Generative AI in Banking: Real Use Cases Beyond the Hype

WAU Marketing
8 hours ago
4 min read

The generative-AI question at your institution isn't "which chatbot do we add." It's "how ready is your data for AI to be worth anything."

There's enormous noise around generative AI in banking, and almost all of it points the wrong way: the assistant that answers questions in the app. That's the visible part and the least transformative. The cases that truly move the needle are deeper, where AI touches the bank's systems, code, and data. And there an uncomfortable truth shows up: generative AI pays off in direct proportion to how accessible and clean your data is—that is, to how modern your core is.

What the big players are already doing (and it isn't a chatbot)

It's worth looking at real cases, not promises. JPMorgan rolled out its internal generative-AI platform, "LLM Suite," to about 140,000 employees for tasks like drafting documents, analyzing transcripts, and comparing contracts; the bank projects up to $2 billion in AI-related value, as CIO Dive reported. Morgan Stanley put an assistant on top of its knowledge base and a tool that generates meeting notes, with 98% adoption among its advisors, according to InvestmentNews.

None of those cases is "the customer chatbot." They're tools that connect to internal documents, data, and workflows. And that only works if the data is available via API, not buried in a core that only speaks in batches.

The case we care about most: AI that modernizes the core

There's a generative-AI use that closes the loop on everything we've written in this blog: modernizing the core itself. Citi equipped about 30,000 developers with AI-assisted coding tools and retired 2,000 legacy applications, CIO Dive reported; it later extended agentic AI to 40,000 developers, with an internal tool that completed hundreds of thousands of code reviews, according to American Banker. Generative AI is being used to speed up exactly the work of getting the bank out of its old systems.

Here's the honest caveat, because "beyond the hype" also means not believing everything: a controlled trial by METR found that, among experienced developers working on real code, using AI tools increased task time by 19%, even though they believed they'd gone faster. The lesson isn't "AI doesn't work"; it's that the result depends on context, data, and discipline—not on switching on a license.

Risk and compliance: where generative AI pays off fast

In fraud prevention and compliance, the real numbers are striking. Mastercard reported that its generative-AI model improves fraud detection by 20% on average and cuts false positives by more than 85%, according to Fortune. Commonwealth Bank of Australia reports a 50% drop in scam losses and a 40% reduction in call-center wait times with generative-AI support. These are fronts where the bank recovers its team's time and cuts measurable losses—not demos.

LATAM: advancing, but with a deeper gap

The region is moving. Nubank acquired Hyperplane, a startup that builds foundation models for banks from their own data, TechCrunch reported. In Mexico, Banorte was the first bank authorized by the CNBV to run its AI models in the cloud, according to IT Masters Mag. In Colombia, 73% of financial institutions already have AI-based implementations, per Asobancaria figures reported by La República.

But there's a structural brake worth naming. The Economic Commission for Latin America and the Caribbean warns that the region holds 6.6% of global GDP and just 1.12% of global AI investment: we consume more solutions than we build. For a regional bank, that makes it even more decisive to have the foundations—data and core—ready to leverage the AI that is within reach.

Why it all comes back to the core and the data

This is the part the noise omits. The Central Bank of Brazil, in its survey of financial-system institutions, found that 45% flag poor data quality as an AI risk, and that only 26.7% of regulated entities use AI models, as Finsiders Brasil reported. Translated: AI's bottleneck isn't the model, it's the data. And the data lives in your core.

McKinsey estimates generative AI could add $200 billion to $340 billion a year to global banking, equal to 9 to 15% of the sector's operating profits, in its analysis of the value of generative AI in banking. That prize isn't captured by whoever buys the most expensive model; it's captured by whoever has the architecture to feed it: data accessible via API, in real time, governed.

How we see it at WAU

At WAU we don't sell you a chatbot. We design the core that makes generative AI possible: data exposed via API, in real time, with governance and traceability, so that when you connect AI—for compliance, for productivity, to modernize your own code—it has something to feed on. AI doesn't fix a legacy core; a modern core is what makes AI useful.

If you want to separate the hype from the cases that actually pay off at your institution, let's talk. We'll help you see what your architecture needs for generative AI to deliver. 👉 Book a conversation with our team.