Beyond Microsoft 365 Copilot: Why Foundry Is the Next AI Move

Microsoft 365 Copilot has been one of the fastest enterprise software adoption stories of the last decade. Twenty million paid seats globally, 250% growth in seat additions year on year, and case studies showing serious returns. British Columbia Investment Corporation reported 10 to 20% productivity gains for 84% of users in their pilot. Forrester’s Total Economic Impact study puts the three-year ROI at 116% for a composite enterprise. The returns are real.

But Copilot is the start of the Microsoft AI story for most organisations, not the end of it.

The businesses getting the biggest returns from Microsoft AI today are the ones using Copilot as the foundation for something more ambitious. Custom agents, multi-step workflows, AI that runs core operations rather than just supporting them. And that work is happening in Microsoft Foundry.

Here’s why Foundry is where the next wave of value sits, and how to think about getting there.

Key Takeaways

• Copilot adoption has scaled fast. Twenty million paid seats globally, with strong ROI for organisations that have done the rollout work properly
• The opportunity has now moved one step further up the stack. From productivity assistance to custom AI agents built in Microsoft Foundry
• Foundry is where the work Copilot can’t do alone gets built. Custom agents, multi-step processes, regulated industry workflows, integration with line-of-business systems
• Most enterprises getting full value from Microsoft AI run both. Copilot for general productivity, Foundry for the work that’s specific to how their business runs
• Copilot adoption is the foundation for Foundry success. Organisations that move into Foundry well usually got their Copilot rollout right first
• Foundry needs a different operating model. Skills, governance and engagement are different to a Copilot rollout. This is what most organisations underestimate
• Australian and New Zealand businesses have local Foundry capacity. Microsoft’s A$25 billion data centre investment is expanding the local footprint significantly through 2029

 

Copilot Is Working. There’s Just More on the Table.

Most Microsoft customers we work with have rolled out Copilot. Many of them are getting real value from it. Meetings get summarised. First drafts of emails and documents land in minutes instead of hours. Data analysis that used to need a power user is now in reach of anyone in Excel.

And the rollouts that have gone well share a few characteristics. They picked specific use cases per role before issuing licences. They built champions networks rather than relying on one-off training. They had visible executive use setting the tone. They measured engagement, not just licences sold. They treated adoption as a programme, not a launch event.

If you’re still working on getting Copilot adoption to where you want it, those are the levers that move the needle. We help clients with this work all the time.

But here’s the bigger conversation we’re having with senior leadership teams across our client base. Even a Copilot rollout running at full tilt only addresses part of the AI opportunity in your business.

Industry data shows around 36% of employees with Copilot access use it weekly. The organisations that have nailed adoption do significantly better, but the gap between Copilot’s potential reach and where most enterprises sit today points to something worth thinking about. There’s a lot of value still on the table.

Some of that value is in continuing to push Copilot adoption. The rest of it sits one step further up the stack.

 

Copilot Is the On-Ramp. Foundry Is the Destination.

Think about what Copilot does well.

It summarises a meeting. It drafts an email. It explains a spreadsheet. It writes a first cut of a deck. It accelerates the general knowledge work that every office worker does. That’s enormously valuable, and it’s only going to keep getting better.

But it’s the easy 20% of the AI opportunity in your business.

The harder 80% is the work specific to how you actually operate. The processes that span Dynamics, your data lake, your line-of-business systems and three or four SaaS tools. The workflows where someone reads an unstructured document, checks it against a policy, makes a decision, updates four systems and notifies a customer. The decisions that need to happen consistently across thousands of cases, with a clear audit trail, in a regulated environment.

Copilot wasn’t built for that work. Foundry was.

What Foundry Is

is Microsoft’s platform for building, deploying and governing custom AI agents and applications. It sits at the layer beneath the Copilots you see in Word, Excel and Dynamics. It’s where the agents themselves get built.

