AI: The Smart Newcomer Poised to Remove Everyday Hassles

With AI able to function on many levels, from answering the most basic of bank customers’ questions to revealing complex behavioural insights, it is clear the technology is an invaluable business tool when employed efficiently. Here, two BNP Paribas industry experts discuss AI’s myriad use cases across all sectors.

From its beginnings in the mid-1950s through developments such as IBM’s Deep Blue chess champion and the board game programme AlphaGo, AI has moved out of the realms of science and advanced technology and is being “democratised”, says Su Yang, Head of AI, Transaction Banking and Head of AI and IT Innovation, BNP Paribas.

“Now with programmes such as ChatGPT, many more people are utilising AI-based products,” he continues. “People can see that some of the better products they are using are AI-based. As financial institutions and corporates, we need to understand that our employees are using AI in their daily lives; employees and our clients need access to these products.”

Yang notes that as with the word ‘innovation’, the definition of AI is “so vast that many different elements can be put into it, depending on who you are talking to. Today, AI is mostly data science – all the algorithms that can learn from data and detect underlying patterns and that can be reused for performing a serious task.”

Steven Lenaerts, Head of Global Channels and Digital Onboarding, BNP Paribas Cash Management, says there are two essential pillars of AI: technology and data. “The technology is intended to mimic human intelligence and draws from data in order to achieve that,” he explains.

Within the technology there are many different tools that can be defined as AI, including expert systems, ML, supervised and unsupervised training, game theory, neural networks, natural language processing (NLP) and robotics. In summary, says Lenaerts, any of these technologies can be embedded into a client’s business processes; clients need to be able to do something useful with the outcomes.

“If you want to define AI in a business-friendly sense, it is about mimicking human intelligence, using mathematical models to get there, and depending on the outcome that you expect, it might have a different name, but it all boils down to the same thing.”

In search of credible and sustainable data

However, to build the right models, “you can’t leave it all to the machine”, he points out. “AI should be enriched with human feedback and data-quality measures. That human input can begin with selecting the right data sources on which to draw. If the data is wrong, the outcome will be wrong. Data needs to be credible and sustainable.”

The value of people is well-evidenced in BNP Paribas’ work with a global company that manufactures lithography machines producing computer chips. AI was used to improve the company’s FX exposure forecasting, with the data scientist and core team looking at the data, analysing it and trying to identify patterns and trends. “A data scientist tries to see which data science or machine learning model will deliver the most accurate results,” Lenaerts continues. “We ran about 20-30 different models and found one that outperformed the others. It was developed in house, using existing algorithms that are on the market. The project was a good blend between the treasury team – which has the business knowledge and knows what data and characteristics should go in the model and the AI technology – and the data scientist. Having a good  data scientist who can tune and tweak the model until the best outcome results is crucial.”

Yang agrees that when it comes to the various elements of ML, data is “extremely important and very central to AI. If you don’t have data of good quality, or if it is labelled incorrectly, you will end up with the wrong results”, he warns.

To reduce the risks of the wrong outcomes or too much variation in results, AI users should ensure that their business case is well formulated. Yang continues: “Depending on the business case, you can pinpoint the specific AI technology and the initial conditions that will help reduce variation. Adopting a ‘secure-by-design’ methodology will also protect the bank and its corporate clients’ data.” Secure-by-design principles prioritise the security of customers as a core business requirement rather than as a technical feature.

Any discussion about AI will include the topic of generative AI (GenAI), adds Yang. This technology, which can generate text, images, videos or other data – typically in response to prompts – learns the patterns and structure of input training data and generates new data that has similar characteristics. The most common application of GenAI is chatbots.

“GenAI, such as ChatGPT, is an example of the fundamental AI research field meeting a more practical field – natural language processing – to produce something that is of high value,” says Yang.

AI is at the heart of BNP Paribas’ transformation strategy in all areas of transaction banking. In cash management and payments, the bank is applying ML to its fraud detection activities. The AI tool recognises outlier transactions and has enabled the bank to uncover frauds totalling millions of euros, helping corporate clients to both prevent and identify fraudulent transactions.

“BNP Paribas has invested in AI technologies for a while now and they are used extensively in the field of fraud detection,” says Yang. “Significantly, 80% of fraudulent transactions are now detected in real-time via a mix of rule-based and ML models, which were developed entirely in-house.”

