AB-731 Licensing and Cost Models Explained (2026)

Last updated: October 2026. Written by the Examinotion Team and fact-checked against the official Microsoft Learn AB-731 study guide and Microsoft Learn licensing and billing documentation.

TL;DR AB-731 tests how generative AI is paid for. Know the three Copilot routes (included, pay-as-you-go, paid add-on), the two Foundry Tools models (pay-as-you-go and commitment tiers), how tokens drive cost, and which budgets only alert rather than block spending. Learn how each model works, not price points.

Exam AB-731: Microsoft Certified: AI Transformation Leader is Microsoft's beginner-level certification for business leaders who plan and steer AI adoption. Its skills outline, last updated on 22 July 2026, asks you to explain generative AI cost drivers, understand Copilot licence types and understand Foundry Tools subscription models [1]. Those three bullets are often underrated, because most candidates prepare for strategy and responsible AI and then meet a licensing scenario on exam day.

This guide explains each costing and licensing idea at the level AB-731 tests it, using Microsoft's own documentation. You can check the live sources on the Microsoft Learn AB-731 certification page and the official AB-731 skills outline. For the wider exam picture, start with our AB-731 complete study guide, and use the AB-731 practice tests to check what you retain.

Where licensing and cost sit in the AB-731 skills outline

AB-731 has three domains, and cost and licensing appear in two of them [1].

Domain (as published)	Weight	Where cost and licensing appear
Identify the business value of generative AI solutions	35-40%	Cost drivers in generative AI usage, including tokens and ROI
Identify benefits, capabilities and opportunities for Microsoft's AI apps and services	35-40%	Capabilities of Copilot versions, Foundry benefits such as scalability
Identify an implementation and adoption strategy for Microsoft's AI apps and services	20-25%	Copilot licence types, Foundry Tools subscription models, impacts on data, security, privacy and cost

The exact wording of the three bullets matters, because the exam is written from them [1]:

"Explain the cost drivers in generative AI usage, including tokens and return-on-investment (ROI) considerations"
"Understand Copilot license types, including pay-as you go, monthly, and included with Microsoft 365 subscription"
"Understand Foundry Tools subscription models, including pay-as-you-go and commitment tiers"

Microsoft's change log for July 2026 lists three Minor rows, including the adoption-planning group, and it does not say which individual bullets changed [1]. Treat all three bullets as in scope. The exam lasts 45 minutes [2], a score of 700 or greater is required to pass [1], and Microsoft ties the price to the country or region in which the exam is proctored, so check your own region's price when you book [2].

A naming note before you go further. Microsoft's pages now mix "Microsoft 365 Copilot" and "Microsoft Copilot", and Microsoft's current pages use the names Microsoft Foundry (formerly Azure AI Foundry) and Foundry Tools (formerly Azure AI Services) [1][14]. The study guide itself uses both Copilot names, so be comfortable with either.

Generative AI cost drivers: tokens, models and usage

Generative AI is usage-based by nature. A language model reads your input and writes its output in small units called tokens, and the bill follows the tokens. Microsoft puts it this way: models process inputs by breaking them down into tokens, and the billing unit and rate can vary by model, deployment type and meter [14].

The main cost drivers to know:

Cost driver	Why it moves the bill	Exam-style takeaway
Input tokens	Longer prompts and more grounding content mean more tokens read	Concise prompts and relevant context cost less
Output tokens	Output tokens need more processing capacity than input tokens [11]	Long generated answers put more load on capacity than long questions
Model choice	Rates vary by model [14]	A larger model is not automatically the right one for a simple task
Deployment type	Pay-per-token versus reserved capacity [13]	Match the type to how predictable the workload is
Usage volume	More users and more calls mean more tokens	Pilot first, then scale on evidence
Extra services	Hosting, fine-tuning and tools add separate charges [14]	Fine-tuned model hosting is charged hourly even when unused [14]

Standard, provisioned and batch deployments

Microsoft Foundry offers several ways to pay for model capacity, and AB-731 expects you to know the broad logic rather than the product SKUs [13].

