A brand agent in Copilot – fix the brand portal first

What is a brand agent in Copilot?

A brand agent in Copilot is an AI assistant that answers brand questions from the approved content in a brand portal. Employees ask it in Microsoft 365 Copilot instead of emailing the brand team.

The need comes from everyday life in large companies. Brand teams answer the same questions every week. Which logo version applies here? How much clear space does the logo need? Where is the current template? Can the logo go on a photo? A surprising number of requests are about signage, office glass, trade fair stands or packaging. With thousands of employees, this turns into a helpdesk that can't keep up.

A brand agent catches the routine questions, points people to the right assets and passes difficult cases to the brand team. If it should create presentations too, the result depends on a well-built PowerPoint master.

How Frontify connects to Copilot via MCP

Frontify connects its brand portal to Copilot through its own MCP server. MCP stands for Model Context Protocol, an open standard that lets AI agents work with external systems. In Microsoft Copilot Studio, the server is added to an agent as a tool using the MCP onboarding wizard. That is how both Frontify and Microsoft describe it in their documentation.

The server isn't switched on by default. An account admin enables it in the Frontify settings, at no extra cost and without going through the customer success manager. The default is the "discovery" pack with 25 read-only tools covering guidelines, libraries, assets and templates. Other packs can change data in Frontify.

On the Microsoft side, access runs through Power Platform connectors. Any data policies your IT team has set there also apply to the brand agent. The technical setup itself is manageable once it is clear who in IT owns Copilot Studio.

Why the brand portal decides whether the agent helps

A brand agent is only as good as the content it answers from. A properly configured agent uses approved content from the brand portal and nothing else, and it doesn't invent rules. That is the right approach, with one direct consequence: whatever isn't in the portal, the agent can't answer.

Take an example. A regional office asks how the logo should be applied as vinyl on a glass door. If the portal has no rules for glass surfaces, the question stays open or ends up back at the helpdesk. Add up a few dozen gaps like this and you get an agent that answers politely and rarely helps. Employees notice quickly and go back to email.

Then there is the question of currency. If the portal holds two versions of the same template, the agent may point to the old one. Frontify itself warns that AI output can look correct and still be wrong. So every agent project starts with work on the content.

The content audit: testing helpdesk questions against the portal

A content audit shows which brand questions an agent could answer today and which it couldn't. We put the most frequent helpdesk requests next to the brand portal and check each one. Is there a clear answer? Can someone find it without prior knowledge? The result is a prioritised list of gaps.

The form of the content matters too. According to Frontify, AI tools with image analysis can read images and diagrams in guidelines. An agent still answers more reliably when a rule is written out as text: one rule per page, a clear title, measurements and colour values spelled out. Embedded PDFs and rules that only exist in a graphic are the most common weak spots.

The audit also produces a test catalogue of real questions, each with the expected answer. The agent is checked against this catalogue before launch and after every major change to the portal.

What a brand agent shouldn't decide

Approvals stay with people. A brand agent can explain which rule applies and where the right asset is. It shouldn't sign off packaging for production or a campaign for print.

Uploaded designs are a similar case. Copilot can analyse images and give pointers, for example on clear space that is too tight or a colour outside the palette. That is not a reliable brand check, and the agent shouldn't be introduced internally as one. Present it as a reviewer and you create expectations it can't meet.

Write access calls for care as well. Frontify offers packs that let an agent upload assets, move them or add metadata. Frontify recommends starting with the read-only pack and backing up important data before enabling anything else. For a brand agent that answers questions, read-only access is enough.

Brand agents in several languages

Copilot Studio supports multilingual agents, with one limitation brand teams should know about. Every agent has a primary language that can't be changed after the agent is created. Other languages are added as secondary languages, and content and topics are edited in the primary language. For agents that use generative orchestration, Copilot Studio translates generated responses automatically. Messages you write yourself have to be translated and uploaded.

For a brand agent used around the world, the primary language should therefore be the language the brand portal is maintained in, usually English. Whether answers in Japanese or Portuguese are as precise as in English only shows in a test with real questions from those countries.

Brand terms need particular attention. Names of design elements, product lines or templates should stay the same in every language, even when a translation seems obvious. A short glossary in the agent's instructions sets out which terms stay untranslated.

Presentations from a brand agent: the master decides

Many companies want the brand agent to create presentations too. The result is only as good as the PowerPoint master behind it. Copilot works with what the template offers: real placeholders in the layouts, theme colours rather than hard-coded values, clean font settings and layout names that say what each slide is for. Without that foundation, Copilot designs off-brand, however well the agent is configured.

That is why the brand agent and the master belong in the same project. The agent answers questions and points to the right template. The master makes sure the slides Copilot produces from it look like the brand.

We explain how a master needs to be built for this in our article on the AI-ready PowerPoint master. How the brand kit in Copilot works with the template is covered in our article on Copilot brand kits and PowerPoint templates.

Frequently asked questions about brand agents in Copilot

What is a brand agent in Copilot?

A brand agent is an AI assistant in Microsoft 365 Copilot that employees can ask about logos, colours, templates and other brand rules. Its answers come from the company's brand portal. Requests it can't resolve with confidence go to the brand team. The agent is built and configured in Microsoft Copilot Studio.

How do you connect Frontify to Copilot?

Through an MCP server that Frontify provides itself. An account admin enables it in the account settings, and there is no charge for this. IT then adds the server to an agent in Microsoft Copilot Studio as a tool, using the MCP onboarding wizard. With the default pack, the agent can read content but not change anything. Power Platform data policies apply to the connection.

Why does the agent not answer some brand questions?

Because the brand portal gives it nothing to go on. The agent sticks to approved content and doesn't add rules of its own. If there are no guidelines for vehicle livery, for instance, the question goes unanswered and is passed to the brand team. Comparing helpdesk requests with the portal shows which frequent questions run into gaps like this.

Can a brand agent review or approve designs?

No, approval decisions remain with the brand team. Copilot can analyse uploaded designs and flag issues such as colours that don't match or a misplaced logo. That helps while designing, but it isn't a reliable brand check. Internally, the agent is best introduced as a source of information on rules and assets.

Does a brand agent work in several languages?

Yes. Copilot Studio works with a fixed primary language, in which content is maintained, plus secondary languages. If the agent uses generative orchestration, Copilot Studio translates generated responses automatically. The language the brand portal is maintained in is the natural choice for the primary language. Before roll-out in a country, employees there should test the agent with their own questions.

How long does the technical setup take?

The technical setup itself is manageable, as long as IT has access to Copilot Studio and ownership is clear. The timeline is set by the content work: the audit, revising the portal pages and testing against the catalogue of real questions.

Assessment and implementation of a brand agent

We split brand agent projects into two phases. The assessment shows whether an agent is worth it and what needs to happen first. We review the use cases, the content in the brand portal, the technical setup in Copilot Studio and the PowerPoint master. You come away with a concept and a list of gaps, a test catalogue of real questions and a recommendation on which content to revise before launch.

In the implementation, we build the agent's instructions and knowledge structure together with your IT team. We work through the portal content the audit flagged as missing with your brand team, and adjust the master where needed. Testing runs against the catalogue until the answers are right. At the end, your brand team has an agent it can maintain itself, with a short guide for adding new content.

Sources

• Frontify Help Center: Frontify MCP
• Microsoft Learn: Connect your agent to an existing Model Context Protocol (MCP) server, 2026
• Microsoft Learn: Configure and create multilingual agents, 2026