88% of enterprises use AI. Fewer than 10% have scaled it to deliver real value. Dalibor Hlava, Chief AI Officer at Galytix, joined Martin Zika at Innovation Week Studio in Prague to explain what is holding banks back, and what it takes to get AI working in production.
Banks have been investing in AI for years. The pilots look promising, the demos are convincing, and the vendor promises keep getting bolder. Yet when it comes to actually running AI inside a live credit workflow, at scale, inside a regulated institution, most of it does not make it. Dalibor and his team at Galytix went looking for the reason. They tested generic AI tools including Microsoft Copilot and Claude against a purpose-built Digital Credit Officer across more than 200 real credit memorandums, measuring not just output quality but whether the results would hold up in a production environment. The models were smart enough. That was never the issue. What kept failing was reliability, explainability, and auditability. And those are exactly the things a regulator will ask about first.
1:58 Why generic AI agents cannot handle production-grade banking work, and why better models alone will not fix that
5:52 The five structural weaknesses his benchmarking uncovered, and which one surprised him most
9:39 Why putting a human in the loop does not solve the problem, it just moves it
12:13 What problem structuring means, explained for a CRO not an engineer
18:27 The risk-return frontier: three archetypes of AI deployment and where banks go wrong
21:20 How to think about the build vs buy vs partner decision when your data is messy and your methodology is custom
24:54 What the future of banking AI looks like, and why it is a network of specialised digital officers, not one universal super-agent
"The winning systems will not be the ones that are the smartest. They'll be the ones that are defendable in production. The ones that actually survive the audit." — Dalibor Hlava, Chief AI Officer, Galytix
Dalibor Hlava is Chief AI Officer at Galytix, where he leads the development of specialized AI systems for regulated financial institutions. He has been working at the intersection of AI and banking since the early days of the current generative AI wave and recently authored a research piece on how domain specialised ‘Digital Officers’ can successfully scale AI in Credit.
The Spotlight Series Podcast Was Recorded at Innovation Week Studio, Prague
Galytix (GX) is a specialised AI firm built exclusively for financial institutions. Over the past decade, GX has trained its AI on deep credit and risk domain knowledge and successfully deployed it at 30+ financial institutions worldwide. GX's AI is designed for consistency, robustness, and regulatory trust and is fully adapted to the data, risk frameworks, processes, and compliance standards of the institutions it serves. GX AI assistants operate inside complex, heavily regulated institutions, supercharging prospecting and productivity, sharpening decision quality, and delivering a step-change in client experience.
If you're evaluating AI for credit or risk and want to see how a specialised system performs against your own workflows, request a demo today.