A few things make it different to the Copilot you’re already running:

• Custom agents, built for your processes. Not the pre-built ones in Dynamics or Microsoft 365, but agents designed for your specific workflows, your data, your business rules
• Choice of model. You’re not locked into a single AI model. Foundry lets you pick the right model for the right job. Anthropic, OpenAI, Microsoft’s own MAI models, Meta, others. Mix and match per use case
• Integration with everything. Over 1,400 connectors into business systems including Dynamics 365, SharePoint, SAP, Salesforce and ServiceNow. Agents can read, write, and act across systems
• Enterprise governance baked in. Every agent gets a managed identity through Microsoft Entra. Centralised observability, guardrails, policy enforcement, audit trails. This matters in financial services and other regulated environments
• Distribution through the surfaces you already use. Agents built in Foundry can be published into Microsoft 365 Copilot, Teams, and custom apps. Your users don’t need a new tool. They access the agent through the interface they already know

In plain English, Foundry is where Microsoft AI moves from productivity assistant to operational backbone. The place where it actually changes how the business runs.

 

What Foundry Looks Like in Practice

Two examples to make this concrete. One financial services, one general.

Financial Services: A Hardship Triage Agent

A regulated lender receives several hundred hardship requests a week. They arrive through email, web forms and inbound calls. Each one needs to be read, categorised, prioritised, and routed to the right team within tight regulatory timeframes.

Today, that’s a manual job for the operations team. Read the request, decide if it looks like genuine hardship or a routine query, decide if there are vulnerability indicators, decide on urgency, log it in Dynamics, route it to a case manager. It’s slow, inconsistent across staff, and ASIC is paying close attention to whether hardship is being identified and handled well.

A custom hardship triage agent built in Foundry reads the incoming request, extracts the key information, checks for indicators of vulnerability against the lender’s own policy framework, scores urgency, and creates a properly categorised case in Dynamics with a clear audit trail of how it reached its conclusion. A human still makes the decision, but they start with a structured, evidenced summary instead of an unread inbox.

Copilot can’t do that. The pre-built Customer Service agents in Dynamics get part of the way, but the policy-specific logic, the local regulatory context, the vulnerability frameworks are unique to the lender. That’s Foundry work.

General: A Supplier Onboarding Agent

A mid-sized professional services firm onboards around 40 new suppliers a month. The process involves collecting documents, running ABN checks, verifying insurance certificates, screening against sanctions lists, setting up the supplier in the finance system, and getting sign-off from three different approvers.

Today it takes about three days per supplier, mostly spent waiting for emails to be answered.

A Foundry agent collects the documents, validates them, runs the external checks, populates the supplier record in Dynamics, drafts the approval requests, and chases approvers automatically. The same three days of work happens in a few hours. The finance team handles exceptions instead of working through every supplier from scratch.

This is well past Copilot territory. It needs orchestration across systems, business logic specific to the firm, and an audit trail. Foundry.

 

Why Copilot Adoption Sets You Up for Foundry

None of this is an argument against Copilot. It’s an argument for taking the next step.

The organisations that move into Foundry successfully are usually the ones that got Copilot adoption right first. They built AI literacy across their workforce. They got people comfortable with prompting and reviewing AI output. They built governance muscles. They know what good adoption metrics look like.

That foundation matters more than people expect when you start building custom agents. Foundry needs users who trust AI output enough to act on it, governance processes that can scope and approve new use cases, and an organisation that knows how to measure whether an AI capability is delivering value. Those things are built during a good Copilot rollout.

If Copilot is still finding its feet in your business, the five things that consistently move adoption are:

• Pick specific use cases before licences. Identify five to ten high-value tasks per role, document them, train against them, measure them. Generic rollouts get generic results
• Champions, not training videos. Peer modelling beats top-down training every time. Identify your power users early and make them visible. Their success stories scale adoption faster than any policy
• Executive modelling that’s visible. If your CEO doesn’t use Copilot in their meetings, your middle management won’t either. Leadership use sets the tone
• Measure what matters. Active weekly use, time saved on specific tasks, repeat use cases. Seat counts are not adoption
• Treat it as a programme, not a project. Continuous enablement, regular check-ins, targeted re-engagement of low adopters at the 30 and 90 day marks. Adoption isn’t a launch. It’s an ongoing discipline

Done well, this is what puts the Forrester ROI numbers within reach. And it sets the foundation for the bigger move into Foundry.

 

Where 365 Mechanix Comes In

This is where more of our work is heading. Three years ago, most of our engagements were Dynamics implementations. Two years ago, Copilot rollouts. Today, an increasing share of our work is in Foundry. Custom agents built for specific business processes, governance frameworks for AI in regulated industries, hybrid Copilot-plus-Foundry programmes that move organisations from productivity gains to operational change.

