CG TECH

Insights

Copilot Pay-As-You-Go: Who Owns Your AI Spend?

Three business leaders reviewing AI costs and usage charts on a laptop in a Sydney office overlooking the Harbour Bridge and Opera House.

If your business treats Microsoft Copilot as a fixed monthly cost per person, there’s a change coming on 2 November 2026 that deserves attention.

From that date, new Microsoft 365 Copilot Business licences purchased through Microsoft’s Cloud Solution Provider, or CSP, channel will include usage-based billing by default. This applies to new standalone licences and bundles, with the required Azure subscription setup included.

That doesn’t mean Microsoft is suddenly charging by the prompt every time someone uses Copilot in Word, Excel, Outlook or Teams. The change applies to eligible services where the amount of work can vary, including Copilot Cowork and the Work IQ APIs.

The change makes sense, but it also creates a new question for business leaders.

Who owns your AI spend once the cost can move each month?

If nobody owns that question, the first serious discussion about Copilot consumption may happen after the invoice arrives.


What’s changing on 2 November?

Microsoft 365 Copilot Business is still a per-user subscription. It continues to include Copilot Chat, Copilot in Microsoft 365 apps, Work IQ grounding and Microsoft agents such as Researcher, Analyst and Facilitator.

Usage-based billing sits beside that subscription.

It covers eligible services where Copilot may carry out longer or more complex work. Instead of charging the same amount for every task, Microsoft measures that work in Copilot Credits.

One credit currently costs US$0.01 under Microsoft’s published pay-as-you-go rate. Pricing is in US dollars and can change, while Australian charges may also depend on the reseller, currency treatment, taxes and the customer’s agreement.

The key point is simple: this isn’t a replacement for the Copilot licence.

It’s another billing method for services that sit outside the included subscription allowance.

Copilot Cowork is a good example, it requires a Microsoft 365 Copilot licence, but the subscription doesn’t include a pool of Cowork credits.

Cowork usage is charged according to the work it performs.

Businesses that are still getting familiar with the service can read more about how Copilot Cowork works and where it may fit.


Why Microsoft is moving this way

It’s easy to look at usage-based billing and assume it’s simply a way to increase revenue. I don’t think that tells the full story.

Fixed per-user pricing works well when people use a product in roughly similar ways. Email, file storage and everyday Copilot tasks fit that model because the cost of serving one user doesn’t usually swing too far from another.

Agent-style work is different.

An agent that prepares a short weekly update doesn’t perform the same amount of work as one that reviews six months of data, checks several information sources and prepares an executive report.

Charging the same flat amount for both would hide the real cost of the work. Usage-based billing makes that cost visible, which can be helpful if businesses use the information properly.

It gives leaders a better way to ask:

  • What did this AI task cost?
  • How much staff time did it save?
  • How often are people using it?
  • Which agents are producing useful outcomes?
  • Which tasks are consuming credits without giving enough back?

Those are better questions than asking whether Copilot is generally “worth it”.

Copilot doesn’t produce one single return across a business, so its value needs to be measured use case by use case.


What could a Copilot task cost?

Microsoft’s Copilot Credits Guide provides planning ranges for light, medium and heavy Copilot Cowork tasks.

A light task, such as preparing a recurring weekly status update from a person’s priorities and calendar, may use around 70 to 200 credits. At the published pay-as-you-go rate, that’s roughly US$0.70 to US$2.

A medium task, such as collecting emails, calendar items, CRM data and file content for a customer briefing, may use around 400 to 600 credits. That works out to about US$4 to US$6.

A heavy task, such as reviewing six months of exported product usage data and producing a leadership report, may use more than 1,500 credits. That puts the starting cost above US$15.

These are rough planning estimates, not fixed prices.

Actual use depends on the models selected, the context retrieved, the tools called and the time needed to complete the work.

None of these amounts looks alarming on its own.

The bigger issue is frequency.

A US$5 task performed once a week is different from the same task being performed every working day by 40 people. A useful test can become a meaningful monthly cost once it spreads across teams.

That’s why businesses should estimate both the cost per task and the likely volume.


The setting I’d check first

When a business creates a Copilot spending policy, Microsoft includes a setting called Auto-apply new services.

It’s turned on by default.

When enabled, future supported Copilot services and agents are automatically added to that policy as Microsoft makes them available. This saves administration time, but it may also mean new forms of usage are approved without a separate review.

For some businesses, that’s fine.

They may prefer faster access and trust their existing limits to control the risk.

Others will want finance, IT or a business owner to review each new service before staff can use it. In that case, switching off Auto-apply new services gives the business a chance to assess the purpose, likely cost and data access first.

Neither choice is automatically right. What matters is that someone has reviewed the setting and made a decision.

That’s much easier to defend than leaving it on because nobody knew it was there.


Put controls in place early

Microsoft has added cost controls to the Microsoft 365 admin centre. They’re useful, but they still need to be configured around the way your business works.

Start with a spending policy

A spending policy determines which users and groups can access supported services, which services they can use, how much the policy can consume and which billing method pays for it.

