Build finance and operations agents on your own Microsoft stack

Copilot Studio is Microsoft’s low-code platform for building custom AI agents. It sits on the Power Platform and Dataverse, and agents you build run inside your own Microsoft 365 tenant rather than on a third-party AI vendor’s servers. You can connect agents to your Microsoft data (SharePoint, Dataverse, Dynamics 365) and to hundreds of connectors, then publish them into Microsoft 365 Copilot, Teams or other channels. As of 2026, capabilities like computer-using agents, agent-to-agent communication and apps inside agents are generally available, and Microsoft Agent 365 provides a central control plane for governing them.

It is a strong fit when you already run on Microsoft 365 and Dynamics 365. Because agents live in your tenant and inherit your existing Microsoft data processing terms, finance and operations teams can automate reconciliation prep, variance commentary, approval routing, AP triage and reporting without introducing a new vendor or a new data processing agreement. The trade-off is that Copilot Studio is best for workflows anchored in Microsoft data and tools. For heavy free-form reasoning, some teams still pair it with other models, which is where the data residency questions below become important.

It can, but only if you configure it correctly and stay aware of two 2026 changes. Copilot Studio is in scope for the Microsoft EU Data Boundary when your tenant has an EU or EFTA billing address and all your environments are created in an EU region. However, Microsoft 365 Copilot “Flex Routing” was enabled by default on 17 April 2026 (message centre item MC1269223), which permits Copilot language model processing to occur outside the EU Data Boundary during peak demand. Data at rest stays encrypted within the EU, but admins who require strict residency must actively disable Flex Routing. We help clients check and document this.

Microsoft now lets you select Anthropic Claude models in Office apps and in Copilot Studio, and Claude is available in Microsoft Foundry. This adds capability, but it changes your residency position. As of mid-2026, Anthropic models are out of scope for the EU Data Boundary, run on Anthropic-hosted infrastructure rather than Azure EU regions, and EU support is listed as coming during 2026. For EU and UK tenants, Anthropic models are excluded from the default rollout and an admin must explicitly opt in. The capability is real and useful. The point is to opt in deliberately, with your data protection officer informed, not by accident.

Yes, and we do this vendor-neutrally. The honest answer is that there is no single right tool. Native Copilot Studio agents on Azure OpenAI keep you inside the EU Data Boundary and your existing Microsoft agreement, which suits regulated finance workloads. Opting into Claude buys stronger reasoning for specific tasks but moves that processing outside the EU Data Boundary until Microsoft’s EU support lands. We map your workloads, your regulatory exposure and your appetite, then recommend a configuration. Often it is a mix: native agents for data-bound finance work, a carefully scoped model choice for the rest.

Cowork is Microsoft’s autonomous agent for long, multi-step jobs. You delegate a whole task and it works across your apps, files and the web before returning the finished output, rather than a draft. It became generally available on 16 June 2026. It is strong for research briefings, turning meeting transcripts into action lists with owners and dates, and first-draft reports from your own templates. Settle two things before switching it on. First, residency: Cowork’s advanced work runs on Anthropic’s Claude models, which are outside the EU Data Boundary and off by default for EU and UK tenants, so enabling them moves that processing out of the boundary. Second, cost: Cowork is metered and billed per task, so you need pay-as-you-go or a prepaid commitment on top of your Copilot licence. It is off by default with spend controls at tenant, group and user level. We help finance teams point it at lower-risk work first and govern the most sensitive jobs carefully.

We start with a short audit of your finance and operations workflows and your current Microsoft licensing and tenant configuration. We identify two or three agents that will pay back quickly, build them in Copilot Studio against your Dataverse and Microsoft 365 data, set governance and residency settings appropriately, and train your team to maintain and extend them. Most first engagements deliver working agents within a few weeks. We deliberately leave capability in-house rather than creating a dependency.