Case studies/Logistics and freight
A global freight forwarder running 28,000 packets a month, with a four-hour clearance window between vessel discharge and filing. We rebuilt the packet desk as a pipeline: LlamaParse on the intake, CargoWise as the system of record, a queue the ops team now runs without us.
| Entry | Consignee | HTS | Conf. | Status |
|---|---|---|---|---|
| 88721345 | Pacific Pier Imports | 6109.10 | 0.96 | posted |
| 88721346 | Northbay Trading | 8517.12 | 0.91 | pending |
| 88721347 | Alpine Freight Co. | 4202.92 | 0.94 | review |
| 88721348 | Ridgeline Ops | 8471.30 | 0.97 | posted |
| 88721349 | Harbor & Co. | 6203.42 | 0.89 | review |
| 88721350 | Cedar Logistics | 4202.92 | 0.95 | posted |
| 88721351 | Whitepine Imports | 8517.62 | 0.92 | pending |
| 88721352 | Sunset Brokers | 6109.10 | 0.98 | posted |
At a glance
A single desk, a single system of record, a fixed clearance window. Everything else was implementation detail.
The engagement
The stack
ISO 27001 · ISO 9001 · DPA and NDA signed at kickoff.
Before, the packet desk
The desk worked. It worked the way desks work when the process has been stable for fifteen years and nothing inside the four-hour window gets to change. These were the three patterns we found in discovery.
Filers opened the BOL PDF, the commercial invoice in the carrier portal, and the draft customs entry in CargoWise. They cross-checked pieces, weight, origin, and HTS code by eye. On a clean packet, 14 minutes. On a packet with a mismatch, up to 40.
Pre-build baseline: 14 minutes per clean packet, 40 minutes per exception packet.
When a filer needed an HTS code fast, the fastest path was the last invoice from the same shipper. The product master was a separate system, slower to search, and not everyone on the desk had access. Classification drift over six months was measurable.
Pre-build baseline: ~3% of lines flagged by CBP audit for HTS inconsistency across the prior fiscal year.
Weight and piece discrepancies between the BOL, the commercial invoice, and the entry packet were the filer's responsibility in theory. In practice, they surfaced a week later during billing close-out, which meant customs amendments and, on bad months, penalty exposure.
Pre-build baseline: approximately 11% of packets required a post-filing amendment in the prior fiscal year.
What we built
The pipeline follows the same five stages we run on every logistics engagement. The details below are the ones we actually implemented for this hub, not a generic template.
Carrier portals polled on a 10-minute cadence. Broker email routed via secure mailbox. Dock scanner drop to SFTP. All normalised to a single packet ID before extraction.
Document type tagged on ingest. Handwritten dock BOLs routed to the LandingAI model, typed BOLs routed to LlamaParse. Classification confidence below 0.90 holds the packet for human tagging.
Shipper, consignee, pieces, weight, country of origin, and line-level HTS. LlamaParse primary, AWS Textract fallback for packets LlamaParse cannot parse inside the SLA.
Pieces and weight agree across BOL, CI, and entry. HTS verified against the client's product master, not the prior shipment. Below 0.90 confidence, the packet holds for filer review.
Clean packets posted to CargoWise via the shipment API. Source documents attached to the consol file. Exceptions routed to a named filer queue with the flag in plain English.
After, the numbers the desk signs off
Same desk, same filers, same carriers, same broker relationships. The pipeline compressed the packet and fed the filer a clean draft. What changed was the cycle, not the team.
Filers still own the exception queue. They still sign off every packet with a borderline HTS confidence. The difference is that on a clean day, the pipeline posts the draft into CargoWise while the filer is still on the first coffee. On a bad day, the filer sees one flagged field instead of three tabs.
From the desk
We held our process. The pipeline just made the desk faster on the hard days, and the hard days are the ones that used to cost us.
Customs operations leadGlobal forwarder, Houston hub
Handover
The engagement ends at a clean handover. The ops team runs the pipeline; Hexaa stays on call for a fixed retention period, then steps back.
Related cases
Each links to a named client, a named document, and the system the clean data lands in. We publish only what the client signed off to publish.
POD capture and billing close-out for 200 drivers across three terminals. Friday cycle time moved from two days to same-day.
→Logistics · 2025Customs broker · HS classification against product masterHS proposals against a 42,000-line product master, flagged under threshold for filer review. Built on LlamaParse, integrated with the broker's in-house entry system.
→Construction · 2025Commercial GC · submittal queue against the primeSubmittal log reconciliation against the prime contract spec. 1,200 submittals per project, Procore as the system of record.
→Free 30-minute call
You'll leave with a clear next step.
The BOL says one weight, the commercial invoice says another, the entry packet carries a third. The pipeline compares all three at the field level, flags the mismatch, and holds the packet until the exception clears. The filer sees one discrepancy, not three tabs.