Standard (pay per token): you pay for the tokens you use. Microsoft positions this for variable or unpredictable traffic, with no reserved capacity [11].
Provisioned (reserved capacity): you reserve processing capacity and are charged hourly whether or not the deployment is handling requests, which suits predictable, high-volume workloads [11][12].
Batch: Microsoft lists a 50% discount against Global Standard in exchange for a 24-hour target turnaround (it can take longer), so it suits work that is not time-sensitive [13].

The leadership-level rule of thumb is simple. Uncertain or spiky demand points to pay-per-token. Steady, heavy, latency-sensitive demand points to reserved capacity. Work that can wait points to batch.

Building an ROI case

AB-731 also asks about return on investment. Microsoft's Cloud Adoption Framework says an AI strategy should define how the organisation measures the value of its AI investments, and that the challenge is deciding where AI can deliver value and whether that value justifies the cost [15]. It also notes that Microsoft Copilots typically provide the shortest timelines to return on investment, in days to weeks, and advises adding 20-30% contingency time to estimates [16].

For the exam, remember the shape of a sound business case:

Pick a specific, measurable use case rather than "use AI everywhere".
Establish a baseline (time per task, error rate, cost per transaction).
Run a pilot and measure the change against the baseline.
Count the full cost: licences or usage charges, change management, training, and governance.
Scale only where measured value exceeds total cost.

The wrong answers in scenario questions usually skip the baseline or the pilot. Our AB-731 AI adoption framework guide covers the surrounding adoption stages in more depth.

Copilot licence types: included, pay-as-you-go and monthly

The outline lists three Copilot licence types: "pay-as you go, monthly, and included with Microsoft 365 subscription" [1]. No single Microsoft page lists those three exact labels together, so here is how Microsoft's licensing documentation maps onto them. This mapping is our reading, and you should confirm it against the pages in the Sources list.

Outline label	What it most likely means	What Microsoft's documentation says
Included with Microsoft 365 subscription	Copilot Chat	Microsoft 365 Copilot Chat is included in your Microsoft 365 subscription at no additional charge [9]
Pay-as-you-go	Metered access billed through Azure	Optional pay-as-you-go access to work-based chat [9], and to agents without a full licence [3]
Monthly	The paid per-user Copilot licence	The Microsoft 365 Copilot add-on licence for frequent users [9]. Microsoft's licence options page lists other paid routes with their own terms, so check the commitment period on the page [8]

The "monthly" label is the loosest of the three, because Microsoft's pages do not define it and some paid routes carry annual commitments [8]. Read scenario wording closely. If a question describes a small group of heavy users who need Copilot embedded in Word, Excel and Outlook all day, the paid licence is the intended answer. If it describes an organisation that wants to test agent scenarios before committing, pay-as-you-go is the intended answer. If it describes everyone getting Copilot Chat as part of their existing subscription with no additional licence charge, that is the included Copilot Chat tier. Note that Microsoft says agent usage in Copilot Chat can still incur usage-based charges in some cases [9]. For the product differences behind these tiers, see our Microsoft Copilot types explained comparison.

How Copilot pay-as-you-go works

Microsoft describes pay-as-you-go as flexible, usage-based billing that lets an organisation use AI capabilities without the commitment of a full licence [3]. It applies to Microsoft Copilot Chat, SharePoint agents and the Microsoft Copilot Retrieval API (in preview) [3]. Microsoft lists two common use cases [3]:

Establish usage patterns: administrators can understand adoption of new apps to decide whether prepaid licences make financial sense for the organisation.
Scalability: the organisation scales usage with demand and pays only for what is used.

