Case studies/Logistics and freight
A Chicago-based mid-market broker running 6,200 rate confirmations a month, with four minutes of copy-paste per confirmation into MercuryGate. We built a pipeline that parses the confirmation, matches the lane, rate, and accessorials to the booked load, and posts the record without the keystroke round.
| Load | Carrier | Lane | Booked | Rate con | Status |
|---|---|---|---|---|---|
| MG-73812 | Knight-Swift | Chicago, IL → Dallas, TX | $2,450 | $2,450 | posted |
| MG-73815 | Werner Enterprises | Joliet, IL → Atlanta, GA | $2,180 | $2,180 | posted |
| MG-73819 | Schneider National | Gary, IN → Memphis, TN | $1,820 | $1,895 | variance |
| MG-73821 | Heartland Express | Rockford, IL → Kansas City, MO | $1,640 | $1,640 | posted |
| MG-73824 | Crete Carrier | Chicago, IL → Denver, CO | $3,180 | $3,180 | posted |
| MG-73827 | Marten Transport | Aurora, IL → Nashville, TN | $2,020 | pending | awaiting countersign |
| MG-73829 | USA Truck | Bolingbrook, IL → Houston, TX | $2,640 | $2,640 | posted |
| MG-73832 | Covenant Logistics | Chicago, IL → Charlotte, NC | $2,310 | pending | awaiting countersign |
| MG-73835 | Prime Inc. | Joliet, IL → Phoenix, AZ | $3,420 | $3,420 | posted |
At a glance
A broker with a TMS that worked and a keying habit that did not scale. The pipeline had to fit into MercuryGate exactly as the desk already used it.
The engagement
The stack
ISO 27001 · ISO 9001 · DPA and NDA signed at kickoff.
Before, the broker desk
Brokers were keying the same data twice: once into the sales board, once into MercuryGate. The keying was not the problem. The error pattern that followed was.
Each rate confirmation carried seven fields the broker had to re-enter into MercuryGate: origin, destination, pickup window, delivery window, line-haul rate, fuel, accessorials. On a clean confirmation, four minutes. On a confirmation with a carrier-specific layout, longer.
Pre-build baseline: 4 minutes per confirmation, measured across a two-week sample on the Chicago desk.
Detention, layover, lumper, and stop-off charges were keyed from memory when the confirmation layout buried them. When the carrier billed back, the broker reconciled. The reconciliation queue ran two days behind.
Pre-build baseline: approximately 4% of loads required a carrier-side rate correction on billing.
When a rate did not match, the broker had three places to look: the portal confirmation, the MercuryGate load record, and the email thread with the carrier rep. Three tabs, two logins, one carrier on hold.
Pre-build baseline: mean time to clear a rate exception was 18 minutes, sampled across January 2025.
What we built
The pipeline follows the same five stages we run on every logistics engagement. The details below are the ones we wrote for this brokerage, against the carrier templates they actually receive.
Shipper portals polled on a 5-minute cadence. Carrier email routed via secure mailbox. Sales-board confirmations published as events. All normalised to a single confirmation ID.
Each confirmation tagged by carrier template. 40 named templates across the carrier base. Unknown layouts route to LlamaParse's generic extractor with tighter confidence gates.
Origin, destination, pickup window, delivery window, line-haul rate, fuel, accessorials. LlamaParse primary, Azure AI Document Intelligence fallback for layouts outside the template set.
Lane and rate cross-checked against the booked load in MercuryGate. Accessorial codes mapped to the broker's canonical set. Below 0.92 confidence, the confirmation holds for broker review.
Matched confirmations posted to MercuryGate via the load API with the source PDF attached. Exceptions surface in a named broker queue with the discrepancy in plain English, not three tabs.
After, the numbers the broker desk signs off
Same brokers, same carriers, same TMS. The pipeline read the confirmation, matched the booked load, and posted the record. What changed was where the broker's attention went.
Brokers still own every rate exception. They still negotiate, still confirm, still write the email. The difference is that a clean confirmation takes 20 seconds to post, not four minutes, and the exception queue shows one flagged field instead of three tabs. The ops team runs the pipeline today, with Hexaa on call for a fixed retention period.
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.
16,000 PODs a month captured from dock scans, driver photos, and faxes. Billing close-out moved from 48 hours to same-day.
→Logistics · 2025Global forwarder · port-side customs packets28,000 customs packets a month reconciled inside a four-hour clearance window. LlamaParse on intake, CargoWise as the system of record.
→Distribution · 2026B2B distributor · Prophet 21 PO flowFax and email POs routed into Prophet 21 without the copy-paste round. Sales desk keeps the customer call, the ERP carries the record.
→Free 30-minute call
You'll leave with a clear next step.
The rate confirmation lists a line-haul, a fuel, and a set of accessorials. The booked load in MercuryGate lists the same three fields. The pipeline compares them, flags any delta, and holds the confirmation until the broker clears the exception.
