In This Article
On June 16, 2026, Microsoft moved its most capable agentic Copilot work, Copilot Cowork, to general availability, and it announced Microsoft Scout, an always-on personal agent, alongside it. Both run on a new billing model. Instead of a flat per-user license, the heavy agent work meters against a usage currency called Copilot Credits. For a bank, a credit union, or a mortgage lender, that one change reframes the whole conversation.
The question was never whether agents are useful. An agent that drafts a loan-file summary, reconciles three systems, and prepares a board packet overnight is obviously useful. The question a 150-seat institution now has to answer is a cost-control question and an access-control question at the same time: who is allowed to spend Copilot Credits, how much, on which workflows, and how do you prove all of it to an examiner six months later.
If you already piloted Cowork through Microsoft's Frontier program, there is also a near-term clock. Frontier customers currently using Cowork have to set up usage-based billing in the Microsoft 365 admin center by June 30, 2026. This guide is the cost-governance companion to our walkthrough of what Copilot Cowork does for financial institutions. That piece covers the capability. This one covers the meter.
What Cowork and Scout Actually Get Done
Start with the work, because the work is the reason any of this is worth governing. Copilot Cowork is an agentic system built for complex, long-running, multi-tool tasks. It pulls organizational context across Microsoft 365 and connected systems and runs a chain of steps toward an outcome, rather than answering a single prompt and stopping. Think of the difference between asking a chatbot a question and handing a capable analyst a project.
For a mortgage lender, that looks like an agent assembling a post-closing package: reading the loan file, checking the servicing system, flagging missing documents, and drafting the exception memo. For a credit union, it can mean an agent that reconciles month-end figures across the core and a reporting tool and produces the first draft of the board report. The value is the hours it gives back to people who are currently copying and pasting between four windows.
Microsoft Scout is the always-on version of the same idea, currently limited to Frontier customers. Scout is a personal agent that spans cloud, desktop, and web. It connects Teams, Outlook, OneDrive, and SharePoint along with your chats, emails, calendar, and contacts, and it can act on browser activity, local resources, and connected systems. An always-on agent that reads a loan officer's email and a CFO's SharePoint is powerful, and that is exactly why the security and governance layer underneath it is not optional for a regulated institution.
Why This Matters for Financial Institutions
A flat per-user license is a known number. You multiply seats by a price and you budget for the year. Usage-based agent work is a variable number that moves with how hard your team leans on it. For an institution that answers to the FFIEC, the NCUA, or the OCC, a variable cost that scales with autonomous AI activity is a line item your CFO and your examiner will both ask about. Governing it well is the difference between agents that pay for themselves and a surprise bill nobody approved.
Copilot Credits: A Usage Currency, Not a Flat License
Here is the model, stated plainly. Copilot Credits are a common currency for eligible Microsoft services that bill by usage, starting with Cowork and the Work IQ API. You buy and allocate credits, and agent work draws them down as it runs. To use Cowork at all, you need two things together: a Microsoft 365 Copilot license for the user, and usage-based billing with Copilot Credits set up for the tenant. The license is the seat. The credits are the fuel for the heavy, long-running agent jobs.
Credits come in two purchasing shapes. You can prepay, which buys a known pool of credits up front and gives finance a fixed number to plan around. Or you can run pay-as-you-go, which bills for what you consume. Most institutions will want a deliberate mix: a prepaid base for predictable workloads, with pay-as-you-go as the controlled overflow valve rather than the default. Microsoft does not publish a flat per-credit retail figure in an open source, so do not budget off a number you saw in a forum. Model your expected spend with Microsoft's official Customer Cowork Estimator at aka.ms/CustomerCoworkEstimator before you commit.
The contrast with the licensing your team already understands is the whole point.
Predictable Microsoft 365 Copilot Licensing
A fixed per-user, per-month price you multiply by seats. Finance can forecast it a year out. It covers the everyday Copilot experience in Word, Excel, Outlook, and Teams. The number does not move when one analyst has a busy week.
Usage-Based Copilot Credits
A variable pool that agent work draws down as it runs. Cost scales with how much autonomous, multi-step work your team triggers. Powerful, and worth it, but it needs budgets, caps, and clear ownership or it turns into a bill nobody forecast.
