Skip to content
← Back

From inbox to order in Business Central

Dutch B2B supplier (name under NDA). Built via an Amsterdam AI consultancy. Live since July 2026.

Challenge

Orders came in by email: as free text, PDF, or Excel. The order desk retyped each one into Microsoft Dynamics 365 Business Central. Mail arriving after office hours waited until the next day.

Approach

An AI pipeline reads the order mailbox every five minutes. It extracts the customer, items, quantities, and units from the mail and its attachments, matches them against Business Central, and gets the order ready. A staff member checks the order and releases it; nothing goes out automatically. Price requests, changes, and internal mail are recognised and skipped. Every mail gets an audit row. Customer-specific rules live in Business Central itself, so the client adjusts them without a developer.

Stack: Python, Azure Container Apps, Microsoft Graph, Mistral (Azure AI Foundry), Business Central API, Terraform, GitHub Actions.

My role. I designed and built the system end to end: code, infrastructure, CI/CD, evaluations, handbook, and handover. The AI consultancy handled the client relationship and project management.

Result

Live since July 2026. Moved to the client's own Azure environment and handed over in September 2026, with a handbook for the order desk. The client adjusts its own rules in Business Central.

Highlights

  • Units. Pieces, boxes, or pallets: one missing unit makes an order a thousand times too large. Every order therefore gets a value check and a limit based on historical orders.
  • Never a duplicate order. A write that might already have succeeded is never blindly retried. When in doubt, a manual check follows: one visible task beats a silent duplicate order.
  • Recognising customers. First an exact match, then fuzzy matching, and a language model decides the doubtful cases. Branches of the same chain are never merged automatically.
  • In the client's own environment. Data, logs, and AI costs stay in the client's own Azure environment.

Technology

PythonAzureMistralBusiness CentralMicrosoft GraphTerraformGitHub Actions
© 2026 Storm IT SolutionsAmsterdam
NL