Starling Bank, one of the UK's first digital banks, moved its data warehouse to BigQuery and its machine learning work to Vertex AI on Google Cloud, and now uses Gemini in customer-facing and internal tools, according to an undated Google Cloud case study: the first phase of Scam Intelligence reached consumer testing in four weeks, Gemini summarising call transcripts saves 8,000 hours a month, a Gemini help centre cut customer service referrals by 50%, and Scam Intelligence has increased the rate at which customers cancel marketplace payments by 300%.
Frédéric Laurent, Deputy CIO at Starling, said: "We lacked the infrastructure to support the necessary data science work." He added: "We built our Scam Intelligence feature really quickly thanks to Vertex AI."
Starling built its core platform on cloud infrastructure from the start, but as the team used machine learning more heavily, the existing data and ML stack became a bottleneck. For a regulated bank the requirement was that data scientists could iterate quickly without risking data exfiltration, which is why Vertex AI Workbench was central.
BigQuery is the data foundation, with Vertex AI for model development and Vertex AI Workbench as the secure environment for data scientists. Scam Intelligence sends a picture of a marketplace advertisement into Gemini's multimodal models, which look for red flags such as a suspiciously low price or pressure tactics from the seller; the prompt is continuously evaluated and managed through Vertex AI Pipelines, and Model Armor provides guardrails so the solution can only be used as intended. Spending Intelligence turns a natural-language question, such as how much was spent on coffee last week, into a structured function call that triggers Starling's analytics engine, and Gemini is fed customer-facing content to answer help centre queries.
Starling sells the model onward: with Engine by Starling it provides its cloud-native core banking platform to banks worldwide, whose data model and event-based architecture let them offer AI-powered customer services and deploy AI agents alongside staff. Krishna Aswani, Regional Manager - North America at Engine by Starling, said: "Underpinning any AI solution is the need for a robust, modern data foundation, which is how we've architected the Engine platform from day one."
Laurent said: "Google Cloud has brought us enterprise data and AI capabilities where previously we were limited in what we could achieve". Aswani added: "By leveraging our technology and everything we've learned with Google Cloud, we can help clients build and deploy generative and agentic AI solutions safely, at scale." Engine's API-first and MCP-ready capabilities give banks foundations for an agentic workforce.
Details