The credits side is not a flaw. Metered usage is the right model for work that genuinely varies. It just demands that you treat agent spend like any other variable operating cost: with a budget, an owner, and a ceiling. That control lives in one place.
Who Can Spend Credits, and on What
Before you worry about the total, decide who is allowed to spend at all. This is access governance, and it is where Microsoft 365 Copilot Credits are managed through the Cost Management dashboard in the Microsoft 365 admin center. From there you allocate credits, apply policy-based access and limits, and decide which groups and which workloads can draw from the pool. Microsoft documents the dashboard, its Overview and Consumption tabs, and its budget and alert controls in its usage-based billing overview for Copilot Credits.
The instinct at a small institution is to turn it on for everyone and see what happens. That instinct is expensive, and for a regulated lender it is also a control gap. An agent's spend is tied to what the agent is permitted to read and do, so an ungoverned rollout is two problems wearing one coat: an open budget and an open data boundary.
Credits are enabled tenant-wide with no spending policy. Anyone can trigger long-running agents against any data they can already see. Spend is invisible until the invoice arrives, and when an examiner asks who authorized an autonomous action on member data, nobody has the answer.
Credits are allocated to named groups under a spending policy. Microsoft Entra ID controls who can invoke agents, Microsoft Purview classifies and protects what those agents can touch, and data loss prevention rules keep regulated data inside the boundary. Spend and access are both attributable, by design.
That governed column is not theoretical work you build from scratch. It is the same identity, data-classification, and protection stack you already run, pointed at a new kind of consumer. Microsoft Entra ID decides who can launch an agent. Microsoft Purview classifies the SharePoint and email content an always-on agent would otherwise read freely, and our deeper look at Microsoft Purview for AI agents walks through exactly how those labels and policies extend to agent activity. Pair that with data loss prevention for financial services so an agent that drafts a loan summary cannot quietly route regulated data somewhere it should not go.
Budgets, Hard Caps, and the Audit Trail
Now the governance layer, the part that makes a variable cost examiner-safe. The Cost Management dashboard does more than show you a number after the fact. It gives you reporting and insights, budgets, alerts, and hard caps that stop spending before it runs past a ceiling you set. A budget tells you where you are. An alert warns you that you are getting close. A hard cap is the brake that prevents an overspend from ever happening, rather than explaining one in hindsight.
For a financial institution, hard caps are the headline control. A budget with no cap is a speed limit with no enforcement. Set a ceiling per group or per workload, wire alerts at sensible thresholds below it, and you have converted an open-ended usage cost into a bounded one your CFO signed off on. That same dashboard, with its Consumption tab and attributable usage, is the evidence trail an examiner expects when they ask how autonomous AI activity is controlled and accounted for.
As a Tier-1 Microsoft Cloud Solution Provider, ABT manages the Microsoft 365 tenant for more than 750 banks, credit unions, and mortgage companies. The pattern we see work is to start agent usage as a tightly capped prepaid pool against one or two high-value workflows, instrument the Consumption tab, then expand caps deliberately as the return proves out. Banks that meter and cap on day one treat agents as a controlled operating cost. Banks that switch everything on first end up explaining a bill to a board.
Governing the credits is half the job. Governing what the agents can read is the other half, and it is where ABT wraps Cowork and Scout in the M365 Guardian operating model. Guardian ties together Microsoft Entra ID, Microsoft Purview, Microsoft Defender, and data loss prevention so an always-on agent reading email and SharePoint stays inside the data boundary your examiner expects. Our overview of agent governance for financial institutions covers how that operating model applies once agents are in the tenant at scale.
The Cost-Governance Setup, Before June 30
Pull it together into a sequence. If you are a Frontier customer already running Cowork, the first item has a June 30, 2026 deadline. If you are planning a fresh rollout, the same steps apply on your own timeline, just without the clock.
Set up usage-based billing
In the Microsoft 365 admin center, configure usage-based billing with Copilot Credits. Frontier customers already using Cowork must complete this by June 30, 2026.
