Skip to content
DataDream
← All articles
AI Tools7 min

Copilot Studio: a guide for Dutch companies (2026)

Laurens van Dijk, oprichter van DataDream

Laurens van Dijk

Agentic Engineer, DataDream

Share

Building agents in the Microsoft environment

Copilot Studio is Microsoft's platform for building AI agents. For organisations that already run fully on Microsoft 365, with Teams, SharePoint and Power Platform, it is the most obvious way to make agents without hiring developers first. For organisations that do not run on Microsoft, another route often makes more sense. This article covers what Copilot Studio can do, how the costs work and when you are better off choosing something else.

What is Copilot Studio?

Copilot Studio is a low-code platform within Microsoft Power Platform. You use it to build AI agents without programming. The agents run on OpenAI models and, since September 2025, also on Anthropic models. They use your Microsoft 365 data through Microsoft Graph, and you deploy them in Teams, on a website or in Power Apps.

For many practical small-business uses that works well. An HR agent that answers questions about leave based on the staff handbook in SharePoint. An agent that drafts quotes from Dynamics 365. An IT helpdesk that classifies tickets and routes them. You need no Python, no servers of your own and no outside party for that.

Where Copilot Studio is strong

A first version, fast. With the visual editor and Power Platform's ready-made connectors you quickly have a working prototype. Inside a Microsoft environment it is the fastest way to test whether an idea works.

The link with Microsoft 365. SharePoint, OneDrive, Outlook, Teams, Excel and Power BI are available straight away, without managing access keys yourself. An agent that answers from your own documents is mostly a matter of configuration.

IT governance. In the Power Platform admin center, IT decides who may build agents, which data they may use and how long conversations are kept. For organisations with data loss prevention rules and other compliance requirements, that saves a lot of setup work.

Data within the EU. Copilot Studio complies with the EU Data Boundary when your tenant has a billing address in the EU and all your environments are in a region inside the EU. One caveat: Copilot Studio can be set up so that generative AI features also run in other regions, for example when capacity is tight. Check that setting if your data must not leave the EU. For the broader AI Act context, see /ai-act.

Where Copilot Studio falls short

Costs at high volume. Copilot Studio bills in Copilot Credits, per action of the agent. According to Microsoft, a classic answer costs 1 credit, a generative answer 2 credits, an agent action 5 credits and an answer grounded in your tenant's data 10 credits. One question can therefore cost several credits. With large numbers of conversations, such as in customer service, costs add up. Then compare it with a custom solution that calls an Anthropic or OpenAI model directly.

Limited autonomy. Copilot Studio agents are mostly conversations with fixed steps in them. An agent that gets a goal, makes its own plan, chooses tools and tries another approach when something fails is harder to build than with your own Python setup or a framework such as LangGraph. For tasks with many steps and unpredictable input, custom work often fits better. See /ai-agents for that approach.

Tied to Power Platform. An agent built in Copilot Studio runs only inside Power Platform. If you later want to move to another platform or your own setup, you rebuild it.

How billing works

Copilot Studio bills per use, not per user. You buy Copilot Credits in monthly packs or pay afterwards per use. Current prices are on Microsoft's pricing page.

An important exception, with conditions. When a staff member with a Microsoft 365 Copilot licence uses an internal agent, and the agent works with that staff member's authenticated identity, Microsoft charges no credits according to its billing documentation. Limits apply. Flows the agent starts are only included when they run through the "When an agent calls the flow" trigger; flows with other triggers consume credits at the standard rate. Usage is subject to fair-use limits that Microsoft can change. Computer-using agents are not included. And an agent that starts on its own without a licensed, signed-in staff member, or a customer-facing agent on your website, always costs credits. So do not budget an internal agent for colleagues with Copilot as free by default; check per component whether the exception applies.

To work it out yourself:

  1. Estimate how many conversations the agent has per month and what happens in an average conversation: classic answers, generative answers, actions, searches in your own data.
  2. Add up the credits per conversation using the rates above and multiply by the number of conversations.
  3. Set the price of the credits against that, and compare it with what the same conversations cost through a model's API, plus building and maintaining your own solution.

For an internal agent with a few hundred questions a month, Copilot Studio is usually perfectly affordable. With thousands of conversations a day, it pays to calculate first.

When is Copilot Studio a good choice?

