Five Ways Generative AI will Change the Payments Industry

The payments industry makes extensive use of AI/ML technologies in a wide range of applications – including targeted marketing, process automation and customer profiling. The use of AI and ML in fraud prevention is particularly common, as it enables fraud prevention teams to scour large amounts of transaction data to identify fraudulent behaviour and identify emerging threats.

There are many ML methods utilised across the payments industry, however, there has not been an established clear leader. That may be about to change with the emergence of generative AI, and many experts believe the use of AI and ML in payments could be about to make a giant leap forward in terms of performance with the implementation of this technology.

Generative AI is a hugely complex piece of technology, utilising massive deep neural networks, composed of billions of parameters to enable complex pattern recognition, and the early signs are that it could be a game changer. OpenAI has commercialised the first massively trained generative AI, GPT-3, which Elon Musk called ChatGPT “scary good” and warned, “We are not far from dangerously strong AI”. He may have called it correctly.

GPT-3’s model is trained on an astronomical amount of data - everything from books, internet articles, social media, and source code. It can also produce its own content in the style of what it has learnt after being given a simple prompt. For example, the AI can be prompted to write technical documentation based on provided code, or it can answer technical questions as well as translate languages (both human and programming languages). This type of ability may not have an obvious use case in payments or fraud prevention, but generative AI offers a myriad of solutions to complex fraud detection, data mining and solution development challenges.

What could this technology be capable of in the fraud and payments industry? Here are five ways Generative AI will change the payments industry.

Decision support - There are numerous tools in use today that assist with decision support. Dashboards are particularly popular, providing a lot of information in one place. The downside is that most decision dashboards only show what they have been designed to show, and many can’t quickly change enough to demonstrate the latest data trends. This can mean users miss critical information which could be useful in providing a more detailed picture for decisioning.

Generative language models could offer a solution by enabling decision makers to simply request the information they need, and have it presented in a form they can easily digest, along with supporting evidence to help make the best decision. The generative models can also react to live data and continually update dashboards for new, or fast-moving trends, enabling even faster decision making.

In real time payments this will enable systems to react faster to large scale fraud attacks, which are often performed against new payment methods. A generative model could also be used to determine how a range of payment methods are likely to be used in a particular geographic region, or by a specific customer group, indicating where further investment may be required.

Knowledge base search. Most of an organisation’s knowledge is stored across many different sources and locations. From documentation, emails, work tickets and source code, there is a lot to go through and can be extremely easy to miss important information – something a large language model would be able to assist with. Providing a detailed summary with references and contextual information to enable a researcher to get what they need very quickly, all from a simple prompt.

A developer will soon be able to ask the AI to ‘provide all internal information for example payment method integration’, which will enable them to quickly understand how previous payment methods have been integrated, and how they can integrate new methods more rapidly.  

Code Generation for Product Prototyping. With a simple prompt, GPT-3 can produce sample code for almost any scenario. Although the full capability of this is currently a way off, in the future we could see Generative AI producing prototype products or features extremely quickly. Developers could then step in to add additional, more complex features. The AI can then take feedback and improve on its previous response, as it has some ‘memory’ of what has been previously asked, to better serve a follow-up request. Ultimately this could lead to developers defining very specific functionality, perhaps by providing a functional requirements document - and the AI would produce the code, create a test environment and deploy the prototype for rapid testing. This could ultimately lead to much simpler payment integrations, enabling rapid time to market for those features.

GPT-3 is also a great tool for assisting with finding bugs in code and offering solutions on fixes. This feature, if implemented into development environments, could rapidly improve both code quality and reduce time-to-market for products and features. This remarkable ability will completely transform software development by massively reducing product development time.

Product support automation. When I asked GPT-3 how it could provide support functionality, it had the following response: 

“As a large language model trained by OpenAI, I am able to understand and respond to a wide range of questions and provide information on a wide range of topics. This makes me well-suited to be used as a support function in a business, where I could be used to help answer customer questions, provide information on products or services, and assist with a variety of other tasks. For example, I could be integrated into a company's customer support system, where I could help to quickly and accurately answer customer inquiries, freeing up human agents to handle more complex issues.”

GPT-3 clearly understands where its uses are. Customer support teams would benefit massively from this AI and could expect to see their efficiency improve dramatically, as the AI could handle most customer interactions, leaving human teams to work on complex cases, or to train and provide more assistance elsewhere in the business.

Fraud detection. When prompted about how GPT-3 can be used for fraud detection, it came up with the following:

“One potential application of GPT-3 in payments is in the area of fraud prevention. The model's ability to process and understand large amounts of data, as well as its natural language processing capabilities, could be used to identify and flag potentially fraudulent transactions in real-time.”

Applying GPT-3 to fraud could yield revolutionary results. The AI could be trained with massive amounts of historic payment information, allowing it to learn how individual cards are usually used, as well as providing analysts with a view of current fraud trends, but this is only the start of its potential. The model could help fraud analysts by filtering away low-likelihood fraud alerts, reducing manual review effort.

Generative AI tools could prove to be an excellent addition to a fraud defence team’s tool kit, providing the capability to interrogate data with human questions as opposed to database queries. For example, fraud managers could ask the AI to summarise all suspicious payment behaviour – a task that currently takes up most of a fraud manager’s time. This could then be taken a step further with fraud managers working with generative models to develop new fraud rules and to apply ML models simply by asking it how the newly developed rules and models would perform and to suggest improvements to boost performance.

The same generative model could be used for gaining business insight, such as understanding when a customer’s behaviour changes, or for mass behaviour changes and providing possible reasons why – all from just a simple prompt.

Generative AI offers us a glimpse into the future of ML fuelled payments, with extensive uses in fraud detection, real-time payments, feature development (particularly integrations) and data analysis. GPT-3 is just the start of what will, undoubtedly become the AI revolution.