AI automation inside a cross-border logistics operator

Since 2026 my work at euShipments has been almost entirely about putting AI into processes that already exist and already matter. Three lines of work run in parallel.

Customer support. Every incoming ticket is classified, routed to the colleague who should own it, and paired with a suggested reply that is built from real data pulled out of the internal systems, so the person answering starts from facts rather than from a blank field.

Finance. Every month, invoices and service breakdowns arrive from the carrier partners. The system checks each line against what the contracts actually say, learns from every new annex, and looks for patterns that no line-by-line review would catch: a charge that is valid in principle but cannot plausibly apply at that frequency. When it finds a mismatch, it produces the specifics needed to dispute it.

Internal knowledge. An internal MCP server lets colleagues connect their own AI assistants to the company's real data. Each person authenticates with their own key, their role decides which tools their assistant can use, and answers come from the systems rather than from the model's imagination.

One decision runs through all three: sensitive data is collected and masked before anything reaches a language model. That layer took longer to design than the AI parts, and it is the reason the systems can run in a real company at all.