Digital payments are growing. According to IBS Intelligence, in the financial year 2023-2024, they grew 44.3% in volume and 16.4% in value. And as digital payments grow, they become more complex and more appealing to fraudsters.
New AI technology is transforming how fintechs work with digital payments, from automated customer personalisation to increased operational efficiency. And, of course, fraud prevention.
So, let’s explore how AI has affected the payments sector in the past, how it’s being used now and how we expect it to change in the future.
A brief history of AI in the payments industry
AI in payments is not new. It’s been around for decades. As far back as the early 1990s, payment companies used machine learning tools – an early precursor of the LLMs we call AI today – to detect card transaction fraud. Simple rule engines evolved into more advanced neural networks that could detect fraud in real time. Then, in 1996, FICO filed a patent for Reason Reporter, an early form of explainable artificial intelligence (XAI).
The success of machine learning and AI in fraud detection led to widespread innovation throughout the industry. The technology was soon adopted in sectors like algorithmic trading, credit scoring, risk management, personalisation and forecasting.
Even three decades ago, machine learning was more efficient at dealing with big data than humans. Over time, digital payment volumes increased. Fully digital payments became the norm. And fraudsters became more advanced. So machine learning-derived technology improved too, quietly powering the new way we interacted with money.
Then, in 2022, OpenAI’s ChatGPT came along and AI was thrown into the limelight.
AI payments: where are we now?
As of December 2024, according to PYMNTS a massive 71% of financial institutions use some form of AI or machine learning for fraud detection. That’s up from 66% in 2023. In the US and Europe, according to McKinsey, 9 out of 10 people say they made a digital payment in the last year.
More and more payments are happening digitally, which makes them very appealing to fraudsters. Payments are also becoming more complex. This means old-style rule-based and machine-learning fraud detection tools are no longer enough. While their algorithms can still be effective with enough processing power, they require too much human intervention to be efficient.
Fraudsters are also using increasingly powerful AI systems to circumvent fraud prevention technology.
Increased liability for banks
Liability on banks to compensate fraud victims is increasing, with the UK and Singapore leading the way. Since October 2024, UK banks must refund fraud victims up to £85,000 within five days.
In December 2024 Singapore launched a Shared Responsibility Framework (SRF) which requires banks to implement cooling periods, provide real-time notifications for suspicious and high-risk activities, provide customers with a channel to report and block transactions and provide real-time fraud surveillance.
The EU is currently working on its regulatory framework that expands fraud liability for payment service providers (PSPs), as well as online platforms and telecom operators. This will cover a broad range of scams, not just payment fraud.
The rest of the world is currently playing catch up. For example, in the US, liability is capped at $50 unless it’s reported in 60 days. There’s currently a push to adopt a similar reimbursement model to the UK, but no concrete plans. Several countries in Asia have plans to introduce increased liability, but most are still in the early stages.
So, any companies that deal with payments must look to a future where they shoulder more responsibility for payment fraud. And that means we’ve seen another step change in how AI is used in payments.
Generative AI fraud detection
A new generation of AI fraud detection solutions has emerged. Using generative AI (the same technology behind ChatGPT and other consumer chatbots), they can analyse patterns and spot anomalies in real-time, across vast numbers of transactions. And, of course, they learn to spot new fraud patterns as they emerge. So they’re constantly improving their own efficiency. Systems like Salve Bridge, Feedzai and HAWK:AI are becoming the de facto standard.
Mastercard recently launched a generative AI fraud detection model that scans a trillion data points every year, and has an average of 20% increase in detection versus other systems. It can be as much as 300% more accurate.
Other uses for generative AI in payments
While fraud detection is responsible for pushing AI technology forward, it’s not the only use for AI in the payments industry. There are other areas where generative AI is transforming the sector.
Generative AI in customer support
Current-generation support bots have improved significantly over the last few years. We’ve reached the point where they can deal with most basic problems quickly and efficiently. This frees up human agents to solve more complex problems that require intervention.
Generative AI-based search functions also provide a powerful way for customer support agents to access vast internal knowledge bases. Because these repositories have been built over years and decades, finding specific information can be difficult.
AI search can understand the whole database and let agents use simple, natural language queries to find what they need.
Generative AI in underwriting
Generative AI is also used in the underwriting process for digital payments. AI and machine learning systems have been used to analyse data and predict risk levels for years. But, they require huge amounts of data to train on and, often, there isn’t enough available. Specifically designed generative AI tools can create vast amounts of synthetic data for training. This means the analytics tools can be significantly more accurate.
AI in payment processes
Payments are increasing in volume and complexity. So, the processes needed to facilitate them are becoming more complex too. Generative AI can be trained to route payments efficiently. It can be used to automate processes like invoicing, too.
AI for data analytics
Enormous numbers of digital transactions produce an equally vast amount of data – and AI is the perfect tool to analyse and understand it. AI tools can pull out information on things like payment trends, transaction values and customer behaviours and present it in a simple, understandable format quickly and efficiently.
Using AI to reduce overheads and boost profits
Combining payments-specific tools with more general business AI tools lets payments companies reduce overheads and increase profits significantly. Buy now pay later (BNPL) giant Klarna has partnered with Open AI to use generative AI to reduce its workforce. In 2024, Klarna cut its workforce by 24%, from 5,000 employees to 3,800. The next target is to reduce employee levels to 2,000.
