Freight settlement doesn't fail because the decisions are hard. It fails because the decisions are simple and there are too many of them, spread across systems that were never built to be automated.
The company books thousands of freight orders every month. Settling each one means logging into the vendor's transportation management system, retrieving the load details, comparing them against internal records, flagging discrepancies, and accepting or rejecting the order. Across more than 50 vendors, each with its own portal, its own data format, and its own workflow quirks, and with no APIs exposed by any of them, this had been a purely manual, one-analyst-per-portal operation. At more than 50,000 cases a month, the team's capacity was permanently stretched, and order backlogs kept accumulating faster than they could be cleared.
Order intake. The agent picks up each incoming load order from the queue, reads it in full, identifies the correct vendor, and retrieves that vendor's standard operating procedure before doing anything else.
Portal execution. It logs into the vendor's portal through browser automation and navigates the interface exactly as a human analyst would, the same clicks, the same screens, following the vendor's SOP step by step, with no manual intervention once it starts.
Audit trail. Every portal interaction is recorded as a full session video: every click, every screen, timestamped. That gives the operations team a complete, reviewable record of any case, at any time, without anyone having to reconstruct what happened after the fact.
Data sync. Load data extracted from the portal is pushed directly into internal systems. There's no separate data-entry step and no manual reconciliation pass required afterward.
Exception handling. Threshold mismatches, missing information, and deviations from a vendor's SOP get escalated with full context and the session recording already attached, so the operations team only steps in on cases that genuinely need a human judgment call.
The deployment now runs at more than 95% browser-agent accuracy across the more than 50 vendor systems it touches, unifying all of them without a single API integration. Turnaround time on settlements dropped 40%, and the team reclaimed more than 20,000 human hours a year that had previously gone into manual portal work.
The interesting part of this deployment isn't freight-specific. It's that the hard problem, operating dozens of unrelated systems that expose no integration surface, with a full audit trail on every action, is exactly the same problem Zamp originally solved inside banking, where sanctions-screening portals and card-network dispute systems present the identical constraint: no API, high stakes, and a documented trail required on every touch. The freight deployment runs on the same browser-automation and audit-logging architecture, pointed at a different set of portals.
For more on how that architecture generalizes across industries, see why Zamp isn't just a finance AI company and how Zamp's AI employees work.
