Scaling AI in European banking: A practical framework for success

  0 Be the first to comment

Scaling AI in European banking: A practical framework for success

Contributed

This content is contributed or sourced from third parties but has been subject to Finextra editorial review.

This article has been co-authored by John Barber, Regional Manager - Europe, Infosys Finacle, Narasimha Prasad Nagaraja, senior director at Infosys Finacle, and Ramprasath Ganesaraja, head of AI research at Infosys Finacle.

A practical guide to reaching the promised land

In our previous article, Forget Fintech Disruption—Here's How European Banks Can Become the Disruptors, we explored the significant potential of AI and generative AI (GenAI) in the banking industry, highlighting the vast opportunities for efficiency, profitability, and enhanced customer experiences. However, we also identified several barriers that prevent European banks from fully realising AI's transformative power. These include siloed AI projects, a talent shortage in data science, legacy system integration challenges, and navigating stringent regulatory requirements. Despite significant investments and promising use cases, many banks struggle to scale AI initiatives beyond the pilot stage and unlock strategic value.

In this follow-up article, we will outline a practical framework designed to help banks overcome these obstacles and scale their AI and data projects successfully. While there may not be a one-size-fits-all solution for these nuanced problems, here is a helpful framework for banks looking to scale their AI and data projects—a four-step ABC framework: 

1. Ready your data

The first step is readying your data—the building block of any data project.

Analogue to Digital: The ‘A’ of readying data is converting data from analogue to digital to ensure no relevant information remains outside the system on paper or in other non-digital formats. Enterprises are primarily successful in digitising their important data—on customer profiles and behaviour, transactions, interactions, and others, making it available for processing.

Business Context: While digitisation is essential, it is insufficient to scale data projects. Unless there is a business context, data, on its own, is of little use and may even be misleading. Metadata such as the location of a purchase transaction, the channel of customer interaction, and other qualifiers, ranging from purpose to persona to market circumstances and so on, gives data its context. Enriching data with context is necessary to build an accurate AI model.

Curation: With ‘A’ and ‘B’ in place, banks and FIs should turn to curation, the ‘C’ of preparing data. This ensures that data is free from anomalies, duplication, information gaps, and biases. At the end of this process, organisations can be confident they have a sound base for their data projects.

2. Outline your use cases

When outlining use cases, financial organisations take note to ensure that A, B, and C impact one or more of the following important business metrics: revenue, profitability, and experience.

Automation: Automation is a proxy for operational efficiency that impacts the business's profitability. The use cases, which should enhance effectiveness and efficiencies, are usually based on optical character recognition (OCR), chatbots, robotic process automation (RPA), and others.

Business Transformation: Business process and business model transformation use cases eventually impact revenue. The use cases typically leverage machine learning (ML) techniques, falling into one of the following four categories: regression, classification, clustering, and forecasting. Some use case examples include credit underwriting, cash flow forecasting, customer churn analysis, and "next best" recommendations.

Customer Experience: Other use cases rely on hyper-personalisation to create seamless, engaging banking experiences. AI, machine learning (ML), and big data analytical methods process various data, including profiles, preferences, risk appetite, financial goals, and others, to refine customer understanding and recommend personalised products and services for individuals.

3. Prepare your organisation

Organisational preparedness is as crucial as data-related readiness to implement projects at scale.

AI Platform: The right AI platform is imperative for democratising and scaling use cases in data projects. The underwhelming performance of AI projects can be attributed to one of five reasons: the absence of a single platform that can deal with everything, from design to deployment; the need for niche skill sets; non-explainability of AI output; concerns related to data privacy; and post-deployment issues such as data drift and concept drift. Banks must choose the right AI platform that can largely address these concerns.

Basic Skills Redefined: Data projects demand digital skills quite different from the traditional skills used for creating "function" and "control" banking products, such as loans, deposits, or anti-money laundering and multi-factor authentication solutions. Financial institutions need to upskill or reskill their employees or hire people with the necessary expertise. Apart from purely technical skills (data science and engineering, for instance), organisations should cultivate skills such as problem-finding and framing, empathy, and creative thinking.

Culture: Creating a culture of data is a significant part of organisational readiness, as it drives new ways of thinking about how data is created, stored, and handled to maximise its usefulness. For example, banks must learn to visualise the art of the possible, understand the most influential variables for a particular prediction or outcome, and learn to handle data project life cycles.

4. Gear up for rollout

The final step in the framework is rolling out the data project. However, a critical task remains: finalising the decision-making authority—AI or human? There are several modes of operating available here:

Augmented Mode: In this mode, bank staff or customers oversee the decisions, whereas AI is the supporting actor assisting them. Consider loan approval, for example. While the loan officers make the final decision, they are supported by AI, which extracts, synthesises, and summarises information from the various documents to provide the insights that help make those decisions.

Bridge Mode: In this mode, AI can execute identified processes, which are reviewed by a "human in the loop" who takes the final decision. This is the bridge between the augmented mode and the centre stage mode.

Centre Stage Mode: Here, AI takes centre stage and is responsible for executing workflows, processes, or transactions in all but exceptional cases where human intervention is required. Returning to the loan approval example, in the centre stage mode, AI independently processes loans for a predefined amount for the respective client segment and refers only a few exceptional or abnormal applications to bank staff for a decision. The use cases can be selected and deployed based on banks' risk management policies.

