in the scenario first, then match it to the letter.
SS5 · Q17
Sensitive Personal Data — Privacy Principle
📋 Original Microsoft Learn Question
Question 17 of 50
A company needs to ensure an AI solution that processes customer support transcripts aligns with Microsoft responsible AI principles and reduces risk related to sensitive personal data. What should the company do?
Select ONE correct answer:
Your comment (SS5): I chose "Limit how long personal data is stored" — it was just an educated guess. Please provide more insight.
Keyword: SENSITIVE PERSONAL DATA → maps to P in FAITHP (Privacy & Security). Data minimisation — storing only what is necessary — is the fundamental NDPR/GDPR control.
EXAM TIP: Sensitive data + privacy risk → minimise data, limit storage time, retain only what is necessary. This is always the P answer.
SS6 · Q18
Responsible AI for Product Recommendations — Trap Alert
📋 Original Microsoft Learn Question
Question 18 of 50
You are deploying an AI service that generates personalised product recommendations based on customer purchase history. You need to apply responsible AI principles before deployment. Which two explanations should you provide? Each correct answer presents part of the solution.
Select TWO correct answers:
Your comment (SS6): Where is the trick in this question? What mistakes can an exam taker make?
TRAP ✗"Removes human oversight" — sounds like efficiency but VIOLATES H in FAITHP. Always eliminate this.
TRAP ✗"Shifts responsibility to Microsoft" — accountability cannot be outsourced (A in FAITHP).
✗"Removes governance controls" — the word "removes" is always a red flag.
✓Transparency in data usage — T in FAITHP.
✓Prevents biased recommendations — F in FAITHP.
MULTI-SELECT RULE: Immediately eliminate any option using "removes oversight," "no human needed," or "shifts responsibility to Microsoft." These are ALWAYS wrong.
SS9 & SS10 · Q24 & Q25
Applying FAITHP to Specific Scenarios
📋 Original Microsoft Learn Question
Question 24 of 50
You are sponsoring an AI solution that will automate eligibility assessments for a financial assistance program. The solution will directly influence decisions that affect customers. You need to ensure the AI solution meets the Microsoft Responsible AI Standard for consistent performance and organisational oversight. Which two principles should you prioritise? Each correct answer presents part of the solution.
Select TWO correct answers:
📋 Original Microsoft Learn Question
Question 25 of 50
You are leading the deployment of an AI solution that generates customer-facing product recommendations. You need to establish governance controls that reduce the risk of biased outputs and ensure that customers understand how AI is being used. Which two principles should you prioritise?
Select TWO correct answers:
Your comments (SS9 & SS10): These are similar to Q19. Once SS7 is answered, these will be clear.
| Scenario Keyword | FAITHP Letter | Principle |
|---|---|---|
| Bias, discrimination, unfair treatment | F | Fairness |
| Governance structure, who is responsible | A | Accountability |
| Accessibility, no one excluded | I | Inclusiveness |
| Explain, customers should know, understandable | T | Transparency |
| Human control, consistent performance, safety | H | Human Oversight |
| Personal data, sensitive info, NDPR, GDPR | P | Privacy & Security |
Section 4 · SS8, SS12, SS16, SS24
AI Governance — Council vs Champions
Two distinct governance structures. Know the difference and you will not confuse them.
SS8 · Q20
The AI Council — Why Business, Legal, and IT?
📋 Original Microsoft Learn Question
Question 20 of 50
You are defining an AI governance framework for how AI solutions will be evaluated, approved, and monitored across your organization. Leadership requires cross-functional oversight and alignment with Microsoft responsible AI principles. You need to establish a governance control that ensures consistent policy enforcement and risk management. What should you implement?
Select ONE correct answer:
Your comment (SS8): I believe the answer is B. Can you explain why business, legal and IT are required as part of the AI council?
| Member | Why Essential | Nigerian Example (GTBank) |
|---|---|---|
| Business Leaders | Define what problem AI should solve commercially. | Head of Retail Banking defines the credit scoring use case. |
| Legal / Compliance | Ensure compliance with NDPR, CBN guidelines. | Legal Counsel reviews AI for CBN compliance before deployment. |
| IT / Technology | Evaluate feasibility, security, infrastructure. | IT Director confirms Azure integration and data security. |
EXAM TIP: AI governance = cross-functional. Any option that puts AI decision-making in ONE department's hands is always wrong.
SS12 · Q28
AI Council vs AI Champions
📋 Original Microsoft Learn Question
Question 28 of 50
Select the answer that correctly completes the sentence. An AI champions program is primarily designed to [Target_A].
