AI agents in Copilot Studio, connected to the systems you already have
Microsoft Copilot Studio lets you stand up an agent quickly, inside Microsoft 365 and Teams. The hard part is everything around it: the connection to SAP, Priority ERP or Dynamics 365, permissions, testing and ongoing cost.
Zooz-tech develops Copilot Studio agents that do real work inside the process, not just answer questions, and supports them after go-live.
Four types of agents we build in Copilot Studio
Agents for internal processes
An agent in Teams and Microsoft 365 that answers questions from your organization's documents and data, and moves the process itself forward: it opens a request, updates a record, sends an item for approval.
Agents that write to core systems
Read and write access to SAP, Priority and Dynamics 365 through an MCP layer or API, with user approval before every write.
Document and email automation
A document that arrives by email becomes records in the system, and an agent answers open-ended questions about the resulting data.
Upgrading existing agents
An agent you have already built in Copilot Studio gets knowledge sources, tools and controls, or joins a team of agents in Groupy.
What the agent connects to
An agent is only as valuable as the systems it can reach. We connect it to your core systems in a way that limits it to defined actions, within the user's permissions.
SAP, including On-Prem
Through an MCP layer that exposes defined entities and actions to the agent, not open access to the system.
Priority
Through the Priority API, with per-user permissions.
Dynamics 365 and Dataverse
Including Dynamics 365 Finance & Operations, through prebuilt connectors or the Web API.
Microsoft 365
SharePoint, Outlook and Teams, with each user's existing permissions.
Azure AI Search
When the knowledge is too large or complex for the built-in indexing, with an index you can also share with other agents.
Any system with an API
Through Power Automate, a custom connector or an MCP layer.
Copilot Studio, or Azure AI Foundry?
Copilot Studio is not the right choice for every agent. Our default: Copilot Studio for internal processes that live in Microsoft 365, and Azure AI Foundry for bots that serve customers and residents.
Copilot Studio fits when:
- The agent serves employees, inside Microsoft 365 and Teams
- Prebuilt connectors exist for the systems you need
- Usage volume is low to medium
- Your organization already governs Power Platform, with environments and DLP policies
We consider Azure AI Foundry when:
- The bot serves customers or residents and needs its own interface and authentication
- Usage volume is high and consumption-based pricing becomes a burden
- The logic is very complex and becomes hard to maintain in a low-code tool
- You need full control over the model, the flow and monitoring
Copilot Studio in real-world use
Invoice agent for SAP On-Prem
An MCP layer that lets a Copilot Studio agent check for duplicates and write to SAP, with user approval before every write.
Copilot StudioMCPSAP MoroccanoilExport process automation
From a single Packing List email to a complete export document package in Dynamics 365 F&O, plus an agent for open-ended questions about the data.
Copilot StudioD365 F&OMCP Zooz-tech productGroupy: a team of agents
Teams of AI agents that cross-reference data and make decisions, and can also bring in Copilot Studio agents that already exist in your organization.
Multi-AgentCopilot StudioWhat we hand over at the end of the project
An agent in your environment
The agent is built in your organization's Power Platform environment, in its region, with development, test and production environments.
Integrations and controls
Connectors, an MCP layer where needed, DLP policies and permissions, and user approval at every point that calls for human accountability.
Testing and acceptance criteria
Conversation scenarios and real questions from the process, which the agent must pass before handover.
Ongoing cost estimate
An estimate of usage consumption (Copilot Credits) based on expected volume, before you commit.
Documentation and training
Documentation of the agent, its sources and integrations, and training for whoever will maintain it in your organization.
What an agent needs to keep working
An agent is not a one-off project. These are the things that need ongoing maintenance, and you can leave them with us or hand them over to your team:
Knowledge sources
Documents and procedures change. Someone has to keep them current, or the agent answers from outdated information.
Usage consumption
Tracking Copilot Credits, especially as usage grows or the agent opens up to more users.
Answer quality
Reviewing conversations and the questions the agent could not answer, and improving accordingly.
Platform updates
Microsoft updates Copilot Studio frequently. We check that the agent keeps working, and put new capabilities to use.
From one process to an agent in production
Milestones, client demos at every stage and acceptance criteria for deliverables. How we manage projects
One process
We pick one process with clear value and define what the agent does, which systems it works with and what counts as success.
Prototype
A working agent on real data early on, so you can see it and try it before moving forward.
Integrations and testing
Connecting to your systems, setting up permissions and controls, and testing against the acceptance criteria.
Go-live and ongoing support
Rollout to a first group of users, stabilization and expansion, with the same team.
Frequently asked questions about Copilot Studio
How much does it cost to run an agent in Copilot Studio?
In Copilot Studio, you pay for the usage units (Copilot Credits) the agent consumes, based on the type of response and action. During the specification phase, we estimate consumption based on expected usage volume, so the ongoing cost is known in advance.
Can a Copilot Studio agent connect to SAP On-Prem?
Yes. In the Gil Car project, we built an MCP layer that lets a Copilot Studio agent check for duplicates and write invoices to SAP On-Prem, with user approval before every write.
Where is the data stored?
The agent runs in your organization's Power Platform environment, in its region. Data in the source systems stays where it is, and the agent accesses it according to the user's permissions.
Can you build a customer-facing bot in Copilot Studio?
A Copilot Studio agent can also be published on a website. For bots that serve customers and residents, we usually recommend Azure AI Foundry, because of the need for a custom interface, authentication and full control. The full comparison is in our guide.
We already have an agent in Copilot Studio. Can you help?
Yes. We add knowledge sources, integrations and controls to existing agents, and can bring them into a team of agents in Groupy.
Further reading: AI agents · Copilot Studio vs. Azure AI Foundry · AI agent for Dynamics 365 · AI workshops for organizations
Do you have a process that is a good fit for an agent?
In a 30-minute call, we will walk through one of your processes and tell you whether Copilot Studio fits it, what needs to be connected and what goes into the first phase.