Reporting and planning with your business data.

From a business question to a verifiable answer

A pilot addresses one agreed workflow, such as a monthly report or an actual-versus-plan comparison. Together, we define metrics, periods, calculation rules and reference results. Your business team checks answers against the underlying data and decides whether the acceptance criteria are met.

MCP is an open standard for access to defined tools, such as document searches or metric queries. The server retrieves approved information from the source system and passes results to the selected AI application. Which data the model receives and where it is processed or logged depend on the agreed architecture.

Finance and production each have their own scope

When a question connects finance and production, such as the cost of downtime, we agree data relationships and calculation rules across both areas. A finance pilot includes neither machine control nor approval for production operation.

Finance, reporting and planning · Dr. Maher Hamid

Reports, actual-versus-plan comparisons, liquidity and document searches based on agreed sources and business rules. Abacus, Bexio, SAP, DATEV and approved planning or document sources may be relevant to the task. The pilot accesses financial systems read-only.

Production and engineering · Schady Hamid, B.Eng

Downtime analysis, OEE and batch context through OPC UA, PLC, MES and control systems. Network segmentation, plant load and operational safety have their own test and acceptance criteria. Control functions are a separate scope with risk assessment, an approval chain and an audit log.

Agree data flows and access before implementation

Depending on the AI application, an MCP server may be called locally or through a remote connection. A locally hosted MCP server does not make a connected cloud model local: query results may be transmitted to its provider. We therefore consider the source system, MCP server, AI application and model destination separately.

Before the pilot, we agree which data may travel over which connections. This includes authentication, user and client permissions, network boundaries, log contents and retention. You approve hosting and model providers; we test the technical controls with your IT team.

A scoped pilot before production approval

1. Define the task and assess feasibility

We agree business questions, data sources, business rules, owners and acceptance criteria. Then we check data quality, interfaces and permissions. Missing access or approvals are recorded as prerequisites.

2. Test the pilot against agreed criteria

Four to six weeks is a planning assumption for a clearly scoped pilot once access, suitable data and required approvals are ready. Implementation takes place in an isolated environment; your business team checks results against agreed reference cases.

3. Approve production operation separately

Moving into production has its own scope and acceptance process: permission tests, monitoring, logging, error handling, operational ownership and handover. Only then do the responsible people give a separate production approval.

Review storage, model destination and contracts together

The original records can remain in the source system while selected results are transmitted to a model and stored in chat histories or logs. Hosting the MCP server locally therefore says nothing on its own about data residency, retention or potential training use by the model provider.

We document the intended data flow and review provider terms, settings, subprocessors and required agreements with your responsible teams. Fully local operation is an architectural option to assess. Data protection and legal assessments remain with the people responsible for them.