Somewhere in a logistics company's inbox, a PDF just arrived. It contains a shipping order — pickup address, delivery address, cargo details, dates, references. Someone needs to open it, read it, and type all of that information into the order management system. Then another one arrives. And another.
This is exactly the kind of problem Intelligent Document Processing (IDP) is built to solve — and one of the most persistent operational bottlenecks in logistics and supply chain, where document volumes are high, the tolerance for error is low, and the people doing the data entry are far too skilled to be spending their time on it.
The Challenge: Manual Order Entry at Scale
For one of our customers — a European logistics group operating across multiple countries — this was daily reality. Incoming order PDFs from dozens of business partners, each with their own document layout and conventions, all needing to be manually re-entered into CarLo, their transport management system (TMS) by Soloplan.
The instinctive answer is “just use OCR.” But document variation is only one layer of the problem. The extracted data still needs to reach a downstream enterprise system in exactly the right structure. And any automated pipeline operating in production needs to handle exceptions gracefully — escalating what it cannot resolve, never dropping documents silently, always leaving a clear audit trail. Solving all three layers together is where the real engineering challenge lies.
The Solution: An Intelligent Document Processing Pipeline on Azure
We designed and built an Intelligent Document Processing pipeline on Microsoft Azure — one that reads incoming order PDFs, understands their contents, enriches them with known master data, and delivers structured data directly into CarLo, automatically.
The backbone is Azure AI Document Intelligence — part of the broader Azure AI Foundry platform — Microsoft's service for understanding the structure and content of documents at scale. But the AI model is only one piece. Around it, we designed an architecture with a clear separation of concerns: document ingestion, sender identification, intelligent extraction, data normalization, system integration, and operational monitoring — each handled as a distinct, purposeful layer.
One of the more deliberate design choices was how we approached the AI models themselves. Rather than relying on a single general-purpose model, we trained dedicated models per business partner — each one tuned to the specific document format that partner uses. The accuracy gains are significant. But this also required a smart routing layer that identifies the incoming document's origin and directs it to the right model, including fallback logic for cases where that identification is not straightforward. We also built in logic that lets the customer onboard new business partners into this process independently, without any code changes.
The output of the pipeline is not raw extracted text — it is a clean, validated, fully structured data payload delivered in exactly the format CarLo expects, ready for processing without any human touch.
Designing for Production, Not Just for the Demo
What separates a proof of concept from a system teams can actually rely on is how it behaves when things do not go as expected.
We put as much thought into the failure scenarios as into the happy path. When the system encounters something it cannot resolve with confidence, it does not guess — it escalates, notifies the right people, and preserves enough context for a human to step in and act. Nothing is dropped silently. Every outcome is traceable. The right team is notified with the right information, automatically.
This operational discipline is, in our experience, what customers remember most. The AI accuracy matters. The integration precision matters. But the trust that builds over time comes from knowing the system behaves predictably even at the edges.
The Impact of Intelligent Document Processing on Operations
The most direct result is time returned to the team. Order data that previously required manual entry now flows into the system automatically. Beyond that, there is consistency — the same process, applied the same way, every time — and scale, as the solution handles growing document volumes across multiple European entities without growing headcount.
For the operations teams, it means less time on repetitive data work, and more on the decisions that actually require human judgment.
A Note on Intelligent Document Processing
This project is a strong example of where AI delivers measurable, near-term value in enterprise operations — not as a distant promise, but as a working system in production today. The technology is mature, the Azure ecosystem provides the right building blocks, and the integration patterns are well understood.
What it requires is the right combination of domain knowledge, architectural thinking, and experience building systems that hold up in the real world. That is precisely what we bring to this kind of engagement. See more of our work on the oneLake reference & case studies page.
Managing high volumes of incoming documents?
Orders, invoices, delivery confirmations — if it arrives as a PDF and needs to end up in a system, we'd be glad to explore what an Intelligent Document Processing solution could look like for you.
oneLake GmbH is a Microsoft Solutions Partner specializing in Data & AI and Azure. We help organizations in Germany and across Europe build intelligent data solutions on the Microsoft cloud.