How Revolut runs AI at scale

2025-10-09 · Source: Air Street Press · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, FinTech & Digital Financial Services · Depth: Advanced, medium

Summary

Revolut, a regulated bank serving over 70 million customers across 40+ countries, manages AI at scale by focusing on a robust control plane rather than just the models themselves. Nikolay Donets, head of ML engineering, highlighted at RAAIS 2026 that the challenge lies in governance, cost controls, measurable fallbacks, and human review for its 200+ products. Since 2022, with the rise of large models via APIs, Revolut shifted from classical ML to a unified approach. They implemented a single gateway and governed AI by use case, aligning with the EU AI Act. This central platform, which now handles roughly twice as many generative use cases as classical ML, ensures "zero effort" upgrades and centralized compliance. Their AI assistant, AIR, evolved from a basic chatbot to a sophisticated system handling 25,000 calls monthly, resolving issues 8x faster, and boosting resolution rates from 17% to 80%. Revolut maintains strict human oversight, ensuring no life-changing decisions are made by AI.

Key takeaway

For AI Architects or MLOps Engineers building systems in regulated industries, Revolut's strategy underscores that operationalizing AI at scale hinges on a centralized control plane, not just model performance. You should prioritize establishing a single, governed gateway for all AI access and implement robust fallback mechanisms for external models. Critically, ensure human oversight remains the final arbiter for any life-changing decisions, even as AI improves efficiency.

Key insights

Scaling AI in regulated environments demands a robust control plane and use-case-centric governance over individual model focus.

Principles

• Govern AI by use case, not model.
• Centralize AI access via a single gateway.
• Implement multi-tier fallback chains for external models.

Method

Revolut established a single AI gateway with a governance layer, shifting from individual team libraries. They adopted use-case-based risk assessment, implemented fallback chains for external models, and embedded AI engineers in product teams while maintaining a central ML engineering group for standards.

In practice

• Deploy a central AI gateway for all product teams.
• Mandate per-model visibility, not just interface monitoring.
• Prioritize smallest effective models over newest.

Topics

• AI Governance
• MLOps
• Financial Services AI
• Generative AI
• AI Control Plane
• Revolut AIR

Best for: CTO, VP of Engineering/Data, Executive, MLOps Engineer, AI Architect, Director of AI/ML

Related on AIssential

• Nikolay Donets of Revolut at RAAIS 2026 — The hard problem in production AI is building a unified platform for all stakeholders, not just isolated models.
• Presentation: The AI Gateway: Scaling Centralized Inference Across Decentralized Teams — AI model gateways provide a critical control layer to manage diverse AI inference across decentralized teams while maintaining centralized governance.
• Why Financial AI Can't Scale Without Unified Governance with James Dean of Google and Mark Crean of Securiti — Effective data governance, not technology, is the primary bottleneck for scaling AI in financial services.
• The 100-agent benchmark: why enterprise AI scale stalls and how to fix it — Scaling agentic AI requires robust governance and unified platforms to overcome significant operational costs and production challenges.
• Operationalizing AI for Scale and Sovereignty — AI factories enable scalable, sustainable, and governed AI by prioritizing data control for secure, tailored solutions.
• We Didn’t Know What We Didn’t Know: Standing Up Enterprise AI Services at Scale — Scaling enterprise AI requires robust infrastructure, clear policy, and strong cross-functional partnerships, not just advanced models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Air Street Press.