Starling Bank has put AI-powered customer tools into production, embedding what it calls Spending Intelligence and Scam Intelligence directly inside its app.
Group CEO Raman Bhatia framed it in the boldest terms possible: "We are defining what comes next in the intersection of digital banking and AI."
That is a large claim for a feature that lets you ask how much you spent on coffee last year.
What the tools actually do
Spending Intelligence lets customers ask plain-English questions about their finances. The demo use case is the coffee query, which is fine, but most banking apps already surface spending breakdowns without requiring a conversation.
Scam Intelligence is more interesting. Customers can upload images of items and ads from online marketplaces and get an AI analysis of scam indicators before making a payment. If the execution matches the pitch, that is a genuinely useful feature in a market where purchase fraud is rising fast.
Starling says the AI work has delivered a 300% reduction in payment fraud. That is a striking number, though the baseline and methodology behind it are not specified.
The deployment story matters more than the product story
The more significant detail is how Starling got here. Bhatia described it as solving the last-mile deployment problem that traps most enterprise AI projects in proof-of-concept purgatory.
The bank says it built its platform around workflow automation and flexibility, which allowed it to move models into production rather than letting them rot in a sandbox. Generative AI tools have been rolled out to roughly 80% of eligible staff. An internal AI Champions programme trains peers and embeds AI into the software development lifecycle.
Cross-functional teams of engineers, product leads, risk officers and compliance staff pursue AI opportunities together.
This is the unsexy part that most banks skip when they talk about AI. Getting a model to work in a demo is easy. Getting it through compliance, into production and in front of customers without breaking anything is where most institutions stall.
The real question
Starling's advantage has always been that it was built as a technology company that happens to hold a banking licence. That makes deploying AI tools structurally easier than it is for legacy banks dragging decades of technical debt behind them.
Whether customers care about conversational finance is uncertain. Whether they care about not getting scammed is not.
The fraud tool is the one to watch.
