MIA doesn't exist, but she talks to 10,000 people a day. About LLM, KB, MCP, and other acronyms

In September 2025, the Saint-Gobain group took a step that might seem, at first glance, like just another digital marketing campaign: launched MIA, an avatar created with artificial intelligence, who presents the video series “Voice of the Future.” But beyond the glossy aspect of technology, Saint-Gobain has done something much more courageous: it has changed the traditional way in which a global construction corporation communicates with its professionals, customers, and partners.

For the wood industry—and for any B2B industry, for that matter—this moment marks the beginning of an era in which communication is no longer just about messages sent and received, but about intelligent, contextualized, and personalized conversations on a global scale.

I am passionate about this field, and I have been following and using artificial intelligence for some time now. I took advantage of this news to delve deeper into this topic and develop a little further how artificial intelligence can help us communicate with partners outside the organization and, why not, with colleagues within it.

But who (or what) is MIA?

MIA is not just a chatbot with a friendly face. It is a complex system that combines several levels of AI technology:

1. AI avatar with real voice and face – MIA presents video content about sustainable construction solutions from around the world, talking about Australia, India, China, and Brazil. It does not read from a teleprompter – it generates customized content for each country and context.

2. Chatbot integrated into the website – In addition to the avatar, Saint-Gobain also offers a text/voice chatbot (via Voiceflow) accessible on all pages of their website. The system answers questions about brands, solutions, and career opportunities without requiring user identification.

3. Confidentiality and security – Conversation data is deleted weekly, and the system does not collect sensitive data. Transparency is total.

For a company with 161,000 employees in 80 countries and a turnover of €46.6 billion, communicating at this level of complexity is a huge challenge. MIA solves this problem by providing consistent, real-time responses in multiple languages, 24/7.

Technological Infrastructure

To truly understand the potential of these systems, we must look beyond the user-friendly interface. An AI system such as MIA operates on multiple technological layers. Below, I will outline some of the existing technologies, as a source of inspiration for those interested in this field.

Large Language Models (LLM)

At the heart of any modern AI communication system lies an LLM—models such as GPT-4, Claude, or Gemini. These are the „brains” that understand natural language, context, and the nuances of conversation. For MIA, the LLM is trained to understand the specific field of sustainable construction, Saint-Gobain materials, and company policies.

Knowledge Base

An LLM alone could say many nice things, but it could also invent information. That is why modern systems use an internal knowledge base—hundreds of thousands of documents, product catalogs, technical guides, case studies. When a user asks about a specific product, the system searches this base and provides accurate, documented answers.

The key technology here is called RAG (Retrieval-Augmented Generation): the system searches for relevant information in the knowledge base and then generates a natural response in the user's language, combining the information found with the explanatory power of the LLM.

Model Context Protocol (MCP) – Universal Connector

This is where it gets really interesting. MCP is an open-source protocol recently launched by Anthropic (the company that created Claude) and already adopted by OpenAI, Microsoft, Google, Amazon, and other tech giants.

Think of MCP as a USB-C port for AI systems. Just as USB-C allows for standardized connection of any electronic device, MCP allows for standardized connection of AI systems to any data source or tool: Google Drive, Slack, GitHub, SQL databases, ERP systems, CRMs, e-commerce platforms, any type of application. From this moment on, a whole world of opportunities opens up.

MCP benefits for business:

• A single standard – No need to build separate integrations for each system
• Security – You control exactly what data the AI accesses and how
• Scalability – Add new data sources without rewriting the entire system
• Interoperability – The same AI system can work with tools from different vendors

For example, a manufacturer could connect an AI chatbot directly to its inventory system, CRM, technical documentation, and product catalog—all through a single standardized protocol.

The list is not exhaustive, but in principle, it can help you build any kind of automation you want. For those less familiar with technology, my recommendation is to use a solution (application) that already exists on the market and adapt your processes to it.

Examples from other industries

Production and Logistics

IBM Watson is used for:

• Predictive maintenance – AI analyzes sensors on cars and predicts malfunctions before they occur
• Quality control – Automatic detection of defects in images
• Supply chain optimization – Intelligent inventory management

Chatbots in manufacturing industries enable managers to update order status via voice, receive delivery alerts, and check equipment indicators—all conversationally, without complex interfaces.

ARIA (BrainBox AI)

In the construction industry, the system ARIA developed by BrainBox AI from Canada analyzes, advises, and automates tasks related to building life. It anticipates needs, optimizes energy efficiency, and interacts with users via text or voice.

For the wood industry, a similar system could manage, for example, wood drying processes, monitoring storage conditions, and optimizing energy consumption in factories.

B2B statistics that speak for themselves

• 58% from B2B companies already use chatbots, according to ProProfs
• 78% of leaders in manufacturing industries confirms that AI increases productivity
• The chatbot market will increase from $8.27 billion in 2024 to $27.29 billion in 2030
• Cost savingsE-commerce saves approximately $301 billion of the $1.3 trillion spent annually on customer service by utilizing chatbots.

