October puts many calendar-year organizations in the final stretch of budget planning, and this year the task is harder than usual because a single line item labeled “AI” does not adequately enumerate the mix of licenses, consumption, governance, platforms, and delivery effort involved in your AI strategy. 

Use this comprehensive checklist to walk through the line items that belong in an AI budget and use the linked Microsoft tools and calculators to do the math. Treat every public list price referenced here as a starting point, then swap in your contracted rates, tax treatment, region, and currency once you model a specific scenario. 

1. Microsoft Copilot Licensing Costs 

Segment Users Before Buying Premium Seats

Microsoft Copilot Chat, often referred to as “Basic”, comes bundled with your eligible Microsoft 365 subscriptions at no extra cost for web-grounded chat. The paid Microsoft 365 Copilot license, “Premium”, adds work-grounded chat (WorkIQ) and Copilot across the Microsoft 365 apps, and Microsoft lists it at $30 per user per month under an annual commitment on its enterprise pricing page. 

We recommend against assigning the paid license to your entire workforce by default. Instead, group people by the work they do and the Microsoft 365 knowledge context they actually need: give Copilot Chat to people who need protected web research, drafting help, and occasional access to metered agents (i.e., frontline workers), and reserve Microsoft 365 Copilot for people who need work-grounded assistance across Word, Excel, PowerPoint, Outlook, Teams, and Microsoft 365 content (often your knowledge workers). Revisit that licensing mix on a quarterly basis as adoption, usage, and business-value evidence come in, rather than treating the first allocation as permanent. Also be sure to factor implementation, role-based training, change management, support, and value measurement into the same budget. The license allocation is the simple transaction; the work that turns access into an improved process still needs its own funding. 

Build a Copilot Credit Pool for Code, Autopilot, and Cowork

Microsoft now separates everyday AI assistance from advanced, long-running agentic work. The user subscription license (USL) covers the first category at a fixed price, while usage-based billing (UBB) through Copilot Credits covers Cowork, Code, Autopilot, and certain frontier models. The premium Copilot user subscription license is required to access any of the UBB experiences. 

Treat each of these three capabilities as their own consumption lines rather than folding them into a flat per-user estimate.  

1. Copilot Cowork consumption varies with the models, runtime, context, and tools a given scenario uses, so model it by role, task frequency, and task complexity with the official Copilot Credit Estimator instead of a flat assumption. Hint: the Premium Copilot experience, especially when factoring in the advanced Anthropic models, has improved greatly. Have users experiment with Copilot before defaulting to Cowork for “advanced” workloads. 

2. Copilot Code and Copilot Autopilot are both usage based, and Microsoft has not yet published a dependable flat per-user rate for either, so start both with a controlled exploration pool, spending limits, and a plan to refine the forecast once your tenant generates real usage data and Microsoft publishes applicable rates.  

Whichever capability you turn on, assign a cost owner before you enable it; Microsoft’s cost-management controls support spending policies, user and group limits, alerts, and reporting by service, agent, user, and funding source. 

2. AI Agent Development and Consumption Costs 

Use Agent Builder for Low-Risk, Low-Cost Use Cases

Microsoft 365 Copilot Agent Builder supports lightweight internal agents, and for a user with a paid Microsoft 365 Copilot license, building and using these agents through the Copilot Chat harness carries no added Copilot Credit charge, subject to license terms and fair-use limits. Agent Builder also keeps the design scope narrower than a full Copilot Studio project: it does not author actions that integrate external services, so once a scenario needs low-code actions, connectors, or workflows, it moves into Copilot Studio territory. These more advanced agents require non-technical efforts, and you will need to budget for the people who define the process, curate knowledge, test answers, manage permissions, publish the agent, support users, and review performance. 

Classify Each Copilot Studio Agent by Harness

Copilot Studio now requires a harness-level budget decision. Under the Standard harness, a licensed Microsoft 365 Copilot user can use an internal agent in Microsoft 365 without an added Copilot Credit charge, but a user without that license consumes credits, and Microsoft also charges credits when a team uses the agent outside Microsoft 365 channels or invokes features the license doesn’t cover. Under the GitHub Copilot harness, building, testing, evaluating, and running the agent will consume Copilot Credits regardless of whether a user holds a Microsoft 365 Copilot license, so fund design, build, test, and production activity for the entire user base that will create or use these agents. 

Use the Copilot Studio pricing page to model Standard-harness activity, the Copilot Credit Estimator for Cowork and GitHub Copilot harness scenarios, and Microsoft’s guidance on managing GitHub Copilot harness costs to track consumption once agents reach production. Keep separate environments for exploration, test, and funded production, and assign credit limits and an accountable owner to each environment and agent. 

3. AI Governance and Agent 365 Licensing 

License the Humans Connected to Governed Agents

Microsoft Agent 365 adds premium observability, governance, and security for agents, and Microsoft licenses it per user rather than per agent. The standalone product lists at $15 per user per month on the Agent 365 licensing FAQ, and it’s included in Microsoft 365 E7; one license can cover the multiple agents associated with that user. 