One of our clients is VP of Engineering at a regtech/compliance fintech, and AI has sped up their development pipeline significantly.
‘We use GitHub Copilot in our front-end development workflow quite often. In the fast-paced world of RegTech, efficiency and precision are critical, and Copilot helps my engineers accelerate development without compromising quality. It autocompletes React components and can suggest Tailwind classes. It also streamlines repetitive tasks so our team can focus on building secure, compliant, and scalable solutions. While it may not handle audits just yet, it certainly helps us stay ahead of regulatory demands.’
AI in payments: what’s next?
So, if the AI technology used in payments is so powerful already, what’s next? Industry experts agree that while AI and LLM tech hasn’t exactly plateaued, the explosive growth we’ve seen over the past few years has definitely slowed. Increased regulation in Europe and the US means that key innovators have more challenges to navigate.
Relatively unregulated development in China continues apace with models like DeepSeek and Manus, which is claimed to be a big step towards artificial general intelligence (AGI).
For the payments sector, this means that the next few years will focus on better, more efficient ways to apply AI technology. Smart fintechs will distil large, complex problems into more manageable ones that smaller, more specifically-trained LLM and AI models can process. This will help reduce unpredictability and hallucinations in AI models.
For example, customer-centric embedded AI tools could automatically sort payments into appropriate categories (such as savings, expenses, taxes and disposable income) to help manage funds.
Agentic AI in payments
After the scramble for generative AI supremacy in 2023 and 2024, we expect to see a shift in focus, both in the payments and wider fintech spaces, to what is known as agentic AI. Rather than a single massive AI model that performs any task moderately, agentic AI tools automate very specific tasks accurately and efficiently.
Several agentic AI models can be linked together, controlled by another AI model, to automate complex workflows. Such AI chains are already being used in customer service situations.
Chargeback claims
Agentic AI solutions can autonomously verify chargeback claims from customers. They can analyse transactions, timestamps and flagged patterns from fraud detection models autonomously and without human intervention.
This frees up human resources to complete the final checks and action next steps much faster. It’s a more accurate and efficient way to process claims than a fully human workflow.
Fraud detection
The same workflow-based technology, using an array of AI tools to complete specific tasks, can also be applied to fraud detection and prevention. A generative AI can ‘oversee’ a suite of specific agentic AI tools, invoking each one as a situation needs it.
So, a machine learning tool might detect a suspicious payment pattern, based on its knowledge of a customer’s payment history. This sets a chain in action – the payment is frozen, and the customer and payment provider are alerted (and asked natural-language questions to verify the payment if appropriate). All without the need for a human agent.
This will result in more fraudulent transactions being identified accurately. But it will also mean fewer false alarms. And as the systems continue to learn, accuracy will increase further.
Personalised payment solutions
One of the key changes we expect to see in an AI-driven payment future is a much clearer focus on providing the best customer experience. One way to achieve this is through super-personalised payment solutions.
So, an AI agent could monitor customer payments. When a customer with limited funds in their account attempts to make a large payment, they could be directed to an appropriate buy-now-pay-later (BNPL) option.
Or, a customer could be directed to a specific reward credit card or cashback solution they already hold or are eligible for, to make sure they’re getting the best value for their purchase.
This level of tailored solutions would be near impossible with a human workflow, but AI agents can make it a seamless experience for customers. It can help them to make smarter financial decisions and get the most from their money quickly and easily.
Instant payments
Instant payments are a fintech buzzword at the moment. Some areas – like most of Asia, Brazil and India – already use instant payments. And that means the rest of the world needs to catch up, while staying safe and compliant.
According to Mastercard, the rise of real-time payments in the Asia Pacific region has led to a new wave of fraud. Because transactions are instant, fraudsters can move funds extremely quickly between many accounts, making them difficult to detect. To combat this, they have launched a new AI tool called Trace.
Trace is designed to track many transactions across a complex payments network – going beyond a single firm’s limited reach. It’s already been rolled out in the UK, where it’s been used by 21 firms, covering 90% of the Faster Payments Service network.
AI tools are also being developed to drive the processes needed to complete vast numbers of transactions instantly and safely. Agentic AI workflows are being developed that integrate machine learning fraud detection algorithms as an integral part of the process, significantly reducing risk.
Conclusion
Since the launch of ChatGPT in 2022, we’ve seen incredible growth in the use of AI. The technology has evolved at a spectacular rate too. If, as has been suggested, that innovation has slowed now then the next phase of AI innovation must be a distilling of that technology to create smarter, more efficient tools.
So, we can expect to see incredibly fast, powerful fraud-detection tools to stay ahead of the fraudsters. But we’ll also see a move away from catch-all generative AI, towards highly-specific agentic AI tools.
The real innovation will not be squeezing an extra few per cent out of generic AI tools. It will be creating bespoke, efficient agentic AI toolkits to automate complex workflows and free up human resources.
If you need the best fintech talent to work on your AI solutions, or your an AI expert looking for your next role, EC1 Partners have you covered. Get in touch to talk about what you need.
