AI and ERP: from understandable hesitation to controlled orchestration

Most organizations see the potential of AI — but hesitate as soon as it concerns their ERP. Rightly so. We understand that. That is precisely why we take a different approach.

That hesitation is understandable

In conversations with managers and directors, we hear it time and again: “AI sounds interesting, but we prefer not to touch our ERP.” And honestly, we fully understand that. Your ERP is the backbone of your organization. It contains your customer data, your financial flows, your logistics. The idea of an AI model roaming freely in there — without clear boundaries, without control, without an audit trail — is not an appealing prospect.

So the question is not whether organizations want to use AI. That desire is present almost everywhere. The question is: how do you do it in a way that is secure, verifiable, and manageable?

Why existing AI integrations fall short

Many AI integrations today work via direct API connections or broad access permissions. This creates a number of problems: you don't know exactly what the AI has done, you cannot easily roll it back, and you have limited control over what is and is not possible. For a chatbot answering frequently asked questions, that might be acceptable. For an agent executing actions in your ERP, that is definitely not the case.

What you need is not a loose integration, but a controlled intermediary layer. A layer that determines what AI can ask, what it can execute, who authorized it — and that records everything.

How Any2info tackles this: agents, MCPs, and a platform that maintains control

At Any2info, we have embraced MCP (Model Context Protocol) as the foundation for secure AI orchestration. MCP makes it possible to expose systems — your ERP, BI solutions, Microsoft 365, and other platforms — via standardized tools to AI models such as Claude (Anthropic), ChatGPT (OpenAI), or Gemini (Google). These models can then not only retrieve information but also perform actions — in the systems you link, in the way you define. But the key lies in how you set that up.

Our platform acts as the orchestration layer between AI and your business systems. In practice, this means:

• MCP server builder: with the Any2info MCP builder, you determine exactly what an agent is allowed to see and do. You configure which tools, prompts, and resources are available — and link them to platforms such as Exact (Globe, Synergy, Online), AFAS, or Business Central. Instead of a generic ERP link, you can set up a targeted MCP for Sales, HR, Service, or Finance. Your processes, your boundaries.

• External MCP servers: in addition to your own MCP servers, you can link external servers — such as Microsoft 365, scheduling software like vPlan, BI platforms like Sumatra, or any other MCP-compatible system. This allows AI to work not only within your ERP, but across your entire software landscape.

• Controlled access via OAuth: every connection between AI and your systems goes through OAuth and standard authentication protocols. No broad access permissions, no open doors — only explicit, managed connections that precisely define who is authorized for what.

• Structured interaction: AI agents do not roam freely in your systems. They act exclusively via predefined data streams and tool calls. What an agent is allowed to ask and execute is determined by you in advance. No surprises, no unintended actions.

• Logging and reporting: every interaction between AI and your business systems is recorded: who asked what, what action was performed, and what the result was. Fully transparent, fully verifiable — also after the fact.

• Humans remain at the helm: for critical actions, the platform includes an explicit approval step. AI analyzes and makes proposals. The human decides. Always.

From individual agents to true orchestration

With this approach, AI does not become a separate tool alongside your ERP — it becomes a controlled layer above it, capable of combining multiple systems, interpreting information, and then acting in a structured manner.

No , but — with full control at every step.

That is what we mean by AI orchestration: not AI that is loosely connected to your systems, but AI that works responsibly, structured, and securely with your existing software landscape.