Boosting the ‘intelligent’ in intelligent payment routing with AI

For those who have not come across the term, intelligent routing refers to a set of rules that banks use to dynamically direct payment transactions to the most appropriate payment scheme. These rules are based on the paying customer’s preferences, such as cost, timeliness, success rates, and other real-time conditions. The objective of intelligent routing is to ensure that the transaction is processed using the most appropriate payment scheme, improving transmission efficiency and speed, lowering costs, and increasing the likelihood of successful payment completion.

The payee also benefits from intelligent routing in terms of cost-efficiency (for example, use of the most favored foreign exchange option), reliability, and more. Intelligent routing routes payments to gateways that historically have better authorization rates for a particular payment type or provides an operating fallback if one payment gateway experiences downtime or fails.

As Gareth Lodge of Celent said at EBAday 2024, “We're seeing things emerging in smart routing. Not just a set of business rules, but much more dynamic based on pricing and time of day to incentivize customers’ behaviors.” The realization is that we now can benefit the entire payment chain by bringing innovations to payment routing beyond a basic set of rules.

How does AI come into play?

While the complex rules currently used in payment routing are efficient, they are manually implemented and relatively static once established. Introducing AI into the process has the potential to improve the timeliness and flexibility of routing, as well as enable dynamic consideration of more relevant processing parameters.

AI enables rules to be dynamically adjusted in real time based on ongoing data, continuously updating which payment gateway is best at any point in time. In addition, by analyzing historical transaction data, AI can predict which payment gateways are more likely to approve a transaction based on factors like geographical regulation, location, time of day, and customer behavior.

Further, AI can be used to analyze the total cost of a transaction, including fees related to cross-border payments, currency conversions, and interchange rates, and forecast which payment processor will provide the most cost-effective option in any scenario.

Overall, the business outcome is accurate and up-to-date routing of each payment through the most efficient payment rail for that payment. This, in turn, improves the customer experience by personalizing routing decisions based on customer preferences and behavior. Further, the improvement is ongoing as AI models continuously analyze customer transactions and routing decisions.

Achieving outcomes from your AI models: Four key strategies

Because AI technology continues to rapidly evolve, many of our clients opt to undertake a short proof of concept (PoC) before making a larger investment. These PoCs focus on finding and fine-tuning large language models (LLMs), evaluating the data they access, and defining the best prompts (the questions you ask the models). Creating a sandbox environment is helpful for these explorations.

Perhaps the most important strategy is data. AI models require a lot of high-quality data—the better the data, the better the model and its results. In many banks, this data is dispersed across the banks’ systems, so there may be some work to do to make data available centrally and ensure its quality.

Once created, the models need to be trained and regularly updated with the latest transaction data. Monitoring model performance is critical to ensure they continue to deliver accurate and beneficial results.

Key strategies for achieving results :

1. Conduct a PoC to fine-tune models and data
2. Create a sandbox environment for PoC exploration
3. Ensure access to high-quality, centralized data
4. Conduct ongoing model training and performance monitoring

Some precautions and regulatory considerations

As we might expect, using AI to enhance payment routing is not easy due to the sensitive nature of payment data and the complexity of payment systems. However, these challenges are not prohibitive. We simply need to do the right thing in the right way.

First, we need to comply with data regulations (e.g., the General Data Protection Regulation in Europe, the California Consumer Privacy Act in the U.S., and other relevant regulations for your bank’s jurisdictions) by ensuring data is used for its intended purpose. Personal data, especially, should not be collected or retained unlawfully.

In addition, we need to avoid discrimination in our models. In the case of payment routing, this may be less pertinent than in other areas, such as loan origination. However, we still need to ask the right questions to ensure all our customers are treated fairly.

Further, AI systems can be "black boxes," making routing decisions difficult to interpret. The European Banking Authority (EBA), the U.S. Federal Reserve, and others are increasingly concerned with how decisions are made by AI systems and whether those decisions can be explained. Routing decisions need to be traceable and explainable, especially when issues arise regarding the correctness of a payment route or its timing.

Finally, we should be aware that AI introduces new layers of processing complexity, which increases the risk of technology failure and cyber attacks. It’s important to implement robust risk management and AI systems testing to mitigate these risks.

Exploring the opportunities of AI-driven routing

AI can play a key role in intelligent routing, improving business outcomes for both banks and their customers. While the opportunities are promising, careful consideration of the whole ecosystem is required to ensure your business case is not overly optimistic or pessimistic.

CGI is collaborating with several banks on exciting AI initiatives, and we’re evolving our CGI All Payments solution to take advantage of this groundbreaking technology. If you’re interested in learning more about our AI work in payments or in banking overall, feel free to contact me.