Infrastructure for Carrier Performance Prediction & Matching

Carrier performance prediction requires documented, current, and findable definitions of what constitutes acceptable carrier performance — on-time thresholds, claim rate limits, acceptance rate floors by lane. Without L3 formality, the AI has no business rules to validate predictions against. The freight baseline confirms that carrier relationship knowledge stays tribal: 'just ask Sarah about that account.' For automated carrier matching, these rules must be in the wiki, not Sarah's head.

Predicting carrier performance requires systematic capture of historical on-time rates, claim events, and acceptance/rejection outcomes tied to lane, load type, and time period. The TMS and EDI baseline provides transactional capture, but the model also needs structured capture of why carriers were chosen or skipped — context that currently lives in phone calls. Template-driven capture of tender outcomes and carrier selection rationale is needed to train reliable prediction models.

Carrier matching requires consistent schema across all records: carrier ID, lane (origin-destination pair), equipment type, on-time percentage, acceptance rate, claim rate, and timestamp. The TMS provides structured transactional fields, which is the minimum needed for ML features. L3 consistent schema ensures every carrier performance record has the same fields populated, enabling lane-level aggregation and ranking without manual normalization before each model run.

The carrier prediction model must query TMS for historical assignments, GPS telematics for current carrier location and availability, and market rate feeds for spot pricing context — all at tendering time. API access to these core systems allows the AI to assemble the carrier profile needed to rank candidates. The freight baseline confirms legacy TMS limits access, but API connections to the primary operational systems are achievable and necessary for automated tendering decisions.

Carrier performance profiles must update when events occur: a carrier's on-time rate degrades after a service failure, acceptance rates shift when spot markets tighten, or a carrier exits a lane. Event-triggered maintenance ensures the prediction model reflects current carrier behavior rather than quarterly snapshots. The baseline confirms scorecards are updated manually on a quarterly basis — for automated matching this latency causes the model to recommend carriers whose recent performance has materially changed.

Carrier performance prediction draws from TMS (load history, carrier assignments), GPS telematics (current location and availability), and market rate feeds (spot pricing context). These systems must exchange data via API-based connections so the AI can assemble a complete carrier profile at tender time. The freight baseline confirms these systems currently operate as islands with point-to-point EDI for transactions. L3 API connections across the core systems are the minimum for automated carrier matching to function without manual data assembly.