Last-minute reference for AB-731: three skill areas, the vocabulary, a product-matching table and the governance checklist. Business value and Microsoft AI apps carry about three quarters of the marks.
Exam facts
| |
|---|
| Duration | 45 minutes |
| Passing score | 700 / 1000 |
| Weights | Business value 35–40%; Microsoft AI apps and services 35–40%; Adoption 20–25% |
| Outline | July 22, 2026 |
| Languages | 13 |
Vocabulary
| Term | One line |
|---|
| Generative AI | Creates new content |
| Machine learning | Predicts or classifies from data |
| Pretrained model | General model, ready to use |
| Fine-tuned model | Pretrained model trained further on your examples |
| Token | Unit of text that drives usage cost |
| Prompt engineering | Better instructions, better output |
| Grounding | Tying answers to your data |
| RAG | Retrieve relevant content, then generate |
| Fabrication | Confident but invented answer |
Which Microsoft product
| Need | Answer |
|---|
| Secure chat for work accounts | Microsoft 365 Copilot Chat |
| Help inside Word, Excel, PowerPoint, Outlook, Teams | Copilot in the apps |
| Answers from my emails, meetings and files | Microsoft 365 Copilot with Microsoft Graph |
| Research report across web and work data | Researcher |
| Analyse spreadsheets and data files | Analyst |
| Low-code agent for a team | Copilot Studio |
| Extend Copilot with your data or actions | Microsoft 365 Copilot extensibility |
| Custom app, model choice, scale | Microsoft Foundry |
| Search and retrieve company content | Azure AI Search |
| Analyse images and video | Azure Vision in Foundry Tools |
Order of preference: buy, extend, build.
Responsible AI checklist
Fairness, reliability, safety, privacy, security, inclusiveness, transparency, accountability. Governance principles and an AI council set the rules.
Adoption checklist
- Adoption team: cross-functional.
- AI champions programme: peers in each team.
- Barriers: skills, fear, trust, policy, data, use cases.
- Check impacts on data, security, privacy and cost before rollout.
Licences
| Product | Models |
|---|
| Copilot | Pay-as-you-go; monthly; included with Microsoft 365 subscription |
| Foundry Tools | Pay-as-you-go; commitment tiers |
Traps
- Fine-tuning for current facts: use RAG.
- Predicting a number: machine learning, not generative AI.
- Always, never, guarantees: usually wrong.
- No developers: Copilot or Copilot Studio.
Practise in the practice test and read exam tips.
Decision shortcuts
| Situation | Choose |
|---|
| Content is frequent, language-heavy, reviewable | Generative AI |
| Need a forecast or a class label | Machine learning |
| Logic is fixed and exact | Plain automation |
| Need current company facts | RAG |
| Need a consistent style or behaviour | Fine-tuning |
| Need to cut cost | Shorter prompts, relevant content only, a smaller model |
Adoption order of play
- Set governance principles and an AI council.
- Pick use cases with clear business value.
- Check data, permissions and security.
- Pilot, measure and learn.
- Train, appoint champions and communicate.
- Scale, monitor cost and review policy.