The default policy can cover the whole business, or it can be adjusted to apply to selected security groups.

I’d avoid starting with unlimited access across every user. Begin with a group whose work suits agent-based automation, then watch what happens before expanding further.

A good pilot group has:

  • Clear and repeatable tasks
  • Enough work volume to test the value
  • Named business owners
  • Approved information sources
  • A way to measure time, cost and output quality

That produces evidence the business can use, rather than a collection of opinions about whether Copilot feels helpful.


Set policy and user limits

Administrators can apply a monthly limit to a spending policy, then add optional per-user limits underneath it.

The policy limit controls the total credits available to that group. The user limit reduces the chance that one person will consume most of the budget.

When users reach their limit, they lose access to the covered services for the rest of the month unless the limit is changed or the monthly period resets.

Microsoft also lets administrators configure threshold alerts. These can notify nominated people when policy usage reaches a set number of credits or a chosen percentage of the limit.

User notifications can be enabled too, giving people a warning before they run out.

I’d set alerts early enough to allow a useful response. An alert at 70% gives the business time to understand what’s driving consumption, while one at 100% simply confirms that the limit has already been reached.


Give people a request path

Limits shouldn’t become a dead end for staff doing valuable work.

Microsoft includes a credit request process that lets users ask for first-time access or a higher limit. Those requests can be managed in the Microsoft 365 admin centre or directed to another approval process, such as an internal service portal.

That creates a healthier model than choosing between unrestricted access and a support ticket with no clear owner.

A request should explain what the person is trying to achieve, how often they expect to run the task and what business result it supports. If the request makes sense, increase the limit. If it doesn’t, help the person find a better way to complete the work.


Measure value by use case

The Microsoft 365 admin centre can report Copilot Credit consumption by spending policy, user, group, agent and service.

That tells you where credits are going, but it doesn’t automatically tell you whether the spending was worthwhile.

The business still needs to connect each use case with an outcome.

If a customer briefing costs US$5 and saves a salesperson 40 minutes, it may be an easy decision to keep using it. If a recurring report costs US$20 and nobody reads it, the issue isn’t the price of the credits. The issue is that the work shouldn’t be running.

This is why I’d give every approved use case a simple record:

  • The task being performed
  • The business owner
  • The people or agent using credits
  • The expected monthly volume
  • The estimated credit cost
  • The time or effort saved
  • The output that needs to improve
  • The date for reviewing the result

This also helps avoid the problems that appear when people use unapproved AI tools without clear ownership. A controlled path to useful automation is one of the better ways to reduce shadow AI and wasted AI spend.


Your information still matters

Copilot Credit consumption varies according to the models used, the amount of context retrieved, the runtime needed and the tools called.

That makes information quality worth watching during a pilot.

If Copilot has several possible sources for the same answer, staff may need to check more material and rerun tasks. Clear ownership, sensible permissions and reliable source material make it easier to get a dependable result.

The cost effect will differ between tasks, so it needs to be tested rather than assumed. Still, businesses shouldn’t treat data readiness as separate from cost management.

A useful knowledge layer can improve the quality of AI output while reducing unnecessary work. That’s why the knowledge behind your AI can matter more than the model itself.

Metering adds another reason to get that foundation right.


What I’d do before November

I don’t see 2 November as a reason to avoid Copilot Business. I see it as a date to prepare for.

If your business expects to purchase new Microsoft 365 Copilot Business licences through a CSP partner, ask whether the order will fall under the new default usage-based setup.

Then take five steps:

  1. Nominate an owner for Copilot consumption and reporting
  2. Decide which users or groups should receive initial access
  3. Set policy and per-user monthly limits
  4. Configure alerts and the credit request process
  5. Review the Auto-apply new services setting

I’d also keep ownership balanced.

The person responsible for promoting Copilot adoption shouldn’t be the only person responsible for approving its spend. Good decisions usually come from bringing business, finance and technology owners together.

The goal isn’t to stop people using AI. It’s to know what the business is paying for and what it’s receiving in return.

Metering can help with that because it puts a visible number beside the work. Once that number is available, businesses can compare cost with time saved, quality improved or revenue supported.

If you’d like a second set of eyes on your Copilot licensing, spending policies, pilot use cases or information readiness, that’s exactly the kind of work we do at CG TECH.

**Recommended alt text:**
Business leaders reviewing Microsoft Copilot usage, licensing and AI spend controls in a Sydney office overlooking the Harbour Bridge.

About the Author

Carlos Garcia is the Founder and Managing Director of CG TECH, where he leads enterprise digital transformation projects across Australia.

With deep experience in business process automation, Microsoft 365, and AI-powered workplace solutions, Carlos has helped businesses in government, healthcare, and enterprise sectors streamline workflows and improve efficiency.

He holds Microsoft certifications in Power Platform and Azure and regularly shares practical guidance on Copilot readiness, data strategy, and AI adoption.

Connect with Carlos Garcia, Founder and Managing Director of CG TECH, on LinkedIn.

Sources

More insights

Want help applying any of this?

Talk to the CG TECH team about what it means for your business.

Contact us
Scroll to Top