The Challenge
Scaling a digital freight platform comes with unique operational complexity. The core problem was not a lack of ambition for AI, but the structural conditions that made scaling it both necessary and difficult. Four challenges stood in the way:
- Heavy reliance on unstructured logistics data: Freight forwarding runs on emails, attachments, and documents. Critical shipment information was buried in text-based workflows that traditionally required manual handling at every step, making automation both urgent and technically demanding.
- Operational bottlenecks at high volume: Operations teams manually uploaded documents and assigned them to the correct shipments; repetitive work at significant daily volume that directly limited how fast Shypple could scale without growing headcount proportionally.
- AI adoption risk: As a fast-scaling tech company, Shypple wanted to avoid the most common AI pitfalls: disconnected experiments, non-scalable prototypes, and accumulating technical debt that would slow down future development rather than accelerate it.
- Speed versus scalability tension: The team needed to demonstrate immediate business value without compromising the long-term architectural robustness required to support multiple future AI use cases.
Our Approach
To address both the urgency and the structural ambition, we combined pragmatic execution with long-term AI architecture thinking — working across three parallel tracks from day one.
Building a Production-Grade AI Infrastructure
Before scaling use cases, we established a production-grade AI infrastructure:
- Scalable model training and deployment pipelines
- Robust monitoring and continuous improvement loops
- Clear ownership and reproducible workflows
This created a foundation where AI models could be reliably deployed and iterated without becoming dependent on individual contributors.
Automating Core Operational Workflows
We focused on the highest-impact use case first: email and document processing.
- Automated extraction of shipment information from incoming emails and attachments
- Intelligent assignment of documents to the correct shipment
- Human-in-the-loop validation where needed
What was once a manual upload and matching process is now largely automated. AI performs the extraction and assignment, while operations staff review and accept AI-generated submissions, or give feedback when needed.
The result: an 80% improvement in process speed, significantly reducing manual workload across thousands of daily documents.
Embedding AI into the Organization
To setup Shypple for long-term succes, we worked side by side with the internal software engineering team and supported in the hiring process of AI engineers. This pragmatic cooperation was appreciated by Shypple. Moaz (AI Engineer @ Shypple): “In my daily collaboration with Enjins, we see what tasks are planned and divide them between us. We’re just working together as one team”.
Within nine months, Shypple not only had production AI systems running, but also an internal AI capability ready to grow.
From design sprint to partnership
