How AI Reduced Freight Rate Processing Costs by 94%
AI DOCUMENT INTELLIGENCE
FREIGHT AUTOMATION
ERP INTEGRATION
~94%
lower cost vs. manual processing
$0.08
average AI cost per rate sheet
≥85%
charge codes auto-accepted
≥90%
field-level extraction accuracy
<20 sec
end-to-end processing time
20+
carrier formats supported
THE BUSINESS OUTCOME
Freight pricing arrives in every format - and every carrier speaks a different language
Freight forwarders receive continuous rate updates from carriers and NVOCCs. The information can arrive in email bodies, multi-sheet spreadsheets, PDFs, scanned documents or screenshots – often within the same email thread.
The real challenge is not simply reading the document. The organization must interpret rate structures, recognize carrier-specific terminology, understand inclusions and exclusions, standardize charge codes, and enter the correct data into a freight management or ERP platform.
01 FORMAT HETEROGENEITY
No two carriers follow the same schema. One update may be an Excel workbook; another may be a PDF or a rate table embedded directly in an email.
02 CHARGE CODE VARIANCE
BAF, FAF, Bunker Adjustment Factor, FSC and BUC may all describe the same underlying surcharge. Static lookup tables are fragile.
03 CONTEXTUAL AMBIGUITY
Phrases such as “EBS Inclusive” change the commercial meaning. A naive extractor can create systematic quoting errors.
04 SCALE BOTTLENECK
Manual ingestion scales linearly with carrier count and shipment volume, forcing operations teams to grow simply to maintain the same turnaround time.
What the solution needed to achieve
- Process heterogeneous carrier communications without building a new template for every carrier.
- Extract structured freight rates while preserving source context and supporting multi-currency information.
- Normalize vendor-specific surcharge terminology into a master charge-code taxonomy.
- Prevent uncertain or hallucinated values from silently reaching the ERP.
- Integrate with existing enterprise applications rather than replacing the client’s current UI or rate platform.
DESIGN PRINCIPLE
Adaptive Tariff Intelligence Pipeline
Eservecloud designed a production-grade, headless REST API that converts unstructured carrier communications into standardized freight-rate data ready for downstream systems. The platform is vendor-agnostic: it evaluates the structure and meaning of each document rather than relying on carrier-specific hard-coded templates.
“Any carrier. Any document format. One normalized output for your ERP.”
Six-stage intelligent processing pipeline
| Stage | Capability | What happens |
|---|---|---|
| 01 | MIME Decompose | Recursively parses the raw email and separates body text, HTML, Excel files, PDFs, images and forwarded messages. Email metadata is retained for context. |
| 02 | Smart Artifact Filter | Determines which artifacts actually contain rate information. Structured attachments take priority where appropriate; body and attachment results can be merged. |
| 03 | Format Parse | Converts each artifact into clean structured content. Excel, PDF and email formats follow specialized parsing paths, with vision fallback for scanned documents. |
| 04 | AI Extraction | Uses a forced structured-output schema. Each extracted rate includes evidence from the source, and ambiguous or missing values are flagged instead of guessed. |
| 05 | Source Fusion | Combines results from multiple artifacts, resolves conflicts using source precedence, and merges lanes by origin/destination context. |
| 06 | Canonicalise & Normalise | Maps carrier terminology to the master charge list, normalizes ports, equipment and dates, and applies confidence-based automation rules. |
Built for real freight operations - not just document extraction
Carrier-agnostic by design
The pipeline processes documents based on structure and semantics. Adding a new carrier does not require a new code path or brittle template.
Evidence-grounded AI
Every extracted rate carries source evidence. A verification step checks that the supporting text can be traced back to the source document before the value is trusted.
Self-improving canonicalisation
Confirmed mappings are written back to the alias store. The same terminology can then bypass expensive AI disambiguation on future runs.
Confidence-driven automation
High-confidence charge mappings can flow directly to ERP. Lower-confidence cases surface a pre-filled suggestion for human review.
Multi-source reasoning
The system can combine an email body with one or more attachments and apply source precedence when the same lane or charge appears in multiple places.
Vision fallback for scans
When a PDF contains no extractable text, the pipeline switches to a vision path, extending coverage to image-only rate sheets.
The feedback loop creates compounding value
During early usage, a portion of charge codes may require semantic disambiguation. As users confirm mappings, those decisions become reusable aliases. Over time, fewer charges require AI reasoning, which lowers marginal cost while increasing automation coverage.
