What challenges did Atos face before AI implementation?

Upskill workforce at scale with hands-on, gamified generative AI fine-tuning rather than only training/certifications. Create domain-specific underwriting capability.

What was the impact of the AI solution?

400+ participants (409 on leaderboard), 4,100+ fine-tuned models created. Top fine-tuned 3B models achieved >93% win rate vs much larger 90B models. Accelerated practical skill-building from months to two weeks and 85% of participants felt more confident implementing GenAI solutions with customers.

Source

Atos AI League: Intelligent Insurance Underwriter fine-tuning with Amazon SageMaker AI

Use case typeAI model trainingUpdated Mar 17, 2026

Organizations pursuing AI transformation can face a familiar challenge: how to upskill their workforce at scale in a way that changes how teams build, deploy, and use AI. Atos partnered with AWS to deliver a hands-on, gamified learning experience through the AWS AI League to accelerate applied AI skills across the organization. Atos selected a use case called the Intelligent Insurance Underwriter to fine-tune a large language model capable of analyzing insurance scenarios and providing underwriting guidance.

Organization: Atos
Industry: Insurance
Location: France
Published: March 2026

Reported outcomes

+93%

win rateOther quantified impact

409 peopleleaderboard participants4,100 modelsfine-tuned models created+85%confidence in implementing GenAI solutions

Strategic outcomes

Employee experienceBuilt domain-specific underwriting skill across the workforceSpeed & agilityCompressed practical AI skill-building to two weeksCost efficiencyDemonstrated lower-cost specialized model development

Primary read

Use case focus

Showing 2 of 2

1AI model training
2Training infrastructure modernization

Upskill workforce at scale with hands-on, gamified generative AI fine-tuning rather than only training/certifications.
Create domain-specific underwriting capability.

Using AWS AI League, Atos participants fine-tuned Meta Llama models via SageMaker JumpStart and built an Intelligent Insurance Underwriter assistant.
Dataset creation in JSONL, fine-tuning workflows in SageMaker Studio, deployment to SageMaker endpoints for inference, and iterative evaluation/leaderboard to tune hyperparameters.

Technologies

Amazon SageMaker AI Amazon SageMaker JumpStart Amazon SageMaker Studio Amazon S3 PartyRock Meta Llama 3.2 3B Instruct

400+ participants (409 on leaderboard), 4,100+ fine-tuned models created.
Top fine-tuned 3B models achieved >93% win rate vs much larger 90B models.
Accelerated practical skill-building from months to two weeks and 85% of participants felt more confident implementing GenAI solutions with customers.

Architecture

Participants fine-tuned Meta Llama 3.2 3B Instruct models using Amazon SageMaker JumpStart and Amazon SageMaker Studio, with datasets stored in Amazon S3. The workflow included JSONL dataset creation, hyperparameter tuning, deployment to SageMaker endpoints for inference, and leaderboard-based evaluation during the AWS AI League. The article also notes use of an AWS provided PartyRock application for dataset generation support.

Sources & evidence1

Groundedness: 4/5Type: Blog PostPublished: Mar 17, 2026Publisher: AWSEvidence: VendorConfidence: Medium

Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Customer evidence