Calculate the license count by identifying the people connected to agent activity: administrators and other people who manage, sponsor, or own agents that use Agent 365 premium capabilities; users who delegate access to on-behalf-of agents; and the owner, sponsor, or manager of any agent that acts under its own identity, including a task agent. 

Every Microsoft Cloud subscription includes a foundational Agent 365 layer with agent identity, registry visibility, basic usage insights, and core administrative actions at no extra cost, while premium analytics, policy, lifecycle, security, and data controls require an Agent 365 license. Agent 365 also gives every agent a distinct Microsoft Entra Agent ID and applies Entra controls such as permissions, consent, Conditional Access, and lifecycle operations, a topic substantial enough to deserve its own article; for this budget, the scope to track is license count and implementation effort. We recommend that every organization purchase at least one administrator license to cover these functions. 

4. Hidden AI Costs: Power Platform, Azure, and Data 

Identify the Platform Services Behind the AI Use Case

AI work often surfaces costs that belong to the surrounding application or data platform, and we recommend accounting for each of these before finalizing and approving a budget. 

1. Power Platform Premium covers makers and users who need premium connectors, Dataverse, attended automation, or production-grade applications and workflows; match licenses to the planned process using the Power Apps and Power Automate pricing pages. Microsoft pools Dataverse capacity at the tenant level and sells additional capacity once entitlements stop covering demand, so review current database, file, and log capacity in the Power Platform admin center and consult Microsoft’s guidance on adding Dataverse capacity when you need to buy more.  

2. Microsoft Foundry solutions, for advanced agent and AI functionality, deserve their own Azure budget for custom models, agents, tools, grounding, evaluation, observability, search, and hosting; because model, region, deployment, token, storage, and supporting-service choices all drive the result, model the specific architecture in the Azure pricing calculator rather than estimating from a flat rate, and check Foundry’s pricing page for the underlying unit costs.  

3. Microsoft Fabric and SQL make up the data foundation that supplies governed, current information to Copilot and custom AI solutions; Fabric costs can include capacity, OneLake storage, SQL storage, and Power BI licenses, while SQL costs depend on the product, deployment model, compute, storage, licensing benefit, and resiliency design you choose, so size these with the Fabric Capacity Estimator, Fabric pricing page, Azure SQL Database pricing, Azure SQL Managed Instance pricing, and SQL Server pricing pages. 

Regardless of the use case, include engineering, data quality, integration, testing, support, monitoring, and change management in the funded business case. 

5. Microsoft 365 E7: Is the Bundle Worth It?

Weigh the Bundle Against the Components You’d Buy Separately

An organization planning broad Copilot use, scaled agent adoption, and an integrated security posture should compare Microsoft 365 E7 against its current license stack. Microsoft lists E7 at $99 per user per month under an annual commitment on its Microsoft 365 E7 product page, and the bundle combines Microsoft 365 E5, Microsoft Copilot, Microsoft Entra Suite, and Microsoft Agent 365 in one SKU; Microsoft’s E3, E5, and E7 feature comparison lays out what each tier includes. 

Run that comparison against your organization’s actual contract rather than a sum of public list prices, and factor in existing E3 or E5 commitments, Copilot add-ons, Entra products, Agent 365, security overlap, deployment timing, and which users genuinely need the full bundle. You can watch eGroup’s E7 webinar to learn more. 

Choosing a Copilot Credit Purchasing Model

Copilot Credits are the currency for AI consumption transactions and Microsoft offers three paths for purchasing Copilot Credits.  

1. Pay-as-you-go bills actual monthly consumption in arrears through an Azure subscription, at a rate the October 2026 Copilot Credits Guide lists at $0.01 per credit, and it fits exploration, variable demand, and overage coverage.  

2. Prepaid monthly capacity packs, described on the Copilot Studio pricing page, bundle tenant-wide credits in packs of 25,000 that you pay for in advance each month; use current pricing and contract terms to size the pack count and decide whether PAYG should cover any overage.  

3. The prepaid annual option, the Copilot Credit Pre-Purchase Plan or P3, uses a one-year pay-upfront pool managed through Microsoft’s credit purchasing guidance, and unused credits expire at the end of the term; the October 2026 guide publishes tiered discounts of 5 to 20 percent against PAYG rates, based on committed credit volume. 

PAYG limits commitment while usage remains uncertain, monthly packs create a predictable recurring capacity pool, and annual P3 trades commitment risk for a published volume discount. We recommend starting with measured usage under spending policies and alerts, then stepping up to a larger commitment once the organization can connect consumption to funded business processes. 

AI budgeting works best as a portfolio decision: fund the right people, the right agents, the governance layer, and the data platform behind a defined process, then use measured consumption and outcomes to adjust the next allocation. And while this guide is comprehensive and provides you with the tools needed to establish a robust AI budget, the process is far from simple. Build your budget and then reach out and we would be happy to review your approach and funding estimates with you to ensure that you are set up for success for the next year. Let’s get the funding and take action! 

Make Every AI Dollar Count

Building an AI budget is only the first step. eGroup can help you evaluate your Microsoft licensing investments, identify cost-saving opportunities, and make informed decisions before committing to your next year of AI spending.