It’s where we think the next two to three years of value sits for businesses on the Microsoft stack, particularly in financial services. And it’s a different kind of engagement to a Copilot rollout. More design, more integration, more careful thinking about data, governance and operating model. The outcomes are also a different order of magnitude.

If you’re at the point of looking past Copilot and asking what’s next, that’s the conversation we’re having most often right now. Get in touch.

 

What This Means for You

If Your Copilot Rollout Is Still Building Momentum

Keep going. The fundamentals around use cases, champions, executive modelling and measurement are worth investing in. But start thinking about what comes after. The organisations getting the biggest returns from Microsoft AI aren’t the ones with the highest Copilot conversion rates. They’re the ones using Copilot adoption as the foundation for something bigger. Plan for that early.

If Copilot Is Working and You’re Wondering What’s Next

You’re in the best possible spot. Your users are AI-comfortable, your governance is established, and you have a real sense of where the productivity gains are landing. The next step is to look at the processes Copilot can’t touch. Identify two or three high-value workflows that are specific to your business, then build for them in Foundry. Start small, prove the model, scale.

If You Haven’t Started Yet

Don’t skip Copilot. The temptation to leapfrog straight to custom agents is real, but the organisations that try usually find they don’t have the AI literacy or governance maturity to make Foundry work. Run a focused Copilot pilot. Build the muscles. Then move.

 

Where This Is All Heading

The story we’ve been telling clients for a while now is that AI on Microsoft is shifting from assistance to action. From summarising meetings to running processes. Copilot is the most visible part of that shift, and it’s where most organisations have started. But it’s not where the story ends.

Foundry is where the work that changes how your business runs gets built. The gap between the businesses building there and the ones still working on getting Copilot off the ground is the one worth paying attention to over the next twelve months.

If you want to talk through where your organisation sits on that journey, or what a move into Foundry could look like for your business, get in touch with the 365 Mechanix team. We work with organisations across Australia and New Zealand to turn Microsoft’s AI capabilities into things that work in production.

 

FAQs

What’s the difference between Copilot and Foundry?

Copilot is the AI assistant you see inside Microsoft 365, Dynamics 365 and other Microsoft apps. It helps with general productivity tasks like summarising, drafting and analysing. Foundry is the platform underneath, where custom AI agents and applications are built. Think of Copilot as a product you use, and Foundry as a platform you build on.

Do we need to choose between Copilot and Foundry?

No. Most organisations getting full value from Microsoft AI run both. Copilot handles general productivity. Foundry handles the workflows specific to how your business operates. They work together rather than competing.

Is Foundry only for big enterprises?

Not at all. The platform is built to scale up and down. Smaller businesses with one or two well-defined, high-value processes can get significant value from a small number of Foundry agents. The complexity scales with what you build, not the size of your business.

Do we need our own developers to use Foundry?

It depends on what you’re building. Foundry supports low-code and code-first development. Simple agents can be built and configured through the Foundry portal. More complex agents with custom logic and deep integrations usually need developers, either internal or through a partner. We do a lot of this work for clients who don’t want to staff it internally.

What about data governance and regulatory compliance?

This is where Foundry’s enterprise positioning matters. Every agent gets a managed identity through Microsoft Entra. There’s centralised observability, audit trails, guardrail policies, and integration with Microsoft Purview for data classification. For regulated industries, this is meaningful, though governance is always something the regulated entity owns, not something the platform delivers on its own.

How long does it take to build a Foundry agent?

It varies. A simple internal-use agent can be built and deployed in a few weeks. A production-grade agent that integrates with multiple systems, handles regulated data, and has full governance wrapped around it is more like a few months. Most of the time is spent on design, integration and testing, not the AI bit.

Is Foundry available in Australia?

Yes. Microsoft Foundry runs on Azure, which has multiple Australian regions. With Microsoft’s recent A$25 billion local infrastructure commitment, capacity is expanding significantly through 2029. For regulated industries that need data residency on Australian soil, the picture is getting stronger.

How does 365 Mechanix help?

We work with organisations across ANZ to design, build and operate AI solutions on the Microsoft stack. Copilot adoption, custom agents in Foundry, Dynamics 365, Power Platform, the full picture. If you’re thinking about where to go next on your AI journey, we’d be happy to have a conversation.