Introduction: Balancing Performance and Responsibility

Crédit Agricole is committed to its societal project and one of the major goals is to “Act for climate and the transition to a low-carbon economy”. Delivering on this commitment requires drastically reducing greenhouse gas emissions at all organization levels.

In a context where the carbon footprint of the digital sector represents 4% of global CO2 emissions, artificial intelligence poses a major environmental challenge. Training a single language model can release as much CO2 as 5 cars during their whole lifecycle.

Faced with this reality, Crédit Agricole’s DataLab Groupe is focusing its resources on digital responsibility. The approach we implemented covers the whole lifecycle of data and artificial intelligence projects, combining technological performance and environmental responsibility.

Incorporating Sustainable Digital Practices to All Stages of Data and AI Projects

Our method is based on a simple principle: incorporating sustainable digital practices into all phases of Data and AI projects, from design to production.

Some Concrete Examples

From project framing: carbon impact serves as a decision criterion equal to performance or cost. Our teams methodically assess various approaches to select the most resource-efficient option.

Minimization principle: when simplicity is an option, we make it our starting point. The team prioritizes straightforward models before exploring complex architectures, expanding scale only when results clearly demonstrate necessity.

Continuous optimization: throughout the project lifecycle (development, training, deployment, inference), teams work to decrease resource consumption while maintaining essential business objectives.

Three Complementary Axes

This approach is structured around three complementary axes:

  • 📊 Data: Minimizing data usage to only what’s essential, efficient annotation, consistent compression and storage optimization
  • 🤖 Algorithms: Optimizing AI models and code efficiency, leveraging established solutions to prevent redundant and unnecessary training
  • 🏗️ Infrastructure: Tracking resource usage (CPU, GPU, RAM) and carbon footprint, streamlining task allocation

In June 2024, AFNOR released the Spec 2314 “Frugal AI” standard, developed through expert industry collaboration. Crédit Agricole’s DataLab Groupe played an active role in its creation and attained 91% alignment with the recommended practices, especially in areas of skills management, AI relevance qualification, and infrastructure efficiency improvements.

Concrete Tools to Measure and Optimize

The DataLab Groupe doesn’t just rely on theoretical principles. The team has developed and deployed a range of technical tools:

🔍 Creedengo (formerly EcoCode), Software Eco-Design

This code analyzer detects inefficient patterns and recommends resource-efficient fixes. Seamlessly integrated with SonarQube and GitLab CI/CD pipelines, it provides ongoing assessment of a code’s ecological footprint.

Crédit Agricole’s DataLab Groupe goes beyond merely using this tool - it ranks among the top open-source contributors developing Python eco-design guidelines. Substantial efforts were made in 2025 to expand the rule set available to the wider community.

Major benefit: Automated identification of poor coding practices during review processes, preventing inefficient code from reaching production environments.

📈 CodeCarbon, Measuring the Real Footprint

CodeCarbon provides an accurate calculation of carbon dioxide (CO2) emissions generated by computing resources running Python applications. This enables developers and data scientists to evaluate multiple implementations and select the most energy-efficient solution.

🌱 Ecologits, the LLM Specialist

Ecologits is a specialized tool for measuring the carbon footprint of Large Language Models, accounting for the model used, inference time, number of tokens, and geographic server location.

These measurements help inform project decisions (comparing options and identifying optimal solutions) while also educating end users about adopting more sustainable practices when utilizing AI applications.

📊 Infrastructure Reporting: Intelligent Capacity Planning

Our reporting tool contains a collection of detailed metrics with multiple granularities (global and per project views) to forecast requirements and prevent excessive resource allocation.

🔬 Innovations Specific to the DataLab Groupe

Beyond the above framework, the DataLab Groupe works on innovative approaches:

  • Synthetic data: Reducing storage and sharing needs using on-the-fly data generation.
  • CPU-based processing optimization: Lowering energy consumption by implementing and optimizing traditional models tailored to operate solely on CPUs without requiring GPUs.

Deployment at Group Scale

Crédit Agricole deploys these practices at the scale of the group via 7 Design Authorities, an AI × Digital Responsibility working group, and a normative framework implementing the European AI Act.

In Summary

This initiative aligns with the Group’s broader social responsibility mission, specifically our goal to support climate action and facilitate the shift toward a low-carbon economy.

At the DataLab Groupe, our focus is on creating high-performance AI solutions while minimizing their environmental impact. We achieve this through comprehensive measurement frameworks, sustainability principles embedded from project conception to delivery, and careful consideration of each technical decision.

The process delivers tangible value:

  • Reduced costs via efficient resource utilization
  • Enhanced performance through optimized algorithms
  • Proactive compliance with emerging European regulations (AI Act)

Our aim is to prove that excellence and environmental stewardship are complementary, helping to promote these sustainable practices within our organization and across the broader industry.

To Go Further

Useful Resources:

  • Crédit Agricole Group’s Societal Project
  • AFNOR Spec 2314 “Frugal AI” Framework
  • CodeCarbon - Carbon Footprint Measurement
  • Green Software Foundation

Are you interested in these topics? Join Crédit Agricole’s DataLab Groupe to contribute to developing more responsible AI. Discover our opportunities on the Group’s career site.