How Generali Built One AI Engine for 40 Subsidiaries — and Saved €200 Million

In 2021, the data and AI team at Generali made a counterintuitive choice: build shared infrastructure across all 40+ subsidiaries before validating a single use case in production. Three years later, the results answer every objection — over €200 million in operating costs saved, claim settlements that fell from several days to seconds, and an AI application portfolio that grew from 5 to more than 50.

For readers who want to go deeper, Grace Certified offers real-world decision-making cases judged by AI.

Generali Group is one of the world’s three largest insurance companies, operating through more than 40 subsidiaries across Europe, Asia, and the Americas — from Generali Deutschland to Argentina’s La Caja. Each entity had spent decades accumulating its own data environment, analytics stack, and vendor relationships. When AI became a strategic priority, each subsidiary was already running independent experiments: building overlapping models, managing duplicate infrastructure, solving the same problems in isolation.

The team saw the trajectory clearly. Left unchanged, this pattern would produce a fragmented AI landscape — dozens of islands with incompatible data standards, compounding integration costs, and diminishing returns on each new investment.

The Architecture Decision: Centralize Before Scaling

Working with AWS, the team designed a centralized Global AI Engine integrating four services: Amazon Redshift for unified data warehousing across all entities, AWS Glue for cross-subsidiary data integration and standardization, Amazon SageMaker for machine learning model development, and Amazon Bedrock for generative AI applications. The platform consolidated data from all 40+ operating entities into a single, governed environment.

The structural consequence was precise: instead of each subsidiary building its own fraud detection models, claims processing pipelines, and document classification systems, the Global AI Engine created shared “AI artifacts” — validated models, inference services, and data pipelines accessible by any entity. Engineering work completed once became available everywhere, eliminating the duplication at its source.

Andrea Pietrasanta, Generali’s Group Head of Data, AI and Automation, and Antonio Montuschi, Head of Group AI and Generative AI Development, led the migration of existing analytics solutions from third-party providers to AWS. The consolidation itself created the enabling condition: uniform data standards across all subsidiaries made shared models viable where fragmented data had previously rendered them impractical.

Three Years of Compounding Returns

The results confirmed the core hypothesis of platform-first AI investment:

• Over €200 million in operating costs saved across the group within 3 years
• 16 flagship AI use cases deployed from the shared engine — each deployment faster and less costly than the previous one
• 21 million shared services processed per month through AI-powered workflows
• AI applications grew from 5 to more than 50 in 3 years, with a target of 200 within the following 3 years
• Claim settlement time fell from several days to 1 day for standard cases — and to seconds for simple health claims handled automatically by AI agents on Amazon Bedrock

The trajectory reveals the compounding economics of shared infrastructure. The first use case required the largest investment: building the Global AI Engine itself. Each subsequent deployment cost progressively less, because data pipelines, model validation protocols, and inference infrastructure were already in place and governed. The 5→50 growth in AI applications over 3 years is the direct evidence of this leverage. The €200 million in savings is the cumulative outcome.

The Principle Every Multi-Entity Organization Should Carry Forward

For any organization managing multiple business units, subsidiaries, or brands, Generali’s model carries a precise implication: the sequencing of AI investment determines the economics more than the speed of any individual deployment.

The dominant pattern in enterprise AI programs is use-case-first — identify a compelling application, build a solution, declare success, and move to the next one. The predictable result is a collection of isolated deployments with inconsistent data standards, duplicated infrastructure costs, and technical debt that compounds with each new project. Each use case is individually successful; the portfolio as a whole fails to generate leverage.

Generali’s sequence inverted this dynamic. The Global AI Engine was the first investment — before any single use case was validated in production. The payoff came as accelerating deployment velocity: as the engine matured, the time and cost required to launch each new use case decreased, while quality and consistency across all subsidiaries increased in parallel.

The €200 million figure also answers the most common objection to platform-first approaches: that the upfront investment in shared infrastructure delays time-to-value. Generali’s trajectory indicates the opposite. A platform built early generates compounding returns across the entire portfolio — returns that a sequence of individual use-case investments cannot replicate at scale.

For boards and C-suite leaders evaluating AI investment strategy, Generali’s 3-year arc provides a benchmark grounded in one of the most data-intensive, compliance-sensitive sectors in existence. The question worth carrying into the next planning cycle: are your organization’s AI programs building toward a shared engine — or accumulating islands?