The setup has points that make good exam questions:

It is off by default. Pay-as-you-go is disabled until an administrator turns it on [3].
Two steps, not one. You create a billing policy, which is a unique billing identifier linking users to billing rules and an Azure subscription, and then you connect it to a Copilot service. Creating the policy alone does not complete the setup [3].
Billing runs through Azure. Usage is charged to the Azure subscription you specify, so you need a subscription in the same tenant, a resource group, and Owner or Contributor rights [3][4].
Limits exist. A tenant can have up to 50 billing policies [4].
Who can manage it. Global Administrator, Billing Administrator and AI Administrator can manage pay-as-you-go, and Global Reader has read-only access [3].
Budgets alert, they do not block. Microsoft states that the system does not enforce a budget or prevent your organisation from exceeding it [4].

That last point is the classic trap. If a question asks how to guarantee spending cannot exceed a figure, "set a budget" is the wrong answer for pay-as-you-go. A budget is an alerting tool, not a hard cap.

Copilot Credits and Copilot Studio

Agents built in Copilot Studio use a common currency. Microsoft says that from 1 September 2025 the common currency for agents changed from messages to Copilot Credits [7]. Credits can be obtained through pay-as-you-go meters, a one-year prepurchase plan or prepaid capacity packs [7].

Three details are worth remembering:

With pay-as-you-go, the organisation pays only for the credits its agents actually use during the month [7].
With prepaid capacity packs, capacity is enforced monthly, unused credits do not carry over, and exceeding capacity can lead to service denial [7].
Agents used within Copilot Chat, Teams or SharePoint for certain answers and Microsoft Graph grounding are zero-rated for users who hold a Microsoft 365 Copilot licence [7].

Microsoft's usage-based billing documentation adds that organisations can combine methods, and that spending policies, organisation-level and user-level limits and alerts are available [6]. Microsoft describes these as spending limits, whereas pay-as-you-go budgets only alert, so do not generalise from one tool to the other.

Older articles and some Microsoft pages still show per-message pricing for these meters [5]. Because the unit changed, avoid memorising any price points. AB-731 tests which model fits which situation, not the tariff.

Foundry Tools subscription models: pay-as-you-go versus commitment tiers

The second licensing bullet covers Foundry Tools, the family formerly known as Azure AI Services. Microsoft says common billing approaches include pay-as-you-go and commitment tiers [14].

	Pay-as-you-go (Standard)	Commitment tier
How you pay	Per use, with no upfront commitment	A fixed fee for a set volume, at a discounted rate compared with Standard [10]
Best for	Unpredictable or experimental workloads	Predictable, steady volumes where a lower unit rate pays off
Cost certainty	Varies with usage	Predictable total cost for the committed volume [10]
Risk	Higher unit rate at scale	Paying for capacity you do not use

The rules for commitment tiers are specific, which makes them popular in exams. These apply to hosted API and connected container plans; disconnected containers use a separate calendar-year plan charged in full upfront [10]:

The commitment period is a calendar month. The first month is pro-rated, and later months are charged on the first day.
Once purchased, the plan is non-refundable.
Usage above the committed quota is charged at the overage rate shown in the tier.
You cannot change the plan within the current month.
You cannot use commitment tier pricing with a multi-service resource, so you must use single-service resources.
Commitment tiers apply to selected features, such as speech to text, text to speech, text translation, some Azure Language features, Vision OCR and Document Intelligence [10].
To stop, you set the plan to not auto-renew before midnight UTC on the last day of the month, and you then revert to Standard pricing [10].

The pattern for a scenario question is straightforward. If a company has a stable, high volume of document processing or transcription, a commitment tier lowers the unit cost. If the company is still experimenting, or volumes swing from month to month, pay-as-you-go avoids paying for unused commitment. For a refresher on what the individual services do, see our Azure AI Vision vs Azure AI Search vs Microsoft Foundry guide.

The word "commitment" means different things

Microsoft uses "commitment" for more than one idea, and mixing them up is an easy way to lose a mark [10][12]:

Term	What it is	Key rule
Foundry Tools commitment tier	Discounted fixed monthly volume for selected services	Calendar month, non-refundable [10]
Azure Reservation for provisioned throughput	A one-month or one-year reservation for reserved model capacity	Discounted hourly rate, but does not guarantee capacity, and cancelling or exchanging might incur fees [12]
Copilot Studio prepurchase plan	A one-year prepaid option bought in the Azure portal	Prepaid credits [7]

AB-731 is a beginner-level leadership exam, so you will not be asked to size a reservation. You should recognise that a commitment trades flexibility for a lower rate.