Your organisation runs fully on Microsoft 365. Identities (Entra ID), files (SharePoint), mail (Outlook), chat (Teams), processes (Power Automate) and data (Dataverse): a Copilot Studio agent uses all of that directly. With your own setup you would have to rebuild those connections.

The task is clearly defined. Answering questions from your own documents, starting a process based on a question, classifying and routing a ticket. No complex multi-step planning and no extensive judgement calls.

You want to work without developers. People who already work with Power Platform, analysts and process owners can build agents themselves. That helps to use AI more widely in the organisation.

The volumes are manageable. With a limited number of conversations, or use by signed-in staff with a Copilot licence within the conditions above, costs stay predictable.

When does a custom agent fit better?

Many conversations on a clear topic. A customer service operation with tens of thousands of conversations a month can do better with a direct link to a model and its own search layer. You also keep control over your prompts and data.

Strict requirements on where data lives. If data has to stay on your own servers, for example in healthcare or finance, an open model (such as Llama or Mistral) on your own infrastructure, or a setup in Azure with a private connection, fits better.

An agent that plans for itself. If an agent has to break down a process, plan subtasks, choose tools itself and try another approach after a failure, custom work in Python with Claude or a GPT model, or a framework such as LangGraph, gives much more control.

Voice at the core. For applications that mainly run over the phone, such as an intake or a reception desk, platforms like ElevenLabs, Vapi and Retell are built for it. See /voice-agents.

Two tracks side by side

You can also use both. Copilot Studio for internal agents, such as HR questions, the IT helpdesk or price lists for sales, where the link with Microsoft 365 is worth a lot. A custom agent for customer-facing uses with many conversations or voice, where cost and flexibility weigh more. One Microsoft environment, two kinds of agents, one set of governance agreements.

DataDream advises on this without a vendor preference: for each application I weigh whether Copilot Studio, an agent on a voice platform or a custom API-based solution fits best. For replacing RPA, see /rpa.

How do you start?

Step 1: map your applications. Do not start from the tool, start from the work. For example: "Our HR team gets a lot of questions every month about leave, holiday pay and sick leave." For each application, determine the number of conversations, how predictable the questions are, which data is needed and which rules apply. The free AI Scan helps you get started.

Step 2: start small. One well-defined application, two to four weeks, with a small group of users. Measure what it delivers in time or quality and how many credits it costs. Compare that with a custom solution for the same use. Without those numbers every choice is a guess.

Step 3: expand or switch. If the trial scores well on value and cost, take on the next application. If it does not work because it is too expensive, too limited or too slow, build the same application as a custom agent, before the organisation is tied to the wrong platform.

For advice on choosing between Copilot Studio, custom work or a mix, see /ai-strategie. For custom agents, see /ai-agents. Want me to look at which route fits your application? Book a free call.

Curious what AI can do for your business?

Take the free AI Scan and find out in 1 minute.

Frequently asked questions

What is Microsoft Copilot Studio?
A low-code platform within Microsoft Power Platform for building AI agents without programming. The agents run on OpenAI models and, since September 2025, Anthropic models too, use your Microsoft 365 data through Microsoft Graph and are deployed in Teams, on a website or in Power Apps.
What does Copilot Studio cost?
You pay per use in Copilot Credits, not per user. According to Microsoft a classic answer costs 1 credit, a generative answer 2, an agent action 5 and an answer grounded in your tenant's data 10. When a signed-in staff member with a Microsoft 365 Copilot licence uses an internal agent with their own identity, Microsoft charges no credits, within fair-use limits. Flows with a trigger other than "When an agent calls the flow", computer use, agents that start on their own and customer-facing agents do cost credits. Credit prices are on Microsoft's pricing page.
When do you choose a custom agent instead of Copilot Studio?
With many conversations on a clear topic, when data must stay on your own servers, for an agent that plans multi-step tasks itself, and for applications that mainly run over the phone. Copilot Studio fits best for well-defined agents inside a Microsoft 365 environment.
Is Copilot Studio GDPR-compliant?
Copilot Studio complies with the EU Data Boundary when your tenant has a billing address in the EU and all your environments are in a region inside the EU. Do check the setting that allows generative AI features to run in other regions. Under the GDPR you remain responsible for a data processing agreement, the permissions on your data and what you state in your privacy statement.