EXCLUSIVE: “A Healthy Approach to AI” – Renato Rocha Souza, RBI in ‘The Fintech Magazine’

RBI’s Renato Rocha Souza takes us on a deep dive of the bank’s business-centred approach to AI adoption… and ponders on the dangers of the ‘culture war’ between developers

Chief information officers agree that all their organisations’ IT work will involve some level of artificial intelligence by 2030, according to a poll by US consultancy Gartner. But 72 per cent of them also admitted that their existing AI investments were losing money for their organisations, or were, at best, breaking even.

This has led to the emergence of another strain of AI – Angst and Indecision among senior executives, which is having a trickle-down effect on teams. So, how can banks ensure they capitalise on the technology’s promise without becoming overwhelmed by the constant hype and noise?

Raiffeisen Bank International (RBI) took steps to avoid AI Type 2 by implementing a roadmap that ensures AI Type 1’s use is strictly business and not technology-driven. Renato Rocha Souza, Director of RBI’s AI Centre of Excellence, believes large doses of humility and realism are required to avoid wasting time and money.

He admits: “A five-year horizon is too long, it’s hard to predict what will happen – we don’t know if we will have artificial general intelligence by then. So, from the beginning we’ve been focussed on people, mindset and culture. Of course, the technology is important, but staff need to be able to envision how AI can be used so we can embed it in the core business.”

Its strategy has been to ensure the technology can be used ‘top down’ to revolutionise organisational processes, and also from the ‘bottom up’ to boost staff efficiency and productivity. Souza says RBI began its AI journey in 2018, and the examination of generative AI started three years ago. While his AI Centre of Excellence leads the development of AI use cases and drives the bank’s technological transformation, an AI Committee works to ensure compliance with regulations such as the EU AI Act, and strategic alignment across the organisation’s various divisions, which span 23 countries with a particular focus on central and eastern Europe.

A use-case group was formed to highlight areas for implementation and provide ideas on how solutions would work. And to ensure staff at alllevels were aware of AI’s potential and the need to be open to its adoption, RBI introduced an AI Pioneer programme, which began with 200 staff members across divisions who allocated 20 per cent of their time to AI.

The programme progressed, and now RBI has what it calls ‘AI Ambassadors’ in every department. All staff are trained in using AI, and, crucially, in using it responsibly.

Souza says: “Our AI Ambassadors have a responsibility to promote reverse mentoring with the business’s leaders. We are upscaling leaders because there are so many layers of understanding needed to implement AI, and to foresee potential consequences.

“We are embedding AI within RBI based on business priorities, not technological priorities. If your managers cannot understand how far they can go in embedding AI in their processes, they risk just playing with nice and exciting tools without having a proper goal.”

Watching the watchmen

AI use cases are assessed and prioritised with what Souza describes as a matrix that considers revenue expectations, regulatory constraints and the technology available. He admits that the speed of technological development means it’s ‘not unusual’ for objectives to change as more powerful solutions emerge.

“AI models are evaluated in three ways, and the first is by considering the model providers,” he says. “Use cases are developed over the layer of IT platforms. We check for model compliance, model bias and ethics. We agree service level and non-disclosure agreements with providers, and we only use the technologies that are approved by our platform teams.

“Next, we have the AI Committee, which is made up of data scientists and other experts, who must approve every use case. Then we have internal audits. Model governance systems track each algorithm deployed in each model. That’s our third layer in this ‘many eyes inspection and evaluation’ of the models.

“No models are run without human oversight, so there’s constant supervision over performance and results of models, so we can check for biassed results, for example.”

Souza adds that to assess a project’s success, financial KPIs are important but not the only measure. “For instance, the AI Pioneers’ objective was to upskill the workforce, and it’s hard to attach financial KPIs to that,” he says. “But you can measure other things, like productivity, satisfaction and retention. So we connect our artificial intelligence KPIs to people as well.”

The bank works with a range of providers, such as OpenAI, Amazon Web Services, Azure, and the Databricks Lakehouse platform, with solutions typically created in-house.

“Internally, our solutions are containerised because we have a pre-occupation to avoid critical mission systems becoming vendor-locked,” he explains. “That doesn’t prevent us from developing use cases with fintechs, but we evaluate and provide oversight of models from third parties.

“A model we’re considering right now is an AI agent that takes collections for sales teams. So, if a customer owes something to the bank, they can negotiate with an agent instead of a person. With customer-facing models, we must be extremely careful to prevent hallucinations and bias. Because we are governed by the EU AI Act, there must be human oversight, so human input is a given for use cases.”

Acknowledging bias

Souza stresses that caution is required when choosing AI platforms to underpin services, since platforms are shaped by governmental regulations and the societal norms from which they draw data. He doesn’t mince his words, arguing that differences across jurisdictions and platforms amount to a ‘culture war’, and a failure to appreciate the impact of inherent biases means such factors will affect a bank’s own AI models.

“We know models are biassed because they follow the data that they were trained on,” he says. “We see differences in how the US regulates AI, and the emphasis in China, for example. So we must take care not to outsource creative aspects, because the more we outsource decisions, the more we delegate them to AI.

“If you’re using Grok, for example, you buy into certain decisions that are acceptable to Grok. Likewise, if you’re using models based on OpenAI, or Gemini.

“At RBI, we’re not changing the future of our business, but we are changing our future business with AI. We believe it’s important for us to plan the AI transformation with people at the centre. We can automate, but in the future, when agents outweigh employees, humans must remain in the pilot seat and AI in the co-pilot seat, and not vice versa.

“There’s a huge discussion around the consequences of AI adoption that goes far beyond our business processes, though.”

Souza admits the deployment of AI to run low-knowledge, data-intensive tasks will disrupt a whole layer of junior positions, and businesses will inevitably prioritise the hiring and cultivation of ‘senior and high expertise professionals’. Ultimately, some organisations will one day be run entirely autonomously by agents, although he believes that it won’t happen overnight.

“We’re going to get there, but not as fast as some might wish,” he says. “We’ll have a smooth progression by implementing low-risk, high data-intensive agents, or level-one agents, then two, three, four and five until we have more complex, high-risk use cases that will always demand human oversight.

“Looking at society in general, what consequences will this deep dependence on AI models have on our lives? What geopolitical effects will result if we depend on the models of one particular provider?

“We must understand that we are deciding our future without ultimately knowing what kind of decisions are being taken by the technology we adopt. This is the silent battle that is going on.”

This article was published in The Fintech Magazine Issue #37, Page 12-13

The post EXCLUSIVE: “A Healthy Approach to AI” – Renato Rocha Souza, RBI in ‘The Fintech Magazine’ appeared first on FF News | Fintech Finance.