On anomaly alert

Lenaerts notes that both cyber fraud and risks are on the rise and banks and corporations must be vigilant. “We all know the practices that we need to put in place, but it’s a constant race against the fraudsters. Having an AI tool in the background that monitors what is happening, which detects patterns and scores the patterns on fraud sensitivity, could enable a company to prevent fraud and ensure that the people who need to investigate are looking at very specific cases rather than all of the cases.”

This selectivity is based on pattern recognition of broad market data. “Obviously, the more payment data that you have, the better you can do it,” says Lenaerts. “But the data needs to be contextualised with the historical behaviour of the client so you can determine if it is a normal pattern or an anomaly. An exotic payment may not necessarily be a fraudulent one – it might be the corporation’s business.”

AI tools are also used to automate the extraction of information from some paper-based documents, which “a lot of corporations are happy to hear”, adds Yang. “With letters of credit, for example, around 80% are processed automatically. AI has helped in many other ways, too, to conciliate and process data from internal and external parties. This helps us to facilitate financial reporting for chief financial officers, for example.” The bank is also looking to apply AI to its KYC processes and financial forecasting.

It is crucial for a bank to pick the right use cases for AI, says Lenaerts. “Banking covers a very broad space and there are numerous use cases. In transaction banking, servicing is something that sets a bank apart from its rivals and here AI comes in, because it can help us to be more precise, more efficient, and to deliver better quality at a lower cost,” he says.

From customer insights to capturing needs

An “obvious” tool here are chatbot virtual assistants that enable a bank to filter common queries and respond without the intervention of a human. “You can go a step further, using AI to make service request analyses, which enable a bank to pinpoint the issues and priorities of the person making the enquiry. From that, you can optimise the routing of the query within your organisation, identifying the right individual to deal with it. As a global transaction bank, the organisation is large, and you want the right people to pick up on the right question and not have multiple handovers,” says Yang.

Banks such as BNP Paribas are confronted with a “sheer volume of information that needs to be processed and validated” and AI is playing a crucial role, Yang adds. “We can use AI to create customer insights, based on past performance and ensure that we are properly capturing their needs in order to fine-tune our offer in specific segments. The beauty of AI is that it can go from the automation of straightforward tasks through to expert rules where the tool is trained on advanced mathematical models to derive insight and knowledge.”

This insight detecting can, by the tone of a message, determine whether a client is unhappy. Whether the bank has done something wrong or not is irrelevant, notes Lenaerts. “What we want to do is control the reaction and ensure that we are proactive in reaching out to that client. Sentiment detection enables us to calibrate the action to the response. An AI tool can help us to filter through the hundreds of service requests that come in every hour and prioritise a request because it needs urgent attention.”

Corporate treasurers are adopting AI tools for cash forecasting, using advanced analytics on AR data. “Most corporate treasurers know the maturity date on accounts – it’s all about payment disciplines, centralisation, and orchestrating payment runs,” he says. “The part where most corporate treasurers are still struggling is AR.”

Training an AI model to look at the right data, such as historical payment behaviour, credit exposure and past payment behaviour, calibrated over time for a large number of clients on a rolling basis will improve forecasting in accounts receivable, says Lenaerts.

Don’t be afraid to experiment

“Forecasting is all about data analytics, so why not use AI? But [it must be] AI that is based on the proper data, including outstandings, payment terms, historical payment behaviour and financial health of the company in question,” says Lenaerts. By doing this, companies can optimise their liquidity and short-term investments, taking into account treasury policy and the instruments in which they can invest.

AI also can be used to train interest risk and counterparty risk models across FX, counterparty, and interest rate risks, helping corporate treasurers to better predict exposure.

In addition to risk and liquidity management, AI is finding a role in regulatory compliance, notes Lenaerts. “Just like a bank, a corporate – especially one that operates globally – is confronted with myriad regulations. It is a hassle to monitor all the regulatory evolutions and measure the impact they might have on the company and its daily operations.” This is where AI can assist, monitoring regulatory sources, detecting upcoming changes, and interpreting any impact.

When it comes to AI, Yang urges corporate treasurers not to be afraid to experiment. “The technology has arrived at a point where the time to market is extremely short. Corporations that know how to navigate through the innovation process will be able to differentiate themselves,” he says.