The story
Overview
HarborCart's catalog and order volume had outgrown the manual rules used to choose warehouses, carriers, and split-shipment logic.
TENSVA designed a fulfillment intelligence service that scores options in real time and surfaces only the exceptions that still need a person.
The challenge
Ops specialists were making hundreds of routing decisions a day. The rules were tribal knowledge, so quality depended on who was on shift.
Goals
- Reduce avoidable fulfillment exceptions
- Keep humans in the loop for true edge cases
- Integrate without replacing the existing commerce platform
- Give leadership a clear view of routing quality
Approach
We audited historical exception tickets, then ranked the decision types that could be automated safely. The first release targeted warehouse selection and split-order logic, with a review queue for low-confidence cases.
The solution
A Python decision service sits beside HarborCart's order pipeline. Node adapters pass order context in, AWS hosts the scoring jobs, and an ops console lets supervisors override or teach the system.
UX / Architecture
The console is built for speed: a queue of exceptions, a recommended action, and the reason behind it. Supervisors can accept, override, or send a case back to rules review.
Key features
Confidence-scored routing
Each fulfillment option is ranked with a confidence score and a plain-language reason.
Human review queue
Low-confidence cases stay with ops instead of silently failing.
Exception analytics
Leaders can see which SKUs, regions, and carriers create the most friction.
Results
Dummy sample: HarborCart kept its commerce stack and added an intelligence layer around it. Ops still handles the unusual cases, but everyday routing no longer depends on a handful of specialists.
“Dummy sample: We did not need another dashboard. We needed the system to make better calls. TENSVA built that without asking us to rip out what already worked.”