Freight document automation that shows you where every number came from.
Laneproof extracts structured data from freight documents: rate confirmations, carrier invoices, bills of lading, proofs of delivery, and lumper receipts. Two AI models read each document independently, and every extracted value links back to the exact spot on the page it came from, so you can verify a number in a second instead of reopening the PDF and scanning for it.
This is not generic OCR. Freight documents are financial documents, so the goal is not just reading the page but knowing which readings to trust and flagging the ones worth a second look.
Documents it reads
Rate confirmations
Load number, agreed linehaul, fuel surcharge basis, free time, authorized accessorials, pickup and delivery windows, and the carrier's MC number.
Carrier invoices
Every billed line item with its amount, broken out so linehaul, FSC, detention, lumper, and other accessorials can each be checked separately.
Bills of lading
Shipper and consignee, piece and weight counts, and the notations that decide whether a later claim holds up.
Proofs of delivery
Delivery date and time, signature presence, and any exception noted at delivery, which is what detention and claims arguments turn on.
Lumper receipts
Amount, facility, and date, so the receipt can be matched against the lumper line on the invoice rather than taken on faith.
Batch and email intake
Queue a stack of documents at once, or forward the carrier's email and let the attachments be picked up and processed on arrival.
How it works
- 1
Get the document in
Drag and drop a file, queue a batch, forward an email to your Laneproof address, or post it through the API on the Solo plan and above.
- 2
Two models read it independently
Anthropic Claude and Google Gemini each extract the document without seeing the other's answer. Running two engines is the point: agreement is a real signal, and disagreement is exactly where a human should look.
- 3
The merge engine classifies every field
Each value comes back marked agreed (both models matched), single (only one model found it), or disputed (they returned different values). Nothing is silently averaged away.
- 4
Verify against the source
Extracted values are mapped back to their bounding boxes on the original document, so checking a suspicious number means glancing at the highlight, not re-reading the page.
Manual review vs. your TMS vs. Laneproof
| Capability | Manual review | Typical TMS | Laneproof |
|---|---|---|---|
| Reading accuracy signal | No signal, just fatigue | Single OCR pass, confidence rarely shown | Two engines; agreed / single / disputed per field |
| Freight-specific fields | Operator knows what matters | Generic document storage | Rate con, BOL, POD, lumper, and invoice schemas |
| Verify a number quickly | Re-open and scan the PDF | Document stored, values not linked | Value highlights its location on the page |
| Volume handling | Linear in headcount | Manual entry still required | Batch queue and email intake |
| Feeds reconciliation | Separate manual step | Usually not included | Extracted fields flow straight into invoice matching |
What this does not do
Worth being direct about the boundaries so you can tell whether this fits before you spend a minute on it.
- —It is not a TMS, a document repository, or a replacement for either. It reads documents and hands back structured data.
- —Handwriting on a POD is still hard. Where the models disagree on a scrawled signature or a hand-written time, you will see a disputed field rather than a confident guess.
- —It does not file claims or send disputes. It produces the extracted evidence; the outbound communication stays with your team.
- —Very poor scans degrade accuracy the same way they degrade human reading. The difference is that the disagreement gets surfaced instead of buried.
Frequently asked questions
How is this different from regular OCR?
OCR converts pixels to characters. It does not know that a number near the word 'detention' is a detention charge, and it gives you no way to tell a confident reading from a shaky one. Laneproof extracts into freight-specific schemas and runs two independent models so disagreements are visible rather than hidden behind a single confidence score.
Why use two AI models instead of one?
Because a single model that is confidently wrong looks identical to a single model that is right. Running Claude and Gemini independently and comparing their output turns that invisible failure into a flagged, disputed field that a person can check in a few seconds.
Which freight documents are supported?
Rate confirmations, carrier invoices, bills of lading, proofs of delivery, and lumper receipts. These are the five documents that decide almost every billing dispute between a broker and a carrier.
Can I send documents programmatically?
Yes. API keys are available on the Solo plan and above, so documents can be posted from your TMS, Zapier, Make, or a custom integration. Webhook notifications fire when processing completes.
What does it cost to try?
The free tier covers 20 documents per month with dual-AI extraction and no card required. Solo is $149 per month for 400 documents; Pilot is $499 per month for 2,000.
Find out what you have been overpaying.
Run your last 20 documents through it on the free tier. No card, no integration, no contract. If nothing comes back, you have lost a few minutes.