Most Australian businesses we work with have already made peace with one Copilot decision: free Copilot Chat for casual web-grounded questions, versus the paid Microsoft 365 Copilot licence that reaches into your Word, Excel, Outlook and Microsoft Graph data. That decision tree is well-trodden. The newer, less understood question is what sits between them: metered, pay-as-you-go agents.
Microsoft now lets you run agents on top of Copilot Chat and pay per message through an Azure subscription, rather than buying a flat per-user seat. That changes the commercial shape of a rollout. Instead of one predictable line item per head, you get a consumption meter that can be a few dollars a month for a light user — or more than a full licence for a heavy one. This piece works the AUD break-even so you can see where the line actually falls.
Three Copilot cost models, not two
It helps to name the three things clearly, because Microsoft's marketing tends to blur them. First, Copilot Chat: the web-grounded chat experience that ships with most paid Microsoft 365 commercial plans at no extra per-user cost. It is genuinely useful for drafting, summarising pasted text and general questions, but on its own it does not see your tenant data.
Second, the flat Microsoft 365 Copilot licence — indicatively around AU$45 per user per month (indicative AUD list — confirm at purchase). That is the seat that grounds Copilot in your actual files, mailboxes and Teams chats, and embeds it inside the Office apps. Third, and the focus here, is pay-as-you-go agents: you switch on metered billing so people can run agents through Copilot Chat (or SharePoint agents) without holding a paid seat, and you pay per message consumed.
What a metered 'message' actually costs
Agent consumption is billed through the Copilot Studio meter into your Azure subscription. Indicatively, a standard message is around USD 0.01 and a generative (AI-grounded) message around USD 0.02; there is also a pre-purchase pack at around USD 200 for 25,000 messages per month (indicative — confirm at purchase). Converted and rounded for planning, think of it as very roughly 1.5c to 3c per message ex GST in AUD, but always price it in your own Azure agreement at purchase.
The AUD break-even, worked
Hold the seat at an indicative AU$45 per user per month and treat a blended agent message at roughly 1.5c (a mix of standard and generative, in AUD ex GST, indicative). The flat licence is cheaper once a person consumes more than about 3,000 agent messages a month. Below that, metered is cheaper; above it, the seat wins. At a richer blend nearer 3c a message, the crossover drops to around 1,500 messages a month.
- Light user — a few agent interactions a day, maybe 100–300 messages a month: metered is dramatically cheaper, often a couple of dollars versus AU$45.
- Moderate user — 1,000–1,500 messages a month: still usually cheaper on metered, but the gap narrows and the bill becomes variable.
- Heavy daily user — 3,000+ messages a month: the flat seat is cheaper and, just as importantly, predictable.
- Anyone who needs Copilot inside Word, Excel and their own email: the metered path does not deliver that work-grounding, so the seat is the only option regardless of volume.
The headline figures are indicative and depend on your exact blend of standard versus generative messages, your Azure pricing, and the 1 July 2026 list changes. But the shape holds: metered billing rewards occasional, bursty use and punishes heavy, every-day use. That is the opposite of how a flat seat behaves.
Why the break-even is the wrong place to stop
A pure cost crossover misses what each model unlocks. The flat seat is not just 'unlimited messages' — it is Copilot reading your tenant: summarising the right email thread, drafting from the actual document, answering from your SharePoint. Metered agents on Copilot Chat are powerful for scoped tasks you build deliberately, but they are not a cheaper way to get the in-app, Graph-grounded experience. If someone needs that, volume is irrelevant; they need the seat.
Equally, the metered model's variability is a feature for the right population. A frontline or seasonal workforce that touches an agent a handful of times a week should not be carrying a flat seat 'just in case'. Paying a few cents when they actually use it is both cheaper and more honest than provisioning fifty idle licences.
What we'd actually do
We start by segmenting people, not by buying licences. A small group of power users — the ones already living in Copilot inside Office — go onto flat seats from day one; arguing about their message count is a waste of time. Everyone else starts on Copilot Chat with pay-as-you-go switched on, and we watch the Azure meter for three to four weeks to get real consumption per head rather than a vendor estimate.
Then we promote. Anyone whose metered spend is trending past the seat price — comfortably under that ~3,000-message line, allowing headroom — gets moved to a flat licence. Everyone below stays metered. It is a far cheaper way to land Copilot across a business than buying a seat for every employee on a hunch, and it produces a usage baseline you can defend to a finance team.
The right answer is almost never 'all seats' or 'all metered'. It is a deliberate split, set by how each cohort actually uses Copilot and re-checked when prices move. Get the segmentation right and the licensing maths mostly looks after itself.