Logistics Outsourcing — AI / RAG

AI Document Processing for a Freight & Logistics Operator

A freight brokerage was drowning in paperwork — bills of lading, proof-of-delivery scans and rate confirmations, all keyed into the TMS by hand. Dillo built an LLM and OCR extraction pipeline with human-in-the-loop review, plus a RAG assistant over the company's operations documentation. The system paid for itself in under six months.

85%

of documents auto-processed without human touch

12min→45sec

intake processing per document

<6 mo

to full payback on the project

The Client

A mid-size freight brokerage and logistics operator coordinating thousands of truckload and LTL shipments per month across the US. Every shipment generated a trail of documents — bills of lading (BOLs), proof-of-delivery scans (PODs), rate confirmations, lumper receipts, accessorial paperwork — arriving as faxes, phone photos and emailed PDFs of wildly varying quality.

The Challenge

A back-office team spent its days reading documents and keying fields into the TMS: shipper, consignee, reference numbers, weights, charges, delivery exceptions. Average intake time was about 12 minutes per document, error rates crept up at month-end volume spikes, and carrier payments stalled whenever POD processing fell behind — which strained carrier relationships and tied up working capital.

Off-the-shelf template-based OCR tools had been trialed and abandoned: freight paperwork has no standard layout, and every new shipper or carrier broke the templates. Meanwhile, institutional knowledge — how to handle a detention claim, which customers require which accessorial codes — lived in a sprawl of SOP documents and veterans' heads.

What We Did

Dillo delivered the project through its AI/RAG outsourcing practice — a compact team of an ML engineer, two back-end engineers and a delivery lead, working in two-week increments with the client's ops managers reviewing accuracy at every step.

  • Built a document intake pipeline combining OCR with LLM-based extraction: documents are classified by type (BOL, POD, rate confirmation, receipt), then structured fields are extracted with confidence scores per field — no layout templates to maintain.
  • Implemented human-in-the-loop review: low-confidence extractions route to a side-by-side review UI where staff correct fields in seconds; corrections feed back into evaluation sets that track accuracy over time.
  • Integrated directly with the client's TMS via API, so clean extractions post automatically — triggering carrier-payment and invoicing workflows without rekeying.
  • Added validation rules that cross-check extracted values against shipment records (rate vs. agreed rate, weights vs. tender) and flag genuine discrepancies rather than formatting noise.
  • Built a RAG assistant over operations documentation — SOPs, customer playbooks, accessorial guides — so ops staff ask questions in Slack and get sourced answers with document citations instead of hunting through folders.
  • Established accuracy dashboards and an evaluation harness, giving the client measurable per-field precision before each rollout stage expanded automation.

The Results

  • 85% of documents now flow through with no human touch; the rest arrive pre-filled for a quick review rather than manual entry.
  • Average intake processing fell from about 12 minutes to roughly 45 seconds per document.
  • POD-to-payment cycles shortened noticeably, improving carrier satisfaction scores and easing working-capital pressure.
  • The back-office team was redeployed from data entry to exception handling and carrier relations — no layoffs, more capacity.
  • Based on labor savings and faster billing, the client calculated full payback on the project in under six months.
  • The RAG assistant cut new-hire ramp time in ops, becoming the default first stop for "how do we handle X" questions.

Figures reflect outcomes reported for this engagement; they are project results, not audited benchmarks.

Tech Stack

Python LLM APIs OCR RAG Vector search PostgreSQL React AWS TMS API integration

Services Used

Software Outsourcing — AI, LLM & RAG delivery: extraction pipeline, human-in-the-loop tooling and the ops knowledge assistant.

Ready to build your team?

Tell us what you need and get your first highly relevant candidate CVs within 24–48 hours.