This year, at least three financial companies—the neobank Revolut, credit fintech Affirm, and payment network Mastercard—have announced the launch of their own AI models instead of adopting solutions from third-party developers. AI is no longer just an operating expense; it is becoming a distinct asset class for financial companies. But when, and under what conditions, could this increase the value of their businesses?
AI models as a new asset
At the end of August this year, Revolut unveiled its own AI model, PRAGMA, and announced the creation of a dedicated unit, Revolut Research, to develop it. The neobank's AI analyzes sequences of customer banking events to perform credit scoring and detect fraud, as well as recommend the products best suited to each customer.
PRAGMA was developed in collaboration with Nvidia. The model was specifically designed to decipher complex financial behavioral patterns and assess risk in real time. According to Revolut, tests on historical data showed that PRAGMA was 2.3 times more accurate at assessing a customer's risk of default, helped detect 65% more cases of fraud, and improved the relevance of recommendations by 41%.
Revolut co-founder and CEO Nik Storonsky told Reuters that the company uses 30–40 million transactions a day to develop its own AI models.
It is not the only neobank doing this: in March 2026, payment network Mastercard unveiled its new generative AI model, trained on billions of anonymized transactions.
In September 2026, US credit fintech Affirm also announced that it had launched its own AI model, which assesses customers' creditworthiness more accurately—including that of people with limited credit histories—and enables it to approve loans for them. The previous scoring system rejected some applications from such customers.
Just an expense or a strategic investment?
For banks and fintech companies, the main benefit of using AI is that it makes the risk assessment process less expensive, and therefore lowers the cost of serving customers. This is important for Revolut, for example, as the company plans to go public and, before then, increase its valuation to $150–200 billion.
Clint Suckermane, founder of The AI Value Institute, wrote that financial institutions should stop treating scalable AI initiatives solely as expenses and start viewing them as strategic investments in business development.
But what does this mean for investors in these companies? How the market responds to companies' plans and the actual implementation of AI remains insufficiently studied, says Zihan Tu, a researcher at the University of Warwick's Department of Economics. In his study, published on September 22, 2026, he analyzed announcements by S&P 100 companies about their use of AI in 2023–2024 and assessed how they were associated with short-term market reactions. He found that announcements of actual AI implementation generated 4.6 percentage points more abnormal returns on their shares (above what could be expected given overall market movements and other risk factors) than announcements of intentions to implement AI.
In an earlier study published in 2025, McKinsey estimated that banks that are among the first to adopt AI could increase their return on tangible equity (ROTE) by up to 4 percentage points by transforming their business models and creating additional economic value. By contrast, banks that are slow to adopt AI are likely to see their profits decline over the long term.
What investors should look at
Investors are unlikely to be willing to pay a premium simply because a bank or fintech has its own AI models, Antonis Dragiotis, a Greek expert in risk management, artificial intelligence, and banking technology, and coordinator of the AI Catalyst financial services working group, wrote in a comment to Oninvest.
He says they will need evidence that deploying such models will deliver better financial results. Moreover, as more and more companies adopt AI models, simply having them will no longer be a significant competitive advantage.
International financial technology expert Vadim Ivanenko notes that if an in-house AI model enables a bank or fintech to reduce the cost of serving a customer by, say, 30%, cut losses from fraud, improve the quality of its loan portfolio, increase conversion for financial products, or improve customer retention, that represents real economic value.
He recommends that investors in companies that have announced the introduction of their own AI models or tools analyze the following metrics before and after implementation:
— the operating expense-to-income ratio (cost-to-income ratio) and the cost of serving one customer or processing one transaction;
— fraud losses and the proportion of false positives;
— the cost of risk, credit losses, and the quality of credit scoring models;
— customer lifetime value (CLV), customer retention, conversion, and revenue per customer;
— additional operating profit and the return on AI investment, taking into account infrastructure, development, monitoring, and maintenance costs.
It is important to compare these metrics to distinguish the effects of AI implementation from other factors, Vadim Ivanenko adds.
Independent stock market analyst Vadim Kondratyuk adds that a valuation premium is justified only if AI increases a company's ROTE or reduces its cost of equity—that is, if the market starts to consider it more reliable.
He recommends that investors watch three metrics: changes in the cost-to-income ratio, the cost of risk over the credit cycle (whether it falls thanks to better borrower assessments), and revenue per customer. The latter should grow faster than it does at competitors.
In Revolut's case, the introduction of its own AI model has not yet shown up in its financial statements. In the first half of 2026, Revolut Holdings Europe's net operating income rose 36.8% year over year to €1.24 billion, while operating expenses increased 36.5% to €1.055 billion. Expenses as a share of income fell from 82.9% to 80.2%, but credit losses nearly tripled, from €25.7 million to €76.1 million.
Affirm's shares rose 1.32% the day after it announced the launch of its own AI model. It is also too early to assess the model's impact on the company's metrics, as it has been in use for only a month and a half. But Affirm has already said that the new scoring system increased the number of completed purchases by 3.4% compared with a control group. The additionally approved loans were also of higher quality: borrowers made repayments more reliably than customers who received loans using Affirm's previous models. In the second quarter of 2026 (the latest available data), Affirm's operating profit nearly tripled to $147 million, while its operating margin rose from 6.6% to 12.6%. At the same time, operating expenses jumped 24.5%, and provisions for credit losses rose 42.5% to $223 million.
The situation is somewhat different for Mastercard: the banking metric cost of risk does not apply directly to the company. For a payment network, fraud losses and the rate of false positives (when a legitimate transaction is identified as fraudulent) are more relevant. Back in 2024, the company reported that the introduction of third-party generative AI had increased fraud detection by an average of 20%, and by as much as 300% in some cases, while reducing false positives by more than 85%. In the second quarter of 2026, Mastercard's revenue from value-added services and solutions—including security services and the deployment of its own AI technologies—rose 20% year over year to $3.83 billion.
Who benefits most from AI
“There is an interesting paradox: the more accessible AI becomes, the less value there is in simply using it, and the more value there is in being able to apply it effectively to a specific business,” says Vadim Ivanenko.
He says AI will significantly change the banking industry, but the market currently tends to overestimate the speed of these changes and underestimate the complexity of putting them into practice.
For now, we are at a stage where AI's marketing value for some companies exceeds its proven economic value.
In addition, companies themselves are beginning to slow down AI adoption. For example, in October, US bank Goldman Sachs prohibited its traders and bankers from using third-party AI tools—Anthropic's Claude Code and Microsoft's GitHub—after reviewing their performance and risks.
Ivanenko believes that large banks and financial platforms have an advantage over smaller players thanks to their millions of customers, historical data, and huge transaction volumes. On the other hand, traditional banks often have fragmented IT infrastructure, complex approval processes, and high costs of making changes. But a large bank is not necessarily destined to outperform a small fintech company. Sometimes technological flexibility, data quality, and speed of implementation matter more than sheer scale.ㅤ