Choosing the right model for the scenario

Most licensing questions on AB-731 are scenario questions. Use this decision table as a quick filter.

Scenario signal	Likely best fit
Everyone needs safe chat, no extra budget	Included Copilot Chat
A small team of heavy daily users	Paid per-user Copilot licence
Wants to trial agents before buying licences	Copilot pay-as-you-go
Agents with predictable monthly demand	Prepaid Copilot Credits capacity
Variable AI workload, still experimenting	Foundry pay-as-you-go or Standard deployment
Stable, high-volume service use	Foundry Tools commitment tier
Predictable, latency-sensitive model traffic	Provisioned deployment
Large job that can wait a day	Batch deployment

Exam-style scenarios and common traps

These are original practice scenarios written for this guide, not real exam questions. Each shows how the topics above can be tested.

Scenario 1. A finance director wants to trial Copilot agents with one department, with spend limited to a monthly figure. She sets a budget in the pay-as-you-go cost management page and tells the board spending cannot exceed it. What is wrong with her statement? Answer: the budget sends alerts at spend milestones but does not enforce a limit or stop usage [4].

Scenario 2. A company runs a steady, high volume of speech-to-text through a multi-service Foundry resource and wants a lower unit cost. A colleague suggests simply buying a commitment tier for that resource. What is the issue? Answer: commitment tier pricing cannot be used with a multi-service resource. They need a single-service resource [10].

Scenario 3. A pilot of a customer-support assistant shows a high bill because answers are very long. Which cost driver is most likely responsible? Answer: output tokens. Long generated answers add output tokens, which need more processing capacity than input tokens [11], and billing follows token usage [14].

Scenario 4. A leader wants to license every employee with a paid Copilot licence "to be safe". What does a sound adoption plan suggest instead? Answer: pilot with defined use cases and a baseline, use pay-as-you-go or the included Copilot Chat to learn usage patterns [3][9], and license heavy users once value is measured against the baseline [15].

The common traps to avoid:

Treating budgets as hard caps. Pay-as-you-go and Foundry budgets alert only [4][14].
Assuming commitments are refundable. Foundry Tools commitment plans are non-refundable [10].
Forgetting that pay-as-you-go is billed through Azure. It needs an Azure subscription and a billing policy [3][4].
Memorising prices. Prices vary by region and change, and the exam objectives name models, not tariffs.
Ignoring total cost. A valid ROI case includes training, change management and governance, not only licence fees.

How to prepare for the licensing and cost objectives

A short plan works well for a topic that is mostly vocabulary and trade-offs:

Read the three outline bullets, then the Copilot licence options and pay-as-you-go overview pages.
Build your own one-page comparison of the three Copilot routes and the two Foundry Tools models, from memory, and check it against the tables above.
Practise the scenario filter until you can pick a model in under 20 seconds.
Test yourself with scenario-style questions in the AB-731 practice tests, and read our guide to AB-731 case study questions for longer scenarios.
Fit the topic into a longer plan with our 30-day AB-731 study plan.

Practice reduces risk, but it cannot guarantee a pass. The goal is to understand the trade-offs well enough to reason through a scenario you have not seen before.

Frequently asked questions

Is licensing tested on AB-731?

Yes. The AB-731 skills outline, last updated on 22 July 2026, includes understanding Copilot licence types and Foundry Tools subscription models in the adoption domain, which carries 20-25% of the exam. Cost drivers, including tokens and ROI, also appear in the generative AI foundations content. Expect scenario questions about choosing a model, not price recall.

What is Microsoft 365 Copilot pay-as-you-go?

Pay-as-you-go is usage-based billing that lets an organisation use Copilot capabilities, such as Copilot Chat agents, without buying a full licence for every user. An administrator creates a billing policy linked to an Azure subscription, then connects it to a Copilot service. Usage is billed through Azure, and the feature is off by default.

