Undisclosed Global Supply Chain Management Company

Freight transportation accounts for a substantial share of global greenhouse gas emissions, and for enterprises managing multi-modal supply chains across road, rail, sea, and air, accurate emissions measurement is foundational to both compliance and cost control. This company's operations spanned numerous regions and regulatory environments, yet emissions data remained fragmented across unstructured sources — invoices, receipts, and transport records — with no consistent method to aggregate it at scale. Manual preprocessing was impractical given the volume and variety of documents involved. Without reliable, structured emissions visibility, identifying high-impact reduction opportunities or demonstrating regulatory compliance required significant manual effort and carried meaningful operational risk.

Omdena designed an end-to-end AI pipeline built on large language models and generative AI to transform fragmented logistics documents into auditable emissions intelligence. A vision-language OCR stage handles real-world document quality — including low-resolution scans and mobile captures — feeding into a deterministic extraction layer that normalizes activity data against strict schemas. Emissions are then calculated using published GHG Protocol emission factors, preserving full traceability of inputs, scope assignments, and formulas. A retrieval-augmented generation (RAG) layer retrieves relevant regulatory texts from a vector database to produce citation-backed compliance evaluations. All stages operate as specialized agents within a modular, state-driven orchestration workflow, enabling controlled execution and reproducibility across regions and document types. Outputs surface through a user-facing dashboard designed for non-technical stakeholders.

The system delivered measurable outcomes across both environmental and financial dimensions without disrupting operational performance:

• 10% reduction in carbon emissions across the global supply chain
• ~$5M in annual cost savings, driven largely by routing and planning optimizations that cut fuel consumption alongside emissions
• High adoption rates attributed to transparent, auditable calculations that non-technical users could verify independently
• Operationalized recommendations integrated into day-to-day logistics workflows rather than remaining advisory

The combination of deterministic calculations and explainable outputs proved critical: stakeholders trusted the system's conclusions, which directly accelerated adoption across distributed regional teams.

• Deterministic, GHG-Protocol-aligned calculations outperform opaque ML models for emissions programs where stakeholder trust and regulatory auditability are requirements.
• Unstructured document ingestion is the hidden bottleneck: automating OCR and schema-validated extraction eliminates the manual preprocessing step that blocks most emissions initiatives from reaching scale.
• Environmental and financial outcomes reinforce each other — routing and modal optimizations that reduce emissions simultaneously lower fuel and logistics costs, creating a dual business case.
• Usability determines adoption: building for non-technical users from the start — intuitive dashboards, traceable outputs — converts analytical accuracy into sustained operational change.