The promises and pitfalls of open banking in UK consumer lending

Open Banking has transformed UK consumer lending by giving financial institutions real-time visibility into an applicant’s financial life. Credit and Fraud risk expert Kevin Allen discusses the potential pitfalls of this method, including hidden bias in lending models, and the methods that lenders use to mitigate these shortcomings.

What does open banking show a lender?

The days of self-declared paper payslips and static credit bureau scores are gone, and for the most part have taken manual underwriting with them. For more than eight years now, financial institutions have been able to view an applicant’s finances through Open Banking, with lenders receiving raw, line-by-line transaction data including income credits, standing orders, direct debits, point-of-sale transactions and account balance histories.

The rise of Open Banking has categorically transformed the lending journey. When an applicant grants consent, their transactional history spanning from twelve to twenty-four months is shared, directly from theaccounts they choose to connect.

How did lenders check bank data before open banking?

I started in credit risk in the mid-90s, building application scorecards. During this time, a lender would learn about an applicant from the application form and the credit reference agencies. Income was self-declared and supported by a paper payslip or bank statement that an underwriter would manually check. Of course, this process had risks: it was slow, prone to human error, and susceptible to document fraud. 

By the mid-2010s, applications and statements had moved online. Applicants uploaded PDF statements, and lenders used software to analyse income. Other services took applicants’ online banking login and copied the data from the screen, a method called screen scraping, which required handing a banking password to a third party. New authentication rules introduced in 2019 began to move firms away from screen scraping,  and the FCA gave firms until March 2020 to move across to the banks’ own interfaces.

Along came Open Banking— a standard and secure way for customers to share account data with regulated third parties and make payments through APIs. In practice, a customer starts the connection and approves it with their own bank, and a lender never sees a password. For the lending industry, this was the first time an application came with bank data that showed spending as well as income. This included hidden subscriptions, BNPL (Buy Now Pay Later) transactions, and other living costs. By March 2025, one in five UK consumers and small businesses used open banking. By December 2025, the FCA put the number of users at over 16 million.

What does open banking change for lenders and borrowers?

The obvious benefit of Open Banking is that it has simplified the sharing of financial data for both applicant and lender: the former no longer has to find and upload statements, and the latter no longer has to read them. Besides this, transactional data improves the accuracy of affordability checks. A credit file records how someone has repaid credit- it does not log rent, or food shopping, leaving a lender to estimate those costs. Open Banking enables a fairer and far more accurate picture of an applicant’s finances. 

Transaction data can also help lenders spot signs that a customer may be in vulnerable circumstances. The FCA defines a vulnerable customer as “someone who, due to their personal circumstances, is especially susceptible to harm, particularly when a firm is not acting with appropriate levels of care”. Before Open Banking, a lender would often learn about these circumstances only if the customer chose to disclose them. A bank account can show some of these signs, for example a redundancy payment followed by no salary, or transfers to betting sites that grow month on month. Lenders can use signs like these to offer support, such as more time to pay or a referral to free debt advice. Reading twelve months of transactions by hand is impractical at any scale, so lenders use software and increasingly machine learning to categorise data. In a 2024 survey by the Bank of England and the FCA, 75% of UK financial services firms said they already used AI, up from 58% in 2022. Data bias was among the five risks that firms rated highest.

What are the risks of open banking in lending?

While open banking is a secure and regulated way to share financial data, it risks proxy bias: machine learning models often pick up bias from the data they learn from. When a model is trained on spending habits, it may learn to rely on a variable that closely tracks a protected characteristic such as sex or ethnicity. Suppose a model finds that regular shoppers at a premium supermarket default less often. If that supermarket’s customers skew towards one sex or ethnic group, the model could end up favouring that group. Payments to a political party, a religious organisation, a trade union or a specialist health service carry a similar risk, since each can reveal something about an applicant that should not affect a credit decision.

The obvious answer would be to remove the sensitive field from the data, but research suggests that this does not wholly solve the problem. Lenders can reduce the risk of proxy bias at several stages:

• Feature suppression and redaction strips out sensitive merchant category codes (MCCs) and narrative descriptions, such as trade union dues or religious donations, before the data enters a credit-scoring model.
• Category aggregation groups raw transaction text into broad, neutral labels such as groceries or utilities. The model then sees how much an applicant spends on food, not whether they shop at a premium supermarket or a discounter.
• Fairness constraints can be built into how the model is trained, or the model’s scores can be adjusted afterwards.
• Disparate impact testing checks whether approval rates and prices differ across demographic groups, and which features drive any gap. This is harder when a lender does not hold demographic data about applicants. 
• Explainable models and human oversight keep a model interpretable, so an underwriter can explain a decline to the applicant, and a person can review and correct the decisions the model gets wrong.

In practice, these technical fixes are always supplemented with careful and context-informed human review. Of course, such risks are not exclusive to Open Banking, and exist in any situation where traditional statistical models are replaced with machine learning engines.

The Benefits Largely Outweigh the Risks

The promises of Open Banking outweigh the risk of proxy bias, which lenders can manage. As machine learning sorts transactional data into categories, a lender is given fine-grained insight into consumer behaviour. Beyond basic net income and fixed living costs, Open Banking enables lenders to see daily spending patterns, such as how often an applicant buys takeaways or how much they pay for subscriptions. This level of detail can show signs of financial stress and risk flags well before they reach a credit file:

• Signs of financial difficulty/vulnerability can show up as frequent transfers to betting accounts, regular use of high-cost short-term credit, or a growing overdraft.
• Changes to income can show in the account sooner as a salary stops or falls. A credit file shows nothing until the applicant misses a payment and the lender reports it.
• Undeclared credit, such as a new loan or a BNPL plan, may show up as repayments in the account before it appears on a credit file. This helps a lender avoid lending more than the applicant can afford.

Ultimately, as lenders continue to train, test and monitor the models to read transactional data, Open Banking can lead to better informed and fairer lending decisions than any earlier methods.

If money worries are affecting you, free and confidential help is available from StepChange (0800 138 1111), National Debtline (0808 808 4000), MoneyHelper (0800 138 7777) and Citizens Advice (0800 144 8848).

Sources

• Addleshaw Goddard, Payment services post PSD2, 2018.
• Bank of England and Financial Conduct Authority, Artificial intelligence in UK financial services – 2024, 21 November 2024.
• Competition and Markets Authority, Retail Banking Market Investigation Order 2017, 2 February 2017.
• Experian (Tim Baker), Use of CATO Data in Affordability, Credit Research Centre, 28 August 2019.
• Financial Conduct Authority, FG21/1: Guidance for firms on the fair treatment of vulnerable customers, 23 February 2021.
• Financial Conduct Authority, Open banking and open finance, updated 2 June 2026.
• Financial Conduct Authority, Open banking: a year of progress, 16 December 2025.
• Financial Conduct Authority, Research Note: A literature review on bias in supervised machine learning, 11 December 2024.
• Kozodoi, N., Jacob, J. and Lessmann, S., Fairness in Credit Scoring: Assessment, Implementation and Profit Implications, arXiv 2103.01907, 2021; European Journal of Operational Research 297.
• Navya S. D., Sreekanth D. and Uma Sankari, S. S., Structural Gender Bias in Credit Scoring: Proxy Leakage, arXiv 2601.18342, 2026.
• Open Banking Limited, Impact Report 7, May 2025.
• Open Banking Limited, Why open banking is safe, undated.
• Pinsent Masons Out-Law, PSD2: FCA gives temporary lifeline to screen scrapers, 6 September 2019.