How AI-Driven Supply Chain Optimisation Reduced Carbon Emissions by 10% and Saved $5M

Introduction

International supply chains are under growing pressure to operate more efficiently while responding to rising expectations around environmental accountability. As global trade scales, the movement of goods across ships, trucks, rail, and air networks contributes significantly to overall carbon emissions, making logistics one of the most emissions-intensive components of modern business operations. For organizations managing complex, multi-country supply chains, understanding where emissions originate—and how to reduce them without disrupting service levels—has become a critical challenge.

This case study examines how a global supply chain management company partnered with Omdena to address this challenge through an AI-driven optimization initiative. By applying advanced data analysis and machine-learning techniques to large volumes of operational data, the solution enabled more informed routing, transport, and planning decisions. The result was a measurable 10% reduction in carbon emissions and approximately $5 million in annual cost savings, demonstrating that data-driven sustainability initiatives can deliver both environmental impact and tangible business value

Background: Carbon Footprint in Supply Chains

Freight transportation is a major contributor to supply-chain emissions, driven by the scale and intensity of global trade operations.

Freight transportation and logistics represent a substantial share of global greenhouse gas emissions, driven by the continuous movement of goods across road, rail, sea, and air networks. When warehousing, ports, and energy consumption across logistics operations are included, supply chains emerge as one of the most carbon-intensive components of modern commerce. Despite this impact, accurately measuring emissions across complex, multi-modal supply chains remains difficult, as data is often fragmented across documents, systems, and regions.

For organizations operating at scale, the challenge is not only environmental but operational. Emissions data is frequently incomplete, inconsistent, or embedded in unstructured sources such as invoices, receipts, and transport documents. At the same time, logistics teams operate under strict cost and time constraints, leaving little room for manual analysis or disruptive interventions. As highlighted in the project report, the lack of reliable, structured emissions data makes it difficult to identify high-impact reduction opportunities or evaluate compliance requirements without introducing operational risk

Project Goal: Reducing Emissions at Scale

The organization at the center of this project operates across a highly distributed global supply chain, spanning numerous regions, transportation modes, and regulatory environments. Its primary objective was to reduce carbon emissions at scale without compromising operational reliability or service levels. Achieving this required a practical, data-driven approach capable of translating fragmented operational data into actionable insights.

To address this challenge, the project was structured around the following core goals:

• Establish reliable emissions visibility at scale: Create a consistent method to calculate and aggregate carbon emissions across diverse logistics activities using real operational data.

• Automate data extraction from unstructured sources: Enable emissions analysis directly from invoices, receipts, and transport documents without manual preprocessing.

• Support informed decision-making without operational disruption: Identify meaningful emissions reduction opportunities while preserving delivery timelines and service quality.

• Enable transparency and trust in AI-driven outputs: Ensure that calculations and recommendations were explainable, auditable, and usable by non-technical stakeholders.

• Demonstrate measurable environmental and financial impact: Validate that emissions reduction initiatives could deliver tangible cost savings alongside sustainability gains

The goal was to deploy an AI-driven system that could operate within real-world supply chain constraints and deliver measurable emissions reduction alongside clear business value.

Challenges and Considerations

Dataset creation

A key challenge was assembling a dataset suitable for emissions analysis at scale. Much of the required information was embedded in unstructured documents such as invoices, receipts, and utility bills, often with missing or inconsistent metadata. The system therefore needed to process large volumes of real-world data while preserving enough context to support accurate emissions calculations across different regions and transport modes

Building trust

For teams operating under tight operational and regulatory pressure, trust in automated outputs was essential. Transparency and interpretability were built into the system so users could understand how emissions were calculated and how recommendations were generated. Deterministic calculations and traceable emission factors helped ensure results could be reviewed, verified, and confidently applied in practice

Ease of use

With users distributed across regions and roles, ease of adoption was a critical consideration. The solution was designed to integrate smoothly into existing workflows, minimizing manual effort and technical complexity. By prioritizing intuitive interfaces and automation, the system enabled teams to focus on decision-making rather than data preparation, reinforcing the importance of usability alongside model performance

AI‑Powered Approach

High-level, left-to-right view of the multi-agent pipeline used to process documents, calculate emissions, and generate compliance insights.

