Load tenders and rate confirmations: what freight history teaches AI

What does a load record contain?#

A load record contains the shipper's request, the brokerage's commitment to a carrier and everything that happened between pickup and payment. In most brokerages those pieces sit in different places: tenders arrive by EDI 204 or email, rate confirmations are generated as PDFs from the TMS, and status updates come from check calls, ELD integrations or tracking apps.

Brokerages that receive tenders by EDI usually hold cleaner, structured history than those working from emailed tenders, but emailed tenders often carry the special instructions and back-and-forth that EDI drops. A complete picture usually needs both. The fields that matter fall into a few groups.

• 1Tender fields: origin, destination, stops, pickup and delivery windows, equipment, weight, commodity, special requirements and the shipper's reference numbers.
• 2Tender response: accepted or rejected, and for contract freight, which routing guide position the load came from.
• 3Rate confirmation fields: carrier, linehaul, fuel, accessorials, payment terms and any conditions such as tracking requirements.
• 4Revisions: added stops, rescheduled appointments, detention or layover approvals and carrier changes, each with a time.
• 5Status events: dispatched, at pickup, loaded, in transit, at delivery and delivered, plus delay reasons.
• 6Settlement: carrier invoice, proof of delivery, short-pays and disputes.

Load record fields to AI task#

Each group of load record fields supports a different AI task, and each needs different handling before it leaves the brokerage. The table maps field groups to the task and the usual treatment.

What pricing models learn from tender and rate pairs#

Pricing models learn from tender and rate pairs how a market clears on a given lane at a given moment. The tender shows what the shipper needed and when; the rate confirmation shows what it took to get a carrier to commit. Across many loads, that pairing teaches seasonality, the cost of short lead times and how accessorial risk is priced.

Rejected tenders matter as well. When contract carriers reject freight and loads move to the spot market, the rejection and the eventual coverage rate can show capacity tightening on your lanes before broader market reports reflect it. A history that records only covered loads hides that signal.

The spread between the shipper's rate and the carrier's rate is the most sensitive field in the record. Brokerages usually decide in advance whether margins are removed, expressed only in relative terms or excluded from the package entirely.

What carrier matching and tracking models learn#

Carrier matching models learn which kinds of carriers accept which kinds of freight, and how they perform once they do. A tokenized carrier that hauls the same reefer lane repeatedly, delivers on time and rarely triggers detention claims is exactly the pattern a matching model needs, and the token keeps that pattern intact without naming the carrier.

Tracking models learn from the gap between promised and actual. Pickup and delivery windows from the tender, set against status events and delay reasons, teach ETA prediction and early warning. Delay reasons recorded with a code and a short note are far more useful than a status that simply reads late.

How shipper and carrier names come out#

Shipper and carrier names come out through consistent replacement rather than deletion. Each shipper, consignee, facility and carrier gets a token that stays the same across every record, so a model can still see that one shipper tenders weekly on a lane or that one carrier keeps winning certain freight.

Standard de-identification tooling supports this approach. Google's Sensitive Data Protection API, for example, includes deterministic encryption and date shifting among its de-identification transforms, which lets the same input map to the same token and dates move consistently. Tooling is only part of the job; the harder work is finding where identities hide.

• 1Structured fields: shipper, consignee, bill-to, carrier, dispatcher and driver names.
• 2Authority and account numbers on rate confirmations and carrier packets.
• 3Facility addresses and dock instructions that point to a single site.
• 4Commodity descriptions that name a branded product.
• 5Free-text notes, email bodies and PDF images, including logos and signatures.

Illustrative: a flatbed brokerage maps its rate confirmation archive#

Illustrative: a fictional flatbed and specialized brokerage keeps loads in its TMS, receives most tenders by email and stores every rate confirmation as a PDF attached to the load. Revisions are issued as new PDFs with a suffix on the load number.

Mapping the archive shows that structured TMS fields cover lanes, rates and carriers well, while tarping, permits and escort requirements appear only in the PDFs and email. Those requirements are what make flatbed pricing hard, so the team decides they belong in scope.

The decision is to extract tender requirements and revisions from the PDFs, tokenize shippers, carriers and facilities, generalize jobsite addresses to region and remove driver names and phone numbers from notes. Margins are excluded. The CEO approves a sample for review before any wider preparation.

The review also surfaces a gap: some changes were agreed by phone and never re-confirmed in writing. The brokerage changes its process so every change produces a revised rate confirmation, which helps with carrier pay disputes now and makes future history more complete.

How SourceX handles tender and rate records#

For tender and rate records, SourceX's main job is agreeing preparation rules with the brokerage before any sample exists: which identities become tokens, how facility locations are generalized and how margins are treated. Those rules sit inside the SourceX five-step transaction of Supply, Rights, Preparation, Approval and Delivery, and the brokerage approves the prepared sample before anything moves.

The approved package travels with a SourceX Evidence Packet recording provenance, licensing rights, permitted use, the privacy record and release authorization. In the SourceX Enterprise Data Value Framework, revisions, rejections and delay reasons score on human-generated signal, and recency matters because freight markets shift.