, , , , , , , ,

AI Economics: When Does the Meter Start Running?

rquickenden avatar

Ignite this month has seen huge updates across the technology stack with Copilot being front and centre. It was not just about features though – cost, reporting, governance and control as also a major topic.

These updates pushed the Copilot message well beyond chat and document assistance, with Copilot Home, Office in Copilot, Cowork, Code, Autopilot and Workspaces creating a much broader AI experience.

But amongst all the feature announcements, there is another change organisations need to understand.

The cost of Copilot is now a blend of subscription AI and usage-based AI.

That doesn’t mean Microsoft is suddenly charging for every prompt. The core M365 Copilot licence remains the foundation for everyday AI – Chat and Copilot in Office for things such as drafting, creating, modifying, summarising, research, and analysis etc.

But as new features, frontier models and tasks become more complex, longer running and increasingly agentic, usage-based billing (through Copilot Credits) start to become part of the equation and businesses need to understand and manage this.

Copilot is evolving from something organisations simply licence per user into a broader AI work platform where licences, credits, spending policies and consumption can sit alongside each other. It’s no longer also a Copilot vs ChatGPT type conversation as Copilot value goes way beyond the model – especially in the work context!

When we think of delegating work to AI like Copilot, a common question should be more about what outcome or value I get for my buck or quid!

What Does $50 an Hour of AI Buy You?

There was a brilliant example of this during the Ignite keynote. Judson Althoff demonstrated himself using Copilot to prepare for an upcoming keynote.

This wasn’t simply creating a PowerPoint Deck.

Judson used Copilot and Cowork to research the event, develop the story, create an outline and presentation, build a customer tracker, used Copilot Code to generate a coded 3D experience and set up his Autopilot to keep some of that work moving afterwards without needing to disturb his team over the weekend!

Judson shows us what we created and how – and what he managed to achieve in around two hours. Then came the number that caught my attention.

  • Around 10,000 Copilot Credits – approximately $100 (£80).
  • That’s roughly $50 (£40) an hour of AI consumption.

At first, like many you think – wow $100 is lot for doing “one thing”.

But then you look deeper at the outcome.

In two hours, Judson had researched an event, developed a keynote story, created supporting content and tools, and initiated an agent to continue some of the work afterwards – without pulling his team into meetings on a Sunday.

Suddenly that $100 doesn’t sound particularly expensive.

It potentially sounds like a bargain.

How Copilot Consumption Costs Scale

This is where organisations need to start paying attention though. Not because the example above was not “worth it”, but because we need visibility, control, governance and limits.

Imagine if 200 employees in your organisation each ran an equivalent $100 agentic session every month, that’s potentially $20,000 a month (almost $1/4m per annum) of additional consumption.

That doesn’t make the technology bad value.

But it does mean organisations need to understand when they’re using the AI included within their licences and when they’re moving into consumption-based services, where the value is an what that return is.

Not Everything in Copilot Costs Credits

During this same keynote other demos (of what Copilot does now) Microsoft showcased features such as Chat and Copilot in Office (plus Office in Copilot) are included within the monthly Copilot subscription cost, supporting activities such as drafting, summarising, planning, creating and co-authoring content, graphic, interactive charts and dashboards and analysis without any additional charges.

Microsoft’s Jacob Andreou said that: “Copilot is the only way to draft a PowerPoint presentation without spinning a meter.“

That’s an important distinction. There is the everyday Copilot experience organisations licence for their users that is extremely powerful for everyone’s everyday work.

Increasingly, alongside that, there is advanced agentic work where Copilot can take responsibility for bigger pieces of work, reason across multiple tasks and continue working towards an outcome. This clarified the choice, control and meaning to the Copilot pricing structure.

This is also why things like Cowork, Code and Autopilot matter so much.

