AI can help lenders judge risk quickly, but it can also repeat errors and hidden bias. You can find out whether software made the decision about you, check the data it used, and ask a person to review the outcome.
The BBC asks whether artificial intelligence would be a good judge of who should get a mortgage, a car loan or a mobile phone contract. The question is no longer only theoretical. Lenders have used statistical scoring for decades. Many now add machine-learning models that look at far more variables, and some decisions are made in seconds with no person involved. These systems can be faster and, in some cases, more consistent than human underwriters. They can also be opaque. A model trained on historical lending data can carry over patterns of past discrimination. It can also give weight to proxies such as postcode or spending habits in ways neither the applicant nor the lender fully understands. This guide explains what an applicant can do when an automated system turns them down. It uses the general rights available under UK and EU data protection law. Details differ between countries and change over time, so check current guidance from your national regulator.
Gather what you need before starting
Have these ready:
- The written or emailed decision, including the date, the product applied for and any reference number.
- The name of the lender, and of any broker or comparison site you applied through. The firm you dealt with may not be the one that made the decision.
- Proof of identity and address. You will need it when you ask organisations for your personal data.
- A record of what you entered in the application: your income, employment status, address history and existing debts.
- Time and patience. Firms usually have a set period to respond to data requests and complaints. Expect the process to take weeks rather than days.
It also helps to understand the three things a lender usually draws on: what you told them, what credit reference agencies hold about you, and the lender’s own internal rules or model. A refusal can come from any of the three.
Ask whether the decision was automated
Contact the lender and ask in writing whether a solely automated process made the decision, with no meaningful human involvement. Under UK and EU data protection rules, people subject to significant decisions made solely by automated means have specific safeguards. A refusal of credit is generally treated as significant. The UK has revised its rules in this area, so check the Information Commissioner’s Office guidance for the current position. In both regimes the core protections include being told that automated decision-making took place, being able to give your point of view, and being able to contest the outcome.
Ask the lender to explain the main factors behind the result. Firms are not usually required to disclose the inner workings of a proprietary model. They should still be able to describe the categories of data used and the general logic in plain terms.
Obtain your credit reports
In the UK, the main credit reference agencies must provide a statutory copy of your credit report on request, free of charge. Request one from each agency, because lenders do not all use the same one and the records can differ.
Check each report line by line. Look for:
- accounts you do not recognise, which may point to fraud or identity theft
- debts recorded as outstanding after you settled them
- wrong addresses or links to people you have no financial connection with
- missed payments you believe were made on time
If you find an error, raise a dispute with the agency. Agencies normally add a notice to your file while they investigate. If the record is accurate but needs context, such as a missed payment during illness, you can usually add a short statement known as a notice of correction.
Submit a subject access request to the lender
A subject access request gives you a copy of the personal data a lender holds about you. It can include data used or generated during your application. Send it in writing to the lender’s data protection contact and ask specifically for:
- the data collected from you and from third parties
- any scores, risk ratings or categories assigned to you
- information about any automated decision-making, including the logic involved and its likely consequences
Organisations generally have one month to respond, with extensions allowed for complex requests. A request is normally free. The response may not reveal the model itself, but it can show whether the system relied on information that was wrong or out of date.
Request a human review
If the decision was automated, ask for it to be reconsidered by a person with the authority and competence to change it. Put any corrections or extra context in writing. That might be recent payslips, proof that a disputed debt was cleared, or an explanation of a gap in your address history.
A genuine human review should involve real judgement, not a staff member simply confirming the system’s output. If the reply looks like the original decision restated, say so and ask what the reviewer actually considered.
Escalate the complaint if necessary
If you are unhappy with the result, make a formal complaint to the firm first. Most regulated lenders must have a complaints procedure and send a final response. If that does not resolve matters:
- Complaints about how the lending decision was made, or how you were treated, can usually go to the Financial Ombudsman Service in the UK.
- Complaints about how your personal data was handled, including automated decision-making, can go to the Information Commissioner’s Office.
In the EU, national data protection authorities and financial ombudsman schemes play similar roles.
Avoid the mistakes people commonly make
- Reapplying straight away. Repeated applications can leave several hard searches on your file, which some lenders treat as a sign of financial stress. Use eligibility checkers that run soft searches where they are available.
- Assuming AI made the decision. Many refusals come from fixed lending criteria, affordability rules or bad data. Asking first stops you challenging the wrong thing.
- Checking only one credit report. An error may appear with one agency and not the others.
- Making vague complaints. “The algorithm is unfair” is hard to investigate. “You recorded a default that was settled” is not.
- Missing deadlines. Ombudsman schemes usually have time limits for bringing a complaint after a firm’s final response.
- Expecting to see the model. Lenders rarely share the model itself. Ask for the factors and the data instead.
Recognise when this approach is the wrong choice
This process suits someone who suspects an error or an unfair automated outcome. It is less useful when:
- The refusal reflects genuine affordability. A human review will probably reach the same result. Free debt advice services may be more useful.
- You suspect fraud. If your report shows accounts you did not open, contact the lenders concerned and consider fraud-prevention services before anything else.
- The lender made the decision with significant human involvement. The automated-decision safeguards may not apply, although the normal complaint routes still do.
- You urgently need credit. A challenge takes time. Other providers, credit unions or a guarantor arrangement may be faster.
- You believe there is systemic discrimination. One complaint will not fix a biased model. Reporting concerns to regulators or advocacy groups may have more effect.
Frequently asked questions
Can a bank refuse me credit using only AI?
Lenders can use automated systems to make credit decisions, and many do. Under UK and EU data protection law, people affected by significant decisions made solely by automated means have safeguards. These include being told about the automated processing, being able to give their view and being able to ask for human intervention. The exact rules vary between countries and have been revised in the UK.
Do I have a right to know why I was refused a loan?
You can ask the lender for the main reasons and for information about any automated decision-making. Lenders are not usually required to reveal the full workings of their scoring models, which are treated as commercially sensitive. They should still describe the kinds of data and the general logic used. A subject access request can show the personal data and scores linked to your application.
Can AI credit scoring be biased?
Yes, it can. Models learn from historical data, and if past lending decisions were unequal, a model may reproduce those patterns. Variables such as postcode can act as proxies for protected characteristics. Lenders remain responsible for fair treatment under consumer and equality law. The EU AI Act classifies creditworthiness assessment as a high-risk use, which brings extra obligations for providers.
Does checking my credit report harm my score?
Checking your own credit report is normally recorded as a soft search. Soft searches are visible only to you and do not affect how lenders view your file. Full credit applications usually create a hard search, which other lenders can see. Several hard searches in a short period may count against you, which is why eligibility checkers that use soft searches are often recommended.
Sources and further reading
- Information Commissioner’s Office: guidance on automated decision-making, profiling and subject access requests under UK data protection law
- Financial Conduct Authority: rules and guidance on creditworthiness assessments, affordability and fair treatment of consumers
- Financial Ombudsman Service: information on bringing complaints about lending decisions
- BBC News: technology coverage asking whether AI should judge eligibility for mortgages, car loans and phone contracts
Surfaced from the rss:bbc_tech signal “AI in credit decisions”. AI-assisted draft, editorially reviewed.