Applying the framework to an implementation use case

Now that we have the 4-step ABC framework, let’s see how if they answer questions posed by bankers.

How can we measure the effectiveness of cash flow forecasting, and is it possible to explain the outcome?

Banks can measure the effectiveness of cash flow forecasting by tracking metrics like the following:

  • Mean Absolute Error (MAE): This metric calculates the average of the absolute differences between the predicted and actual values. It's easy to interpret and less sensitive to outliers compared to some other metrics.
  • Mean Squared Error (MSE): Squares the differences between predicted and actual values, then takes the average. It penalises larger errors more heavily but can be sensitive to outliers.
  • Mean Absolute Percentage Error (MAPE): This metric expresses the error as a percentage of the actual value. It's useful for comparing forecasts across different scales, but it can be problematic for values close to zero.
  • Root Mean Squared Error (RMSE): Similar to MSE but takes the square root of the average squared differences. RMSE is in the same units as the original data, making it easier to interpret the magnitude of errors.
  • Mean Scaled Error (MSCE): This metric compares the forecast errors to the errors of a naive benchmark model (often a simple average of past values). It helps assess the improvement the model provides over a basic approach.
  • R-Squared (coefficient of determination): This statistic shows how well the forecast explains the variance in the actual data. It ranges from 0 (no explanatory power) to 1 (perfect fit).
  • Forecast Horizon: Accuracy often decreases as the forecasting horizon (how far into the future you're predicting) increases. Evaluate the model for different horizons relevant to your needs.

By using a combination of these metrics and considering the specific context of the forecasting problem, one can gain a comprehensive understanding of your model's effectiveness. Explainability of AI models can explain the outcomes.

Examples of explainable AI (XAI) techniques in forecasting:

  • Feature Attribution Methods: These techniques assign importance scores to different features based on their contribution to the final prediction.
  • Counterfactual Analysis: This approach allows you to see how the forecast would change if a specific feature's value were different. This can be helpful for understanding the model's sensitivity to changes in the data.
  • Visualisation Techniques: Creating charts or graphs that illustrate the relationships between features and the forecast can improve understanding of the model's behaviour.

Benefits of XAI forecasting models include improved model performance, by understanding how features influence forecasts, you can potentially improve the model by focusing on the most relevant features or addressing biases; better decision making,  XAI can help users make more informed decisions based on the forecasts, by understanding the reasoning behind the predictions; and risk management, to identify potential weaknesses in the model through XAI allows for better risk management and mitigation strategies.

Do we need to have specialised AI talent to build these models and leverage AI?

Rather than focusing on AI models, consider democratising the ML lifecycle i.e., which is making the process of building, training, and using machine learning models accessible to a wider range of people, not just those with extensive data science expertise.

The right AI platform can also help you build models easily. Consider the following while looking for an AI platform that meet your needs:

  • Data democratisation
  • AI ML agnosticism
  • Model transparency
  • Model explainability
  • Data privacy controls
  • Continuous monitoring and adaptability

Not a panacea

The technology, however, is not without its challenges. Gartner’s classification of GenAI in its hype cycle is especially telling. While expectations on the outcome of use cases of Gen AI are high, the technology is not fully tested to see if it'll be a complete success down the road. There is also the unpredictability that comes with the complexities of large language models (LLMs) which can be subject to regulatory scrutiny, hallucinations—where models are prone to make up information and lead to inaccuracies, and growing concerns about magnifying existing biases and preventing financial inclusion. Europe, for its part, has a strong regulatory framework governing AI development.

While GenAI has potential, it is important to understand that realising business benefits is a journey for banks and not an overnight transformation.

Despite its challenges, the journey is worthwhile

The potential of AI and GenAI to transform the banking industry is undeniable.

Europe is advantaged by its rapidly developing AI landscape and its human capital. With the winds of AI innovation blowing, some banks are cautiously dipping their toes in. However, simply investing in these technologies is not enough. European banks and banks worldwide must adopt a strategic approach to unlock their full potential.

The four-step-ABC framework provides a practical roadmap for banks to navigate this journey. With a data-centric mindset and a commitment to building a robust data infrastructure, identifying high-impact use cases, preparing organisations, and ensuring a smooth rollout, European banks can unlock the true potential of AI and GenAI. This will not only bolster their bottom lines but also empower them to deliver exceptional customer experiences in a rapidly evolving financial landscape.

By embracing AI with a clear vision and a commitment to ethical practices, European banks can unlock a new era of growth, efficiency, and customer-centricity, becoming the disruptors instead of the disrupted. The time to act is now.

Comments: (0)

/ai Long Reads

John Barber

John Barber Vice President and Head at Infosys Finacle

Scaling AI in European banking: A practical framework for success

/ai

Hamish Monk

Hamish Monk Senior Reporter at Finextra

What is automation in financial services?

/ai

Hamish Monk

Hamish Monk Senior Reporter at Finextra

Understanding AI for personalisation

/ai

Sponsored

This content has been created by the Finextra editorial team with inputs from subject matter experts at the funding sponsor.