Bank of Cyprus' George Kousis: We are already seeing significant and quantifiable results from AI

"I don’t think there is a limit to the types of decisions that can be entrusted to an algorithm," George Kousis, Executive Director, Technology & Operations, Bank of Cyprus, suggests.

In a recent interview with GOLD magazine, Kousis elaborates, "If you trace the history of using algorithms to replace human action, there was always a level of mistrust whenever a new technology was introduced but eventually technology matured to levels where decisions taken by algorithms were better than human and became the de-facto solution. I think the same will apply with AI algorithms once we overcome some key constraints."

Among other things, Kousis also talks about the specific AI tools the Bank of Cyprus has adopted and shares his expert view about which AI trends and technology are worth watching closely.

What are the specific AI tools that you have adopted? How well did they integrate with your existing infrastructure? Did they require specific customisation to work effectively?

In general, we follow two broad directions in onboarding AI tools. The first approach is utilising tools and applications with embedded AI technology to meet our business needs, while the second involves developing custom applications and integrating them into our existing channels and processes.

Examples of the first approach include a Cybersecurity Network Anomaly Detection system that utilises AI to detect and block suspicious activity and the Office CoPilot solution that utilises the same technology as ChatGPT to support employees’ productivity.

Custom development examples that we have integrated into our processes include Document and Image Recognition which, using AI, enables us to extract useful information from documents and images that would otherwise require manual effort.

Moreover, we are now deploying an AI Digital Assistant that can answer any question related to our internal processes and products. As a first step, this assistant is used by our employees but we are working in parallel to develop a Customer Digital Assistant that we believe will revolutionise the customer experience.

Have these AI deployments led to revenue growth, cost savings or improved customer retention? Can you quantify these improvements? And how long will it take before you see a measurable return on your investment?

We have already implemented and have plans for use cases across all four main pillars: Cost and Process Optimisation, Boosting Employee Productivity and Creativity, Enhancing Customer Experience and Revenue Growth.

We are already seeing significant and quantifiable results in the first two pillars. AI Document and Image Processing has been integrated into key processes like Customer Onboarding and Transaction Processing, saving hundreds of hours of manual labour per month and, most importantly, accelerating customer response times.

Another example is Generative AI technology that we have used to understand and extract information from existing documents that would otherwise require someone reading lengthy and complex documents.

Adding to the impressive time savings, we are also observing improvements in the accuracy of the extracted information. Given that a lot of the investment costs will come down as we build and reuse the same technology for different scenarios, we expect the ROI to further improve and fully justify our investment.

What have been the biggest bottlenecks in integrating AI solutions? How did you overcome these challenges?

General technical limitations can be managed primarily through a cloud-first approach but you still need to have strong technical skills to synthesise solutions and integrate them into your business processes. So, other than formulating a clear AI Strategy, acquiring technical skills must be the first step on an AI journey. The challenge is that talent is scarce and real-world experience is even more difficult to find. We are addressing this challenge through selective hiring and an internal skills development programme to scale up our capacity.

As AI continues to transform the workplace, how has Bank of Cyprus adapted? How have employees responded and what steps have you taken to support this transition?

Employee resistance is best addressed through a structured Change Management Programme. Such a programme includes training for fundamental skills and raising awareness on the risks associated with AI. For example, we are training personnel on how to write appropriate prompts and interact with follow-up questions to our Generative AI Assistant. We also teach them about their responsibility in assessing the answers and validating the response using the references provided by the chatbot. Finally, we reenforce a culture of continuous feedback, so we can tune the AI Solution using the feedback provided by our people.

In your industry, what types of decisions can realistically be entrusted to an algorithm? And do you think that AI will ever be capable of handling the novelty and ambiguity of executive-level decision-making?

I don’t think there is a limit to the types of decisions that can be entrusted to an algorithm. If you trace the history of using algorithms to replace human action, there was always a level of mistrust whenever a new technology was introduced but eventually technology matured to levels where decisions taken by algorithms were better than human and became the de-facto solution. I think the same will apply with AI algorithms once we overcome some key constraints.

The most important consideration is fairness. Because of the way some categories of AI technologies work (like Generative AI), they rely heavily on the data used to train the models. However, this data may hide past biases that may lead to perpetuation of unfair treatment of customers (e.g. gender, race or localised biases). Another challenge is transparency, i.e. how you explain the algorithm’s reasoning for the decision. For both examples, the revolution in the technology triggered by ChatGPT also caused concerns because these two capabilities were missing. But technology is progressing exponentially and already major AI Service providers offer good capabilities to help organisations such as banks to test their models for fairness and transparency.

Additionally, regulation is also adapting to these challenges and the EU AI Act dictates what tests and controls need to be performed, depending on the risk level of the decision to be taken. As a final thought, we must also consider the CoPilot aspect of AI. For many scenarios, AI will propose a decision but the final call will be made by a human.

Which AI trend or technology are you watching closely and which do you think are more marketing fluff than substance?

Agentic AI will be the next evolution of AI technology by moving beyond data processing to systems capable of autonomous goal-setting, decision-making and adaptive behaviour. Agentic AI proposes the creation of small and specialised AI Agents that can proactively interact with their environment, learn from outcomes and interact with each other to pursue complex objectives. We see the emergence of several frameworks that describe how such Agents can be developed, orchestrated and tested. We really think that this trend will mature very quickly and will become the mainstream in AI solutions.

What is the one AI breakthrough you believe will be a game-changer for your industry?

I have already noted that, in my view, Agentic AI will be the next step in the evolution of AI technology. For now, we are planning to deploy AI solutions that will be directly visible to the customers like Digital Assistants. We are very careful with our approach and are rigorously testing these solutions to ensure that our customers will enjoy a seamless, productive and secure experience.

This interview first appeared in the April edition of GOLD magazine. Click here to view it.