New charge appears
AI evaluates meaning
Confidence scored
User confirms if needed
Alias is stored
Future match is instant
Cost decreases
Automation increases
Automation that delivers measurable operational impact
$0.08
all-in AI cost per rate sheet
~94%
cheaper than manual processing
≥85%
charge codes auto-accepted
≥90%
field-level extraction accuracy
<20 sec
synchronous end-to-end latency
~$38–54
monthly API cost at 500 shipments
What these numbers mean for the business
| Operational Benefit | How it is achieved | Business effect |
|---|---|---|
| Lower processing cost | Selective AI usage, prompt caching, model routing and reusable aliases | More volume without proportional operations headcount growth |
| Faster turnaround | Automated ingestion, parsing, extraction and normalization | Carrier rate updates can reach pricing systems in seconds rather than manual cycles |
| Fewer pricing errors | Source evidence, context-aware inclusions and confidence rules | Reduced risk of duplicate surcharges or incorrect charge-code mapping |
| Scalable carrier onboarding | Structure-driven processing instead of carrier-specific templates | New carrier formats can be handled with significantly less engineering effort |
| Controlled automation | High-confidence items auto-load; uncertain items are surfaced | Operations retains oversight without reviewing every transaction |
WHY THE ECONOMICS IMPROVE OVER TIME
Validated against real-world freight scenarios
| THC splitting | Terminal Handling Charges were correctly distinguished as origin-side LTHC or destination-side DTHC using contextual source cues. |
| Inclusive charge detection | “EBS Inclusive” was interpreted as an all-in component rather than an additional surcharge, preventing systematic over-quoting. |
| DG classification granularity | Dangerous-goods PSA Group 2 and Group 3 charges were distinguished from generic DG surcharges and mapped at the most granular master-code level. |
| Multi-currency extraction | A carrier communication containing 17 global charges across SGD and USD was extracted and canonicalised with individual confidence scores. |
| Unknown charge surfacing | Carrier-specific charges missing from the master list were flagged as unknown instead of being silently forced into the wrong category. |
| Scanned document handling | Image-only PDFs were routed to a vision-based extraction path when standard text extraction was not possible. |
The key distinction: the system does not merely extract text. It interprets freight context, validates evidence and prepares data for operational use.
Designed to fit into the existing enterprise landscape
Carrier Email / Rate Sheet
Document Ingestion
AI Extraction
Evidence Validation
ERP-ready JSON
Webhook / ERP
Core API capabilities
| Capability | Method / Path | Purpose |
|---|---|---|
| Ingest document | POST /api/v1/ingest | Accept .eml, .pdf or .xlsx and return a job identifier for asynchronous processing. |
| Poll job status | GET /api/v1/jobs/{job_id} | Expose stage-by-stage progress from queued through completed or failed. |
| Retrieve results | GET /api/v1/jobs/{job_id}/results | Return structured lanes, rates, charges, master codes, confidence and review flags. |
| Webhook callback | Configured per client | Push the completed JSON result to the client’s registered endpoint. |
| Health check | GET /api/health | Expose service, version and dependency health for platform probes. |
A production-grade technology stack
| Layer | Representative Components | Role |
|---|---|---|
| API | FastAPI · Uvicorn · Pydantic v2 | REST services, typed contracts, async job orchestration and OpenAPI documentation. |
| Ingestion | mail-parser · pdfplumber · PyMuPDF · openpyxl | Email, PDF, Excel and image processing across heterogeneous carrier documents. |
| AI Extraction | Claude family · forced tool-use schema | Structured extraction with source evidence and deterministic output controls. |
| Canonicalisation | Semantic reasoning · alias repository | Charge-code mapping, confidence scoring and self-improving reuse of confirmed decisions. |
| Infrastructure | Docker · PostgreSQL · Celery · Redis | Containerized deployment, persistent state, scalable worker queues and reliable asynchronous processing. |
From manual rate entry to intelligent freight operations
Freight rate processing has traditionally required operations teams to read, interpret, reconcile and re-enter pricing information across multiple systems. The Adaptive Tariff Intelligence Pipeline converts that manual workflow into an intelligent automation layer that is faster, more consistent, less expensive and easier to scale.
Ready to automate document-heavy freight operations?
Eservecloud Solutions builds production-grade AI automation for logistics and enterprise operations – including document intelligence, ERP integration and agentic workflows that work with your existing systems.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.