What are Foundry Tools commitment tiers?

A commitment tier is a fixed monthly fee for a set volume of a selected Foundry Tools feature, offered at a discount compared with Standard pay-as-you-go pricing. The commitment period is a calendar month, the plan is non-refundable, and usage above the quota is charged at an overage rate. It suits steady, predictable workloads.

Can you use commitment tiers with a multi-service resource?

No. Microsoft states that commitment tier pricing cannot be used with a Foundry Tools or Microsoft Foundry multi-service resource. You must create a single-service resource for the feature you want to commit to. This is a likely detail for a scenario question, so remember it alongside the non-refundable rule.

How do tokens affect generative AI cost?

Models read input and write output in tokens, and billing follows token usage at rates that vary by model and deployment type. Output tokens need more processing capacity than input tokens, so long answers add load. Careful prompts, the right model size and sensible output lengths are the main levers for controlling cost.

Do Copilot pay-as-you-go budgets stop spending?

No. A budget sends email notifications at spend milestones, but Microsoft states that the system does not enforce the budget or prevent the organisation from exceeding it. Copilot Credits spending policies are different, because they can apply organisation-level and user-level limits. Do not treat a budget alert as a hard cap in an exam answer.

What are the three Copilot licence types in the AB-731 outline?

The outline names pay-as-you-go, monthly, and included with a Microsoft 365 subscription. The included route maps to Copilot Chat at no extra charge, pay-as-you-go is metered billing through Azure, and monthly most likely refers to the paid per-user Copilot licence. Microsoft does not define the three labels together on one page.

How much does the AB-731 exam cost?

Microsoft does not list a fixed amount on the certification page. It states the price depends on the country or region in which the exam is proctored, so the cost you see depends on where you sit it. Check the Microsoft Learn exam page and the Pearson VUE booking screen for your local price.

Conclusion

The licensing and cost objectives on AB-731 reward clear thinking more than memory. Know the three Copilot routes, the two Foundry Tools pricing models, the way tokens drive cost, and the difference between a budget that alerts and a policy that limits. Then practise choosing the right model for a business scenario until it feels routine.

When you are ready to test yourself, start practising for AB-731, use the AB-731 study guide to cover every domain, or browse the full catalogue of Microsoft AI exam preparation courses.

Sources

AB-731 study guide for Exam AB-731: AI Transformation Leader - Microsoft Learn, accessed 2026-10-23
Microsoft Certified: AI Transformation Leader - Microsoft Learn, accessed 2026-10-23
Microsoft 365 Copilot pay-as-you-go overview - Microsoft Learn, accessed 2026-10-23
Set up Copilot pay-as-you-go - Microsoft Learn, accessed 2026-10-23
Meters for Copilot pay-as-you-go - Microsoft Learn, accessed 2026-10-23
Usage-based billing and Copilot Credits - Microsoft Learn, accessed 2026-10-23
Copilot Studio billing and licensing - Microsoft Learn, accessed 2026-10-23
License options for Microsoft Copilot - Microsoft Learn, accessed 2026-10-23
Licensing and cost for Copilot extensibility - Microsoft Learn, accessed 2026-10-23
Foundry Tools commitment tier pricing - Microsoft Learn, accessed 2026-10-23
Provisioned throughput for Foundry Models - Microsoft Learn, accessed 2026-10-23
Provisioned throughput billing - Microsoft Learn, accessed 2026-10-23
Foundry deployment types - Microsoft Learn, accessed 2026-10-23
Plan and manage costs for Microsoft Foundry - Microsoft Learn, accessed 2026-10-23
Cloud Adoption Framework: AI adoption - Microsoft Learn, accessed 2026-10-23
Cloud Adoption Framework: Plan for AI adoption - Microsoft Learn, accessed 2026-10-23

AB-731 Licensing and Cost Models: Copilot, Pay-As-You-Go and Foundry Commitment Tiers (2026)