Ecouter sur Spotify 
Ecouter sur Apple podcast 
Ecouter sur Deezer 
Ecouter sur Ausha In the second part of a three-part series of interviews, SBS CEO Eric Berry explains why banks can’t simply outsource their AI strategy to US cloud giants. Client sovereignty concerns, strict regulation, and the need for deterministic, auditable decisions mean AI in banking has to be layered carefully (Gen AI today, agentic AI within the next year). SBS keeps its core banking products AI-free and lets clients choose their own AI provider, while building toward an “AI Gateway” control plane for agentic systems.
The sovereignty dilemma in banking AI
Most companies are told to run AI on major US cloud platforms. SBS, the banking software provider led by CEO Eric Berry, is taking a more cautious path.
Earlier this year, SBS signed partnerships with three of the biggest US AI providers (Anthropic (Claude), Microsoft (Copilot), and Amazon) to accelerate internal adoption. But a parallel concern quickly surfaced: SBS’s banking clients, particularly in Europe and Africa, are increasingly focused on AI sovereignty and reducing dependence on US-based technology.
SBS’s response is a middle path:
- Use best-in-class US AI capabilities where speed and quality matter most today
- Simultaneously develop open-source LLM alternatives as a hedge against future disruption
- Remain fully agnostic on which LLM clients choose to use in production
Why agnosticism matters: if SBS forced a specific model on clients, it would also inherit the liability for that model’s behavior. Major clients, like BNP Paribas, which works with Mistral, want to retain that decision-making power themselves.
Why AI in banking must stay deterministic
Banking is a regulated industry, and regulation demands determinism. Berry uses a simple example: if you check your bank balance, you expect an exact figure, not an AI-generated approximation.
This principle shapes SBS’s entire AI architecture:
- Core banking products remain AI-free, preserving deterministic, auditable outcomes
- AI capabilities sit in a separate data platform and semantic layer, which supports explainability and full audit trails
- Every AI-assisted decision must be traceable, critical for regulatory compliance and maintaining a banking license
The first live deployment of this model is a Scottish building society, running SBS’s AI-enhanced data platform since July. One result: a customer-behavior analysis that previously took three weeks now takes under one minute, including generating a recommended action.
From gen AI to agentic AI: What’s next for banks
Eric Bierry draws a clear distinction between two stages of AI maturity:
Gen AI (today): Generates insights and recommended actions, but doesn’t autonomously change core systems.
Agentic AI (next 6–12 months): Would allow AI agents to take action directly (updating account records, triggering transactions) with two prerequisites still in development:
- Business APIs that safely expose core systems to automated updates
- Human-in-the-loop governance rules to ensure no fully automated decision happens without oversight
A new challenge unique to agentic AI in financial services: it’s not just about orchestrating your own AI agents, it’s about managing external agents entering your systems. SBS and sister company Axway are addressing this jointly through an AI Gateway, designed as a control plane for agentic orchestration in financial services.
The bigger picture: AI as a trust strategy, not just a tech strategy
Berry frames SBS’s AI strategy around a longstanding principle: being a trusted long-term partner to banks, not just a technology vendor.
In conversations with bank CEOs, the same priority comes up repeatedly: avoiding reputational or systemic failure. Clients want SBS to keep their infrastructure secure, compliant, and reliable first, and to prepare for the agentic AI era second, without rushing.