Select ONE correct answer:
Your comment (SS12): Are AI Council and AI Champions different things? Please define them with examples.
| AI Council | AI Champions Programme | |
|---|---|---|
| What it is | Formal governance board of senior leaders | Network of peer advocates across departments |
| Decision power | YES — final authority on AI going live | NO — enablers only, not decision-makers |
| Nigerian Example | Dangote Cement: CDO + Legal + Operations + Risk | GTBank: 30 AI Champions across all banking divisions |
| Analogy | Board of Directors of AI | Community Development Officers |
EXAM TIP: Council = top-down governance (approve, oversee, policy). Champions = bottom-up adoption (encourage, enable, peer advocacy).
SS16 & SS24 · Q32
Building an AI Champions Network
📋 Original Microsoft Learn Question
Question 32 of 50
Your company is deploying Microsoft 365 Copilot. Early pilot results show inconsistent usage across departments. You need to create a network of internal advocates to promote effective and responsible use. What is the best approach to achieve the goal?
Select ONE correct answer:
Your comment (SS16 & SS24): Breakdown this as a design process — I need it as part of a potential customer deployment document.
1Identify: 2–3 motivated volunteers per department — tech-curious, respected by peers.
2Train: Advanced Copilot training beyond what regular employees receive.
3Define Expectations: Monthly peer sessions, feedback collection, quarterly use case sharing.
4Enable Peers: Lunch-and-learn sessions, Teams Q&A channel.
5Feedback Loop: Champions report to AI Council monthly.
6Celebrate & Scale: Recognise Champions publicly. Track adoption by department.
Section 5 · SS19, SS21, SS22
Pretrained, Fine-Tuned & RAG
One of the most important topics for both AB-730 and AB-731. The building analogy makes it permanent.
SS19 & SS21 · Q37 & Q39
Pretrained vs Fine-Tuned Models
📋 Original Microsoft Learn Question
Question 37 of 50
A company is deciding whether to use a pretrained model or a fine-tuned model. What is a key characteristic of a fine-tuned model compared to a pretrained model?
Select ONE correct answer:
📋 Original Microsoft Learn Question
Question 39 of 50
Your company is deploying a domain-specific AI agent to review legal contracts. Early testing with a pretrained LLM produces inconsistent terminology and formatting, even when using detailed prompts. You are considering whether to fine-tune the model. Which statement best describes how fine-tuning differs from using the pretrained model as-is? Select the BEST answer.
Select ONE correct answer:
Your comment (SS19 & SS21): What is the difference between pre-trained and fine-tuned? Can you fine-tune without pre-training? Use a pharmaceutical company example.
Pretrained model = A ready-built building. Foundation, walls, wiring — all in place, built over years. You move in immediately.
Fine-tuning = Hiring an interior designer to customise the building for YOUR business. The core building is unchanged — it is now configured for your purposes.
Can you fine-tune without pretraining? NO. You cannot customise a building that does not exist.
THE DISCRIMINATOR: "Modifies internal weights using domain-specific training data" = Fine-tuning. "Retrieves information from documents at query time" = RAG. These two phrases are your exam compass.
SS22 · Q41
What is RAG — And Is a Copilot Agent a RAG?
📋 Original Microsoft Learn Question
Question 41 of 50
You have an AI assistant that uses retrieval-augmented generation (RAG) to retrieve information from internal and external knowledge sources before generating a response. Which step in the RAG process primarily reduces the risk of fabricated answers?
Select ONE correct answer:
Your comment (SS22): What is RAG? Does a Copilot Agent with a knowledge base = a RAG? Are there several types?
The Library Pass Analogy: A consultant knows everything up to the day they were trained. You give them a library pass — before answering, they find the relevant documents, read them, then answer. That is RAG.
YES — a Copilot Agent with a Knowledge Base IS RAG in practice. When you connect a Copilot Studio Agent to SharePoint, it retrieves content from your documents before generating its response.
EXAM TIP Q41: Which step REDUCES fabrication? Always the RETRIEVAL step — because grounding in real documents prevents the model from making things up.
Section 6 · SS13, SS14
The Azure AI Ecosystem — Three Layers
How Azure AI, Foundry, and Copilot relate to each other.
SS13 & SS14 · Q29 & Q30
Azure AI + Copilot + Foundry Relationship
📋 Original Microsoft Learn Question
Question 29 of 50
Your company is evaluating Microsoft 365 Copilot. During a planning meeting, stakeholders ask whether prompts and responses generated in Copilot are used by Microsoft to train its AI models. Which statement accurately describes how Microsoft uses Copilot prompts and responses?
Select ONE correct answer:
📋 Original Microsoft Learn Question
Question 30 of 50
Your company is planning a large-scale rollout of an Azure AI service for a production workload that has stable and predictable monthly usage. The finance department requires consistent monthly spending and minimal billing variability. You need to select the most appropriate subscription model. What should you choose?
Select ONE correct answer:
Your comment (SS13 & SS14): How does a customer use Copilot or Foundry to train AI on their own data? What is the relationship between Azure AI and Copilot? Where does Foundry come in?
The Restaurant Analogy — Three Layers
1
Azure AI — Kitchen Infrastructure
Foundational cloud infrastructure. Everything else runs on top of Azure.