To address the challenges of emissions measurement and decision-making at scale, the company partnered with Omdena to design an AI-driven system built for real-world supply chain operations. Rather than relying on isolated predictive models, the solution was developed as an end-to-end, document-driven pipeline capable of transforming fragmented operational data into reliable, auditable emissions and compliance insights.

The system ingests unstructured documents such as invoices, receipts, and utility bills and converts them into structured activity data suitable for emissions analysis. Emissions are then calculated using deterministic logic aligned with established GHG Protocol methodologies, ensuring consistency and reproducibility. A retrieval-augmented generation (RAG) layer connects these calculations to relevant regulatory frameworks, enabling transparent, citation-backed compliance assessments. All components are coordinated through a modular, multi-agent workflow, allowing the system to scale across regions and document types while maintaining traceability and operational reliability

Core system components

1. Document ingestion and OCR

The pipeline processes real-world documents using a vision-language OCR approach designed to handle low-quality scans and mobile-captured images with high reliability.

2. Structured extraction and normalization

Extracted text is converted into standardized activity records through a deterministic extraction layer, with strict schema validation and unit normalization to support downstream calculations and audits.

3. Deterministic emissions calculation

Emissions are computed using published GHG Protocol emission factors, with full traceability of inputs, formulas, and scope assignments.

4. Compliance analysis via RAG

Relevant regulatory texts are retrieved from a vector database and combined with calculated emissions to generate structured, explainable compliance evaluations.

5. Multi-agent orchestration

Each stage operates as a specialized agent within a state-driven workflow, enabling controlled execution, error handling, and reproducibility across the entire pipeline.

Development of Tailored Recommendations

Emissions driver identification

Rather than relying on opaque predictive models, the system identifies emissions drivers through structured activity extraction and deterministic emissions calculations. By breaking emissions down by activity type, energy source, region, and scope, the solution highlights which operational elements contribute most to overall carbon output. This enables decision-makers to focus on the areas where targeted changes are likely to deliver the greatest impact.

Scenario-based evaluation

The system supports scenario-based reasoning by allowing emissions to be recalculated under different assumptions, such as changes in energy consumption, fuel usage, or activity volumes. By comparing outcomes across scenarios, teams can evaluate potential interventions before acting, reducing uncertainty and avoiding unintended operational consequences.

Rule-based optimisation logic

Instead of complex optimisation solvers, the project applies clear, rule-based logic grounded in emissions factors, scope definitions, and compliance requirements. This approach ensures recommendations remain transparent, explainable, and aligned with real-world constraints, while still supporting meaningful reductions in emissions without disrupting service levels.

User-friendly delivery of insights

All recommendations and outputs are delivered through an intuitive, user-facing dashboard that presents emissions summaries, compliance verdicts, and improvement opportunities using clear visualisations. By minimizing technical complexity and manual effort, the interface ensures insights are accessible and actionable for users across regions and roles.

Results and Impact

The implementation of the AI-driven solution resulted in a 10% reduction in carbon emissions and approximately $5 million in annual cost savings, demonstrating that data-driven sustainability initiatives can deliver measurable financial returns alongside environmental impact.

Key outcomes from the project included:

• Measurable environmental and financial impact, achieved without disrupting operational performance.
• High adoption driven by transparency, with explainable calculations and auditable outputs enabling confident decision-making.
• Operationalized insights, where emissions reduction recommendations were implemented in day-to-day workflows rather than remaining theoretical.

Together, these outcomes show that sustainability initiatives are most effective when technical accuracy is paired with usability and trust, allowing analytical insights to translate into sustained operational change.

Related Omdena Project: Last-Mile Delivery Optimisation

While the primary focus of this case study is large-scale supply chain emissions management, similar optimisation principles have also been applied to last-mile logistics challenges.

