Most Australian businesses asking "do we need Copilot Studio?" already pay for Microsoft 365 Copilot and assume the second product is an upsell on the first. It is not. They are priced differently, entitle different things, and are consumed in different units. Getting the distinction right before you commit is worth real money, because Copilot Studio is sold tenant-wide in capacity packs and a reasonable agent design can run for a few hundred dollars a month — or several thousand — depending entirely on how much work it does, not how many agents you build.
Here is what each product is for, what the Australian pricing actually says, and how to size it from trigger volume rather than from an org chart of imagined bots.
The two products do different jobs
Microsoft 365 Copilot is a per-user productivity licence. It puts generative assistance inside Word, Excel, Outlook, Teams and the Microsoft 365 surfaces your staff already use, working against content that user is already permitted to see. The unit of value is the employee.
Copilot Studio is a build-and-run platform. You use it to create agents that answer questions, take actions, call connectors, and run on channels — including channels outside Microsoft 365, such as a website, Teams for external parties, or a voice or messaging surface. The unit of value is the work the agent performs.
That distinction matters because it determines the question you should be asking. For Microsoft 365 Copilot, the question is "how many people will use it and will they get an hour a week back?" For Copilot Studio, the question is "how many billable operations will this design generate per month, and what does that cost?"
What Copilot Studio costs in Australia
Microsoft's Australian pricing page lists Copilot Studio as a tenant-wide licence containing 25,000 Copilot Credits per pack per month at AU$299.30 per pack, observed on 16 October 2026. The equivalent US listing, updated 11 November 2026, shows the same structure at US$200 per 25,000-credit pack per month.
Two things follow from "tenant-wide".
First, you are not buying a seat and you are not buying an agent. One pack sits across the tenant and anything you build draws against it. Building a second agent does not, by itself, require a second licence.
Second, scale is purchased in packs. If your consumption exceeds 25,000 credits in a month, you add capacity rather than negotiate a different plan. That makes the arithmetic clean, but only if you can forecast consumption — which is where most buyers come unstuck.
Before publishing an internal business case on the AU$299.30 figure, confirm whether the price you are quoted is GST-inclusive and whether an existing Microsoft agreement, enterprise agreement, or reseller arrangement changes what you actually pay. List price and payable price are frequently not the same number in Australia.
Credits are consumed by work, not by agent count
A Copilot Credit is spent when the agent does something: holds a conversation turn, generates an answer, invokes an action, calls a connector, retrieves from a knowledge source. Different operations consume different amounts, and Microsoft's documented consumption rates change as the product changes. That is the single most important mental shift for a buyer who has priced SaaS by seat for a decade.
It also means the two questions that feel natural — "how many agents do we need?" and "how many staff will use it?" — are both the wrong starting point. The right starting point is trigger volume:
- How many inbound events per month hit this agent? Emails, form submissions, chat sessions, phone calls, records created.
- How many turns or steps does a typical event take to resolve?
- How many of those steps call an external system versus answering from indexed content?
- What fraction escalate to a human, and does the escalation path itself consume credits?
Only once you have those four numbers can you check them against Microsoft's current consumption rates and derive a monthly credit figure. The pack size and price are verifiable and stable. Per-conversation cost is design-specific, and anyone quoting you a universal "cost per chatbot conversation" has skipped the step that matters. Scoping that model honestly — mapping triggers, turns and connector calls before anyone opens the designer — is the bulk of the work when we design and build Copilot Studio agents for Australian clients, and it is also the step that decides whether the project lands on one pack or four.
Where the entitlement line falls
Microsoft's licensing documentation, dated 27 October 2026, is the authority here. It describes the standalone Copilot Studio subscription as allowing organisations to build agents on supported channels, connect to data using premium connectors, and operate beyond a narrow Microsoft 365-only scenario.
Read that carefully, because the practical line falls around four things:
| Factor | Points toward Microsoft 365 Copilot | Points toward Copilot Studio |
|---|---|---|
| Who uses it | Internal staff, in Office apps | Customers, suppliers, or staff via a purpose-built agent |
| Channel | Microsoft 365 surfaces | Web, external Teams, voice, messaging, embedded |
| Data access | What the signed-in user can already see | Premium connectors, line-of-business systems, external APIs |
| Pricing unit | Per user, per month | Tenant-wide credit packs |
The entitlement depends on your tenant's actual licences and your actual deployment scenario, not on a general rule. Validate your intended design against Microsoft's licensing matrix before you commit to a build — particularly if the agent will face anyone outside the tenant or reach a system that requires a premium connector. A design that assumed existing Microsoft 365 Copilot coverage and then discovered a separate requirement mid-build is an expensive way to learn the distinction.
Where Copilot Studio stops and execution pricing wins
Copilot Studio is strongest where the work is conversational, sits close to Microsoft 365 identity and content, and benefits from being inside the Microsoft governance perimeter your tenant already runs. It is weakest, commercially, where the work is high-volume, deterministic and unattended — a nightly reconciliation, a 20,000-record-a-month integration job, a webhook that fires every time a form is submitted and simply moves data between four systems.
That kind of work is better priced on executions. n8n's published Business plan is self-hosted only at €667 per month billed annually for 40,000 workflow executions, with monthly billing reported at €800. At recent exchange rates that is roughly AU$1,100 a month, and because it is denominated in euros, your AUD cost moves with the rate. Divided across the included volume it works out near two Australian cents per execution. Reported figures from 26 October 2026 put additional capacity in large blocks of €4,000 or US$4,800 per 300,000 executions, which is a similar unit economic at much higher volume.
The estimation trap on that side is the definition of an execution. n8n bills per workflow execution, not per step inside the workflow. A workflow triggered once per customer record consumes one execution per record. The same logic run on a schedule over a batch of 500 records consumes one. Two designs that look identical on a whiteboard can differ by two orders of magnitude on the invoice, which is why you model from trigger frequency and record volume rather than from the number of automations you intend to build.
A sensible architecture for an SME often uses both: Copilot Studio for the conversational front door where a person asks a question, and an execution-priced workflow engine behind it for the unattended bulk movement of data. Deciding which side of that line each piece of work sits on is the design decision that sets your running cost for the next three years.
Two things the pricing page does not tell you
Data location. The Australian pricing page establishes an Australian price. It does not establish that every workload's processing, prompt content, retrieved documents and logs remain in Australia. If you handle health records, client files under professional obligations, or anything where data residency is contractual rather than aspirational, that needs to be confirmed per workload and per connector — not assumed from the currency on the invoice.
Operational ownership. Credits and licences are the cheap part. The expensive part is identity and permissions, exception handling, what the agent does when a connector times out, who gets alerted when it silently stops answering correctly, and who owns the thing six months after launch when the underlying process changes. Budget for that or the agent quietly degrades into something staff route around.
Start with the count, not the catalogue
Before you buy anything, pick your single highest-volume candidate process and count its triggers for one month. Actual numbers from your inbox, ticket queue, form submissions or CRM records — not an estimate. Then count how many steps a human currently takes to resolve one, and how many of those steps touch a system outside Microsoft 365. Those three numbers tell you more about which licence you need than any feature comparison will.
If you would rather not guess at the credit maths, send us the process and the volumes. We will map it against Microsoft's current consumption rates and the licensing matrix, tell you how many packs it realistically needs, flag whether any part of it belongs on execution-priced automation instead, and give you a fixed scope with a two-to-four-week delivery window — or a plain answer that the numbers do not justify building it yet. Ask us for that scope and you will get a figure you can take to a board paper, not a brochure.