The Future of AI in Payments: Predictions for 2025 and Beyond

In 2023, the financial services industry invested an estimated
$35 billion in artificial intelligence, with the banking sector leading at approximately $21 billion. Over the past year, Artificial Intelligence (AIs) and Large Language Models (LLMs) have been transforming payments from simple transactions into critical touchpoints within the customer journey, serving as a foundation for what is to come. We spoke with Fouzi Husaini, Chief Technology and AI Officer at Marqeta, who is helping pioneer the shift from traditional credit and debit cards to versatile, fully digital payment instruments using these latest tools.

As Husaini explains, "AI plays a significant role in further enhancing the payment experience by enabling a high degree of customization."

As we look to 2025, the opportunities for AI are becoming more and more exciting. From hyper-personalized payments and fraud prevention to increased productivity and greater operational efficiency, AI's potential to redefine payments is starting to lay the foundations for the next decade. This blog explores Husaini's predictions on how AI could shape the future of financial services in 2025.

In 2024, AI's focus was on building bigger, better large language models (LLMs) in a competitive "AI race." But as Fouzi Husaini explains: "Now we've kind of plateaued to a point where there's a lot of questioning: is just adding more parameters better? The real value now lies in how we apply AI."

This shift is steering the payments industry toward pragmatic, application-layer innovations. Features like more convenient embedded finance solutions exemplify how generative AI can distill complex problems into smaller, actionable ones, reducing the unpredictability of AI outputs and “hallucinations”—a phenomenon where AI generates false or misleading outputs.

For example, picture a small fintech company that wants to help gig workers streamline their payments from multiple platforms. In our current gig economy, you might have drivers working with a company like Uber, Lyft, or Doordash. By using AI-driven embedded finance tools, this company can enable workers to automatically sort income into categories like savings, expenses, and taxes. With a simple dashboard powered by AI, these workers can visualize their earnings and optimize payment methods, ensuring every transaction maximizes rewards or minimizes fees.

In 2025, the focus won't be on who has the biggest LLM, but on the value and usability their models create.

As we mentioned above, LLMs were the focus of 2023-2024. However, the rise of agentic AI systems is quickly becoming the goal for many businesses going into 2025. Unlike generative AI tools like ChatGPT, agentic AI
is focused on
automating specific, highly targeted tasks, with the ultimate goal of streamlining workflows with minimal human input.

“We're going to see agentic AI systems—worker bees that are very, very specific in what they're good at, with a lot of guardrails, connecting almost like a workflow to complete a larger task," says Husaini.

Agentic AI is already transforming operations. By implementing workflows to streamline customer contact center tasks like fraud claims and chargebacks, companies are reducing operational expenses and improving efficiency across the board. These systems are also simplifying the payment industry’s complex software stacks, making legacy systems more manageable.

Imagine a
global payment processor that wants to integrate agentic AI to handle chargeback disputes. When a dispute is submitted, the AI can autonomously verify the claim by analyzing transaction data, timestamps, and flagged patterns from fraud detection models. It compiles evidence, generates a response, and updates the workflow for human review, all within minutes.

This targeted automation allows for quicker resolutions while ensuring accuracy, seamlessly fitting into the larger customer support workflow.

Fraud detection is one of the most critical applications of AI in financial services. With millions of transactions happening globally every minute, distinguishing legitimate activity from fraudulent activity is a classic "needle-in-a-haystack" challenge.

"The most obvious area to start is risk management, specifically fraud," Husaini explains. "Training models on what good and bad transactions look like to identify fraud in real-time is key."

This challenge is being addressed by a GenAI-driven infrastructure that integrates traditional machine learning techniques with modern AI systems to create a full-spectrum solution. In 2025, the opportunity lies in enabling AI to invoke specialized agents—such as machine learning models trained to detect fraud—and then take adaptive actions based on their outputs. For example, instead of relying solely on traditional validation methods like OTPs, the system could initiate additional verifications, including asking customers natural language questions for a more seamless and intuitive experience.

Taking our example a step further, imagine a neobank using a combination of traditional machine learning and modern agentic AI to protect customers from phishing scams. ML models can detect suspicious patterns, while agentic AI begins a series of actions based on that outcome: pausing transactions, alerting the customer, and initiating personalized next steps, like asking natural language questions for identity verification. This synergy enables real-time, effective interventions that prevent losses and build customer trust.

This is one of the simplest ways AI is changing the financial space, but it’s powerful. When every error results in revenue loss, having an automated system that can learn based on individual spending habits can be a game-changer for security and compliance.

Last year, a lot of the hype around AI stemmed from operational improvements and speed to market. This plays into the "AI race" Husaini mentioned earlier. Before, the focus was on having the biggest and fastest LLM. However, as we shift our gaze to better experiences, one of the more exciting utilities is the impact it may have on the end user.

Simply put, what do your customers see when they interact with your product? Today's consumers expect more than just a payment solution—they want an offering  tailored to their needs, habits, and unique interests. AI enables hyper-personalized payment experiences by analyzing individual transactions and customizing options like rewards and Buy Now, Pay Later (BNPL) offerings.

"If you can do customization down to the actual transaction... you can customize on so many dimensions—preferences, rewards, or Buy Now Pay Later options," says Husaini.

Let's say a digital wallet app wants to leverage AI to offer more personalized payment options for its customers. For a user who wants to purchase a $1,000 laptop but only has $600 in their account, the app might analyze past spending behavior and recommend splitting the payment into a three-month BNPL plan or offer to pull from future earned wages that they might not have access to yet. At checkout, the app could also suggest using a specific rewards credit card to maximize cash back on that purchase. There are so many occasions where customers have the opportunity to leverage a specific reward that they are completely unaware of.

This level of tailored suggestions can help users make smarter financial decisions, from maximizing their rewards cards to finding the best payment plan available.

The future of payments is no longer just about efficiency or innovation—it's about creating smarter, more human-centered systems that redefine how we interact with money. AI is the driving force behind this transformation, powering hyper-personalized payment experiences, streamlining complex workflows, and enhancing security in ways we couldn't have imagined, even just a few years ago.

But this isn't just about technology. It's about what technology unlocks for people—more secure transactions, simpler choices, and more time to focus on what truly matters. As AI shifts from being a novelty to a necessity, the focus will remain on solving real problems and enabling growth, not just for businesses but for everyone who relies on financial services every day.

The year 2025 isn't just a marker for what's possible in AI—it's looking to be a turning point for how technology can make our lives better, one payment, one transaction, and one innovation at a time.