AI-generated last-mile delivery routes highlighting optimised path selection and stop sequencing.

In a related initiative, Omdena partnered with Carryt to address last-mile delivery challenges in congested urban environments. The collaboration resulted in a route-optimisation solution built using Google’s Operations Research tools and OpenStreetMap data, enabling the generation and comparison of optimised delivery routes in Bogotá and similar dense urban settings.

The project focused on improving vehicle and driver deployment while maintaining service quality. By reducing unnecessary travel and optimising stop sequencing, the solution supported lower fuel consumption, improved delivery reliability, and better working conditions for drivers. This initiative illustrates how data-driven optimisation principles can deliver both environmental and operational benefits in last-mile logistics. Read more about this project in Delivery Route Optimization in LATAM Using AI Planning.

Why AI for Supply Chain Management?

Artificial intelligence enables supply chain organisations to move beyond reactive reporting toward proactive, data-driven optimisation. In the context of carbon management, AI provides the scale, consistency, and transparency needed to address emissions while maintaining operational performance. The benefits extend well beyond immediate carbon reductions and cost savings:

• Improved fuel efficiency – Greater visibility into logistics operations enables route optimisation and more efficient use of transport assets, reducing fuel consumption.
• Reduced waste – Analysing production, inventory, and distribution data helps minimise waste from overstocking, packaging inefficiencies, and spoilage.
• Support for renewable energy adoption – AI-driven insights highlight opportunities to integrate renewable energy sources, such as solar-powered warehouses or low-emission logistics options.
• Enhanced supply chain visibility – End-to-end analytics reveal bottlenecks, inefficiencies, and emissions hotspots across complex, multi-stage supply chains.
• Data‑driven decision making – Consistent, structured analysis allows managers to evaluate trade-offs and act on evidence rather than intuition.
• Compliance and reporting readiness – Automated tracking of emissions and regulatory requirements simplifies reporting and improves transparency and accountability.
• Stronger customer and stakeholder trust – Demonstrating measurable progress on sustainability strengthens brand credibility and appeals to environmentally conscious customers and partners.
• Scalability and adaptability– AI-powered systems can evolve with changing operations, regulations, and markets, supporting long-term resilience and sustainability.

Applications Beyond Logistics

Manufacturing

AI can analyse manufacturing operations to identify emissions reduction opportunities by optimising energy usage, material consumption, and production efficiency across facilities.

Energy Production

Optimisation models support cleaner energy generation by reducing emissions from power plants and improving the efficiency and integration of renewable energy sources.

Agriculture and Farming

By examining agricultural inputs and practices, AI helps reduce emissions from crop cultivation, livestock management, and fertiliser use while supporting more sustainable farming methods.

Technology and IT Services

AI-driven optimisation of data centres and IT infrastructure lowers energy consumption and associated carbon emissions, enabling more efficient and sustainable computing operations.

Construction and Real Estate

Assessing building materials, design choices, and operational energy use allows AI systems to reduce emissions throughout both the construction phase and a building’s full lifecycle.

Conclusion

This case study demonstrates how AI can play a meaningful role in transforming supply chains toward more sustainable operations when applied with a strong focus on practicality and trust. By combining structured data extraction, deterministic emissions calculations, and transparent, user-focused insights, the organisation achieved a 10% reduction in carbon emissions and approximately $5 million in annual cost savings. The results highlight that effective sustainability initiatives do not require opaque or overly complex systems, but rather reliable data, explainable logic, and clear pathways from analysis to action.

More broadly, the project underscores a key lesson for applied AI initiatives: impact depends as much on interpretability, usability, and operational fit as on technical sophistication. When emissions data is granular, calculations are auditable, and recommendations are easy to act upon, organisations can deliver measurable environmental benefits alongside real business value. This success illustrates that environmental responsibility and economic efficiency are not competing goals, but outcomes that can be achieved together through thoughtfully designed AI systems