2
Azure AI Foundry — Chef's Kitchen
Where builders BUILD custom AI solutions, fine-tune models, and govern deployments.
3
Microsoft 365 Copilot — The Restaurant
The end product business users consume — AI in Teams, Word, Excel, Outlook.
KEY FACT (Q29): Microsoft does NOT use your Copilot prompts or responses to train its foundation AI models. Your data stays yours.
EXAM TIP (Q30): Stable + predictable + production workload = Prepaid capacity commitment. Variable + experimental = Pay-as-you-go.
Section 7 · SS18, SS23
Data Security — Purview & Microsoft Entra
Two security tools that protect different things. Know which does what.
SS18 · Q33
Microsoft Purview DLP — Configure Before Copilot
📋 Original Microsoft Learn Question
Question 33 of 50
Select the answer that correctly completes the sentence. To reduce data security and privacy risks when adopting Microsoft 365 Copilot, your organization must ensure that [Target_A].
Select ONE correct answer:
Your comment (SS18): I suspect Purview DLP is the answer but WHY? Give examples and an analogy.
The Nightclub Bouncer Analogy: Copilot is a very efficient assistant who can access EVERYTHING in your M365 environment. Microsoft Purview is the bouncer — it labels content (CONFIDENTIAL, HIGHLY RESTRICTED) and DLP policies enforce rules so Copilot cannot surface restricted files to unauthorised users.
Nigerian Example: Templars Law Firm — without Purview labels, a junior associate asks Copilot to "summarise all correspondence related to our biggest deals." Copilot surfaces client files the associate has no business seeing. With Purview DLP, RESTRICTED content is protected.
EXAM TIP: Any question about Copilot + DATA SECURITY = Microsoft Purview. Always configure Purview BEFORE deploying Copilot.
SS23 · Q41
Microsoft Entra — Identity and Conditional Access
📋 Original Microsoft Learn Question
Question 41 of 50
Your company plans to deploy Microsoft 365 Copilot to 200 employees. A company security policy requires that access to Microsoft 365 services be limited to compliant devices and protected by strong authentication controls. You need to ensure that Copilot access meets the requirements. Which two actions should you perform? Each correct answer presents part of the solution.
Select TWO correct answers:
Your comment (SS23): What does Microsoft Entra mean? Is it one of those trick or distraction options?
Microsoft Entra is NOT a distractor — it is the correct answer. Microsoft Entra = Microsoft's identity and access management platform (formerly Azure Active Directory). It is the HR department of your IT systems — deciding which users and devices can access which systems.
✓Conditional Access — only company-managed compliant devices can access Copilot. Personal laptops blocked.
✗Token limit — about response LENGTH not security.
✓MFA — even if a password is stolen, attacker cannot access Copilot without the second factor.
✗Semi-Annual Channel — about update cadence, not security.
✗Web-grounded Copilot — switches to public internet. Would REDUCE enterprise security.
EXAM TIP: "Device compliance + strong authentication" for Copilot = Microsoft Entra: Conditional Access + MFA. Never token limits or update channels.
Section 8 · SS17
AI Challenges — Fabrication, Reliability & Hallucination
Three terms that appear in exam questions. One clear rule keeps them straight.
SS17 · Q34
Fabrication vs Hallucination vs Reliability
📋 Original Microsoft Learn Question
Question 34 of 50
A company deploys generative AI to draft internal knowledge base articles. During testing, the generative AI system produces clear and accurate drafts in some test cases and incomplete or inconsistent drafts during other tests. Which challenge of generative AI is this an example of?
Select ONE correct answer:
Your comment (SS17): What is the difference between Fabrication, Reliability and Hallucination? Where does each apply?
| Term | Definition | Nigerian Example |
|---|---|---|
| Fabrication / Hallucination | AI generates false information presented as true with full confidence. | Copilot says "CBN inflation report shows 8%" — but that statistic does not exist in the report. |
| Reliability Issue | AI performs inconsistently — correct sometimes, wrong other times with identical inputs. | AI drafts excellent articles on Monday, incomplete ones on Friday — same prompts. |
| Bias | AI systematically treats certain groups unfairly due to training data patterns. | AI loan tool consistently rejects applicants from certain Nigerian states. |
| Scalability | AI cannot handle the required volume or speed of processing. | AI system crashes when 10,000 users query simultaneously. |
THE DISCRIMINATION RULE:
"AI states a fact that doesn't exist" → FABRICATION
"AI is correct sometimes, wrong other times with same inputs" → RELIABILITY
"AI consistently disadvantages a group" → BIAS
"AI cannot handle the volume" → SCALABILITY
"AI states a fact that doesn't exist" → FABRICATION
"AI is correct sometimes, wrong other times with same inputs" → RELIABILITY
"AI consistently disadvantages a group" → BIAS
"AI cannot handle the volume" → SCALABILITY