Why did the need arise?

In Academy of Brain’s sales process, the salesperson must identify the client’s challenge and find a suitable training package from a broad course catalogue. The catalogue holds over a hundred training videos — meaning their combinations allow practically limitless options, from which the optimal mix for each client should be found efficiently.

In practice this meant the salespeople (5 people) had to work through extensive documentation: content descriptions, keywords and transcripts. The process was slow and prone to variation:

• an experienced salesperson might know the catalogue well
• a newer salesperson left relevant courses unrecommended

The result was an uneven customer experience and potentially lost sales opportunities.

The other starting challenge was structural: Academy of Brain didn’t previously have an environment or structures on which agent solutions could be built.

What goals were set for the project?

The goal was to build an AI agent that:

• compares the client’s need against the entire course catalogue and recommends suitable trainings with its reasoning
• works as the salespeople’s internal tool within their normal working environment (Teams)
• supports decision-making but leaves the final choice to the salesperson (human-in-the-loop)

Alongside this — and at least as important — the goal was to build the client’s capability to keep developing on their own. Astu Labs didn’t want to deliver just a solution, but to teach Academy of Brain’s own representative to build similar agents independently.

How was the agent built?

What the agent does

The salesperson describes the client’s challenge or need to the agent. The agent compares the description against Academy of Brain’s entire training catalogue using RAG retrieval (Retrieval-Augmented Generation) and returns a recommendation of suitable courses with its reasoning.

This is not a mere search function or chatbot, but an agent that understands context: the nature of the client’s challenge, the content of the courses, and the connection between the two. The agent independently concludes which courses best match the described need — and justifies its choices.

The salesperson makes the final call. This is a human-in-the-loop model, where the agent supports decision-making but doesn’t replace the salesperson’s judgement.

Technologies used

The solution is built entirely on Microsoft technologies with no third-party components. The implementation is instruction-based: the agent’s behaviour is steered with instructions (system prompt), topics have been customised and the language model chosen to fit the use case.

How the project progressed

Active collaboration amounted to roughly one person-week of work from Astu Labs. In calendar time the project took several months, because configuration changes by the client’s IT partner were delayed.

One factor helped: the maturing of the technology from summer 2025 into 2026 made it possible to build a production-ready solution with Copilot Studio.

What did the project deliver?

Concrete outputs:

• A production AI agent in use by the sales team (5 people)
• Configured Microsoft environments that lay the foundation for future agent solutions
• The client’s own capability to build similar agents independently going forward

Business benefits

Consistency: Personal variation in course knowledge between salespeople disappears. Through the agent, every salesperson gets an equally comprehensive view of the catalogue regardless of their own experience.

Time savings: Salespeople no longer need to manually work through extensive documentation. The work used to take days; now the agent makes the comparison in seconds.

Sales opportunities: The agent can surface courses the salesperson wouldn’t have remembered or known — potentially growing sales.

The client’s voice

“With his natural guidance the project moved smoothly from start to finish, and challenges were solved efficiently. It’s now easy for me to continue the development work on my own”

— Academy of Brain’s representative on Astu Labs’ Jouni Manninen and the project delivery

What did we learn?

The project crystallised several lessons that apply more broadly to building AI agents:

• The strength of the implementation is its simplicity — Copilot Studio + SharePoint RAG + instruction-based steering is enough to produce a production-ready agent
• The critical success factor is not the technology but the quality of the source material and refining the agent’s instructions together with business experts
• The client’s own participation in the build is change leadership — it ensures both the transfer of skills and commitment to the end product
• At its best, a partner’s role is to enable the client’s own capability, not to create dependency on an outside supplier

Who is this case useful for?

This case is especially relevant for:

• Companies with a broad product or service offering and a sales team that has to match customer needs to the right solution
• Industries such as education, consulting, software and professional services, where the sales process demands deep knowledge of the offering
• Growth companies and SMEs that want to put AI agents to work without a technical team of their own — while building the capability to keep developing independently

How does an agent project like this usually proceed?

The Academy of Brain case reflects a typical path when building an AI agent to support sales or customer service:

1. Clarifying the business need — what is the agent’s concrete task, and how does it support people (not replace them)
2. Collecting and curating the source material — the quality of the RAG knowledge source determines the quality of the agent
3. Configuring the environment — Copilot Studio, SharePoint, Microsoft 365 integrations
4. Building the agent and refining its instructions together with the client
5. Quality testing and iteration — ensuring reliability
6. Release to production and adoption into the sales team’s workflows

The critical phase is refining the instructions together with the client — that is where both the agent and the client’s own capability are born.

Considering something similar?

If you have a broad product or service offering and a sales team that spends time finding the right solution, an AI agent can make the work more efficient — and more consistent.

→ Book a 30-minute conversation — we’ll go through how an agent would fit your sales process, and what source materials building it would require.

From Astu Labs, Jouni Manninen took part in this case, responsible for the agent’s technical implementation and for guiding the client. Meet Jouni and the Astu Labs team here.