Specific applications for the wood industry

Now for the important part: how could the wood industry use these technologies? Here are some realistic scenarios:

1. AI Assistant for Technical Specifications

A laminated wood or CLT manufacturer could create an AI avatar that:

• Answer technical questions about products 24/7, in any language
• Recommend solutions based on project requirements (dimensions, strength, certifications)
• Generate instant preliminary offers
• Send relevant technical documentation
• Connects the customer directly with the appropriate sales representative

2. Knowledge Base for Distributors and Partners

Imagine a system that brings together:

• All product catalogs updated in real time
• Certifications and compliance for different markets
• Installation and assembly guides
• Case studies and references
• Prices and availability (via MCP connected to ERP)

A distributor in Germany might ask in German: “I need 200 square meters of class 33 laminate flooring, certified for underfloor heating, deliverable in two weeks to Bavaria. What are my options?”

The system would search the database, check real-time stock, calculate shipping, and generate a complete quote—all within the conversation.

3. Training and Onboarding

For a furniture or prefabricated components factory, an AI agent could:

• Train new employees through interactive tutorials
• Answer questions about safety procedures
• Explain the production processes
• Provides solutions for equipment malfunctions

4. Intelligent Customer Service

Instead of contact forms and responses after 48 hours, an AI chatbot could:

• Instant solution 80% from recurring questions
• Transfer more complex situations to real people
• Collect feedback and analyze customer sentiments (reactions)
• Provide support across multiple channels (website, WhatsApp, Facebook Messenger)

5. Marketplace Intelligence

An AI system connected to external data sources (via MCP) could:

• Monitor competitors' prices
• Industry trend analysis
• Identify export opportunities
• Suggest strategy adjustments

All from public sources, of course. We are not talking about illegal activities, such as hacking into competitors' internal networks.

Challenges

It's not all rosy. Implementing these systems comes with challenges:

Costs

A simple chatbot for a medium-sized company with an international presence sells for between $10,000 and $30,000. We are talking here about dedicated applications that run on local servers. A complex system with an avatar, knowledge base, and integrations can cost between $50,000 and $200,000, or even more, depending on its complexity. Annual maintenance can be 15-30% of the initial cost.

But the good news is that by using the tools presented at the beginning, you can build a chatbot for your small organization on your own, at a much lower cost and using fewer resources. You can also use the many cloud-based applications of this type that are now available on the market. The ROI (Return on Investment) can be spectacular: companies report savings of up to 30% in customer service and increases of up to 40% in attracting new partners.

Data and Quality

The system is only as good as the data it is fed. If your knowledge base is chaotic, the answers will be chaotic. Data preparation can take 40-60% of the implementation time.

Language and Context

For specific markets (e.g., Eastern Europe, Scandinavia, Baltic countries), the system must be trained on local terminology, specific units of measurement, and national standards.

Security and Privacy

Perhaps I should have started with security and GDPR. Customer data, prices, strategies—everything must be protected. That is why many companies choose self-hosted solutions (on their own servers) instead of the public cloud.

It's not a question of “if,” but “when.”

Saint-Gobain did not create MIA to be “cool.” They did it because they understood that the future of B2B communication is already here. When you have 161,000 employees, dozens of brands, hundreds of thousands of products, and millions of potential customers in 80 countries, classic human communication simply does not scale. It has already reached its limits.

For the wood industry—whether we are talking about CLT manufacturers in Romania, furniture factories in Poland, or parquet distributors in Germany—these technologies are not science fiction. They are available now, at increasingly affordable costs, with open standards (MCP) that eliminate dependence on a single supplier.

The question is not whether the wood industry will adopt these technologies. The question is: who will be the first (or the first few) to take advantage of them?

A company that is currently implementing an intelligent AI system in the areas of communication, sales, and after-sales service:

• Provides superior experience to its customers
• Dramatically reduces response time
• Scale without hiring proportionally
• Collect valuable data about requests and trends
• Clearly stands out from the competition

Over the next 2-3 years, we will see these systems become standard in B2B, just as websites became standard in the 2000s. The companies that adopt them first will set the industry standards.

MIA from Saint-Gobain is not just a cute avatar talking about sustainable construction. It is a warning and an opportunity: the technology for intelligent, contextualized, and personalized communication on a global scale exists. It is open-source. It is accessible. It is time to think about how to integrate it into our industry.

Because in the age of AI, the question is no longer whether you will communicate better with your customers. The question is: how soon will you start?

For less technical readers:

• LLM (Large Language Model) – An AI model trained on billions of texts (data) that understands and generates natural language
• Chatbot – Software program that simulates human conversation
• AI Avatar – Digital representation (face + voice) of a person, animated by AI
• Knowledge Base – Organized database with information about the company, products, procedures
• RAG (Retrieval-Augmented Generation) – Technique that combines information retrieval with natural response generation
• MCP (Model Context Protocol) – Open-source standard for connecting AI to various data sources and tools