Last week in London, Senior Product Manager Mouna Samrout delivered a keynote at the NexGen Banking Summit that cut to the heart of AI strategy in financial services: smaller, purpose-built models outperform when control, traceability and governance matter most. Her talk, “Banking on sovereignty: Why smaller models will win,” showed how banks gain precision and auditability by prioritising domain training, data residency and transparent decision paths.
Three pillars of sovereign AI
Mouna outlined three pillars that resonated across the room. First, data sovereignty ensures sensitive information remains within the right jurisdiction and supports straightforward audits. Second, domain accuracy comes from models trained on financial language, regulatory texts and real supervision needs. Third, transparent governance provides end-to-end traceability so leaders can evidence how outcomes were reached.
Joining Mouna on the ground was Jon Oller from Aveni’s sales team, meeting banking leaders and relaying their priorities from the expo floor. Jon noted a clear shift: firms are moving past experimentation and are now focused on deploying AI with firm guardrails, measurable outcomes and regulator-ready evidence. For CFOs planning 2026, that means proving value with hard metrics and clear governance.
Mouna closed with a line that captured the theme of the day: “In financial services, the future does not belong to those who have the biggest models. It belongs to those who have the most control over them.” That control means UK data residency, referenceable sources, robust monitoring, and audit trails that stand up to scrutiny.
Across panels and hallway conversations, one message held steady: success now depends on scaling AI with evidence, not assumptions. NexGen showcased that the leaders to watch are building with sovereignty, measurement and governance from day one. Aveni’s approach aligns with that direction, giving banks the control they need to move from pilots to reliable production at pace.
Read the original article here.