Model spend before you commit
Run the official Customer Cowork Estimator (aka.ms/CustomerCoworkEstimator) against your real workflows. Decide your prepaid base versus pay-as-you-go overflow from a modeled number, not a guess.
Allocate credits to named groups
Use the Cost Management dashboard to apply policy-based access and limits. Only the groups and workflows you approved should be able to draw from the pool.
Set hard caps and alerts
Define a ceiling per group or workload and wire budget alerts below it. The cap is the brake that prevents an overspend, not a report that explains one later.
Bound agent access with identity and data controls
Gate who can invoke agents with Microsoft Entra ID, classify and protect what they read with Microsoft Purview, and keep regulated data inside the line with data loss prevention.
Keep the audit trail examiner-ready
Make the Consumption tab and attributable usage part of your standing evidence so you can answer who authorized agent activity, on what data, for how much.
For mortgage lenders that want the productivity without owning every governance decision themselves, there is a complementary path. ABT's MortgageGuide Copilot is a financial-services-tuned agent built and governed by ABT, which gives lenders predictable, controlled agent value alongside the metered Microsoft surface rather than instead of it. Predictable Microsoft 365 Copilot licensing plus managed Copilot Credit governance is the combination that keeps the cost honest and the examiner calm.
The Takeaway
Copilot Cowork and Microsoft Scout are real productivity, and Copilot Credits make their cost real too. Lead with the work, govern with access, and close with control: decide who can spend, point your existing Microsoft Entra ID and Microsoft Purview stack at agent activity, then set budgets, alerts, and hard caps in the Cost Management dashboard. Frontier customers already on Cowork set up usage-based billing by June 30, 2026. Everyone else gets to do it on their own schedule, with the same discipline.
Turn On Copilot Agents With the Meter Under Control
ABT manages the Microsoft 365 tenant for more than 750 financial institutions. We lock in predictable Microsoft 365 Copilot licensing, model and govern your Copilot Credit spend with budgets and hard caps, and wrap agent access in the M365 Guardian operating model so an always-on agent stays examiner-safe. Let ABT set up your cost and access governance before you scale agents.
Frequently Asked Questions
Copilot Credits are a common usage currency for eligible Microsoft services that bill by consumption, starting with Copilot Cowork and the Work IQ API. Agent work draws credits down as it runs. You buy and allocate credits, and manage them through the Cost Management dashboard in the Microsoft 365 admin center, which supports both prepaid and pay-as-you-go models.
You need both. Copilot Cowork requires a Microsoft 365 Copilot license for the user and usage-based billing with Copilot Credits set up for the tenant. The license is the per-user seat for the everyday Copilot experience. The credits are the metered fuel for the heavy, long-running, multi-step agent work that Cowork performs.
The June 30, 2026 deadline applies specifically to Frontier program customers who are already using Copilot Cowork. Those customers must set up usage-based billing in the Microsoft 365 admin center by that date. It is not a universal cutoff for every organization. If you have not been running Cowork through the Frontier program, the deadline does not bind you, though the same setup steps apply whenever you adopt it.
Use the Cost Management dashboard in the Microsoft 365 admin center. It provides reporting, budgets, alerts, and hard caps. Set a budget so you can see where spend sits, wire alerts at thresholds below your ceiling, and set hard caps that stop spending before it passes the limit you defined. For a regulated institution, the hard cap is the key control because it prevents an overspend rather than explaining one after the invoice arrives.
Because credits are consumed by usage, your cost depends on how much agent work your team runs rather than a single flat figure. The right way to plan is to model your expected consumption with Microsoft's official Customer Cowork Estimator at aka.ms/CustomerCoworkEstimator, then decide your prepaid base and pay-as-you-go overflow from that modeled number. ABT can run that modeling with you and structure the allocation so spend stays predictable.
ABT is a Tier-1 Microsoft Cloud Solution Provider that manages the Microsoft 365 tenant for more than 750 financial institutions. We lock in predictable Microsoft 365 Copilot licensing, model and govern Copilot Credit consumption with budgets, allocations, and hard caps, and wrap agent usage in the M365 Guardian operating model using Microsoft Entra ID, Microsoft Purview, and data loss prevention so an always-on agent reading email and SharePoint stays inside your data boundary and examiner-ready.
