Will Polish banks build joint AI? Experts see potential, but also barriers.

The material was created in cooperation with Pekao SA

Is building a common infrastructure and shared AI solutions for the banking sector in Poland, similar to Blik, possible and realistic? Could this accelerate the adoption of artificial intelligence and reduce costs? Participants in the debate "AI for the sector – common infrastructure, competencies, responsibility" discussed these issues during this year's Banking & Insurance Forum in Warsaw.

"A shared AI environment for financial institutions? At first, this idea seemed counterproductive, but the more I think about it, the more it makes sense," said Michael Donahue, CTO of Pentaho at Hitachi.

Donahue pointed to the chaos in international markets regarding the use of artificial intelligence. Everyone is operating alone, as individual contributors, yielding mixed results. AI models come and go, and the lack of a common framework makes "the cost outrageous." He believes a common environment that leverages shared costs but also shared security measures is the way to go.

The remaining debate participants generally agreed with this thesis, but emphasized that the path to full cooperation in the use of AI in the banking sector is not so straightforward. Marcin Zygmanowski, Vice President of Bank Pekao SA, overseeing the Technological Transformation and Innovation Division, noted that creating a single data center for the entire banking sector would be a huge challenge. This is because, for individual banks today, different applications of AI are a differentiator that can provide them with a competitive advantage.

At the same time, Vice President Zygmanowski sees opportunities for cooperation in areas that don't determine an entity's competitiveness. He cited AML, or the mandatory anti-money laundering system, as one example. Banks are required to use a system that consumes significant resources without delivering added value.

"I'm convinced that there are non-competitive areas where we can work together. This includes cyber defense, the previously mentioned AML, KYC (customer verification – ed.), and others," noted Marcin Zygmanowski.

"I agree that, on the one hand, a shared infrastructure makes sense," said Piotr Kusek, a specialist at Comarch. "However, on the other hand, I know that data is extremely important to banks; it's a valuable resource they don't want to share. Therefore, finding a model that can share it seems very difficult," he said. In his opinion, a potential opportunity for the sector would be the so-called federation model, which allows AI models to be trained on data stored locally on devices instead of transmitting it to a central server.

"We often see how difficult it is to unify the operations of a single capital group. And we imagine how difficult it would be to build a common infrastructure for a group of banks that sometimes cooperate and sometimes compete," noted Krzysztof Daniel, head of data strategy at DXC Technology Polska. He added, however, that as the AI ​​revolution becomes commonplace, joint efforts across the sector will follow.

– Talking about the infrastructure itself without answering the question: "why do it?", secondly: "what to do?", and thirdly: "what for?", is a classic mistake – said Aleksander Poniewierski, advisor to the management board at PKO Bank Polski.

As he explained, there are no competitive advantages in data centers themselves, and computing power can be purchased, therefore there are no answers to the questions: "why?" and "what should be done?" – If we also don't know "what's next?", i.e. how to commercialize it, how to make money on it, how to build a so-called sustainable business, then we won't get anywhere – said Poniewierski.

While the prospect of a shared AI infrastructure seems somewhat distant, banks are well advanced in adopting AI, though there are still challenges. Marcin Zygmanowski, Vice President of Bank Pekao, identified four classes of AI solutions that most banks are focusing on. These include various types of co-pilots, i.e., AI assistants, tools designed for this purpose, personal assistants for employees, and assistants for customers.

"What's missing? Probably the most proven use cases that we're ready to implement in production, not just experiment with. This is important because full project implementation can take several months, and in that time, new, "better" models are already emerging," said Zygmanowski. He added that challenges include changing employee awareness of AI use, as well as implementing and monitoring the operation of AI assistants for customers.

Krzysztof Daniel assessed that although the banking sector is subject to strong regulatory pressure, contrary to concerns, this is its strength when it comes to AI implementations (if only due to the maturity of its processes and structured data). Another question is whether the "productization" of AI solutions has a chance of working in banking?

"'Build' or 'buy'? It depends," replied Marcin Zygmanowski. "In areas that differentiate the bank, at the touchpoints of customer contact and service, we seek proprietary solutions; we want to own them. Where non-differentiating measures are applicable, we can buy a ready-to-use solution, although such models are still few and far between," summarized the vice president of Bank Pekao SA.

The material was created in cooperation with Pekao SA