  • Traditional Copilot is largely: “Help me do this”, “create this thing”, “research this”, “summarise and present that…”
  • Cowork moves towards: “Take this piece of work and get it done.“
  • Autopilot takes another step towards: “Keep moving this forward for me, follow up with people and when done, consolidate it to a report while I am away.“

That is a huge technological and cultural change – and it explains why a simple flat per-user licence won’t necessarily be the natural commercial model for every type of AI work.

AI FinOps: Measuring Cost Against Business Value

This is one of the most important conversations organisations are (and need to be ) having right now.

The question therefore is no longer “How many Copilot licences do we need?” or even “how do we ensure we are getting good usage from our license investment”.

It is quickly shifting to:

“Which AI capabilities should people have for their role and work? What should they be allowed to consume, what gates and approvals need to be in place? and “What business outcome should that consumption create”, and “How do we show (and prove) that value“

The aim shouldn’t be to minimise every Copilot Credit. The aim is to prove value in the flow of work or the task and rewrite the commercial business flow and value Agentic Workflows bring.

  • A complex research task that saves someone half a day could be fantastic value even if it consumes considerably more than a simple summary.
  • Equally, an agent repeatedly consuming resources without a useful outcome is waste – however impressive the technology might be.

Tracking this and proving it is not always simple. How much human / team time is needed to complete that task too. If that person is “watching” AI work rather than using that time elsewhere – is that still value? Do we count that cost!?

AI Governance, Spending Controls and Visibility

AI Governance is not just about locking it down and saying no.

Also, this week at Ignite, Microsoft demonstrated IT and Finance visibility into AI spend, including what has been consumed and what is driving that consumption.

New capabilities and granularity on reporting and spending limits was also demonstrated for Usage Based Billing services (like Cowork) such as spending limits for Cowork, with users needing to request an increase when their limit is reached.

Microsoft also showed how business outcome value measurements is also coming into Agent 365. This combination of cost visibility, limits and business outcomes becomes increasingly important as agentic AI scales.

Five Ways to Keep Costs Under Control

  1. Know what you are paying for. Whatever AI tools your organisation uses, make sure the business understands what is included and what is consumption based. Don’t assume every AI license and experience follows the same commercial model. How many different products are being paid for and used?
  2. Start small with agentic use cases. Measure actual consumption rather than trying to predict enterprise-wide costs from day one.
  3. Segment users and use cases. Not everyone needs access to every advanced capability or model.
  4. Put limits and visibility around consumption. Users, IT and finance should understand when credits are being consumed and why.
  5. Measure cost against business value. The useful metric isn’t simply “credits consumed”. It’s what those credits achieved.

And that last point is probably the most important.

From AI Licensing to AI Economics

As much as we were spoiled by the one “all you can eat AI” with Copilot and some other AI tools, usage-based billing is a logical direction for AI.

Businesses, Builders and Leaders want better, faster, more agentic models, need deeper reasoning, are building agents that can perform complex (autonomous) work and AI that can continue working after we’ve closed the laptop.

These capabilities consume very different amounts of compute, run longer (on their own) and run without a human trigger, so expecting everything to sit forever behind one flat per-user licence probably isn’t realistic.

But flexibility can very quickly become complexity. Microsoft must continue to invest in helping customers make the distinction between included and metered AI crystal clear, while giving administrators sensible controls and users visibility of what they’re consuming. It was good to see new features, controls and reporting announced this week as well as just new capabilities.

The biggest change here isn’t really billing – It’s that AI is fast becoming an operating environment for human and digital work.

  • Licences still matter.
  • Credits increasingly matter too.
  • Governance, observability and value measurement matter even more.
  • The role of partners is to help their customers with cost and governance, not just enablement, use cases and building things.

So perhaps the question we should be asking is less about “Is $100 of Copilot expensive?”, it is “Do I know when I’m spending that $100 – and was the work it just did for me worth it?”

If the answer to both is yes, $100 could be an absolute bargain.

Enjoying this article?

Subscribe to get new posts delivered straight to your inbox. No spam, unsubscribe anytime.

No spam. Unsubscribe anytime.

You may also like

See All Posts →

Leave a Comment