AI in banking: The rise of autonomous finance

After two years of explosive progress in generative models, AI has become the defining force behind innovation within financial services.

According to Japanese consultants NTT Data, a remarkable 91 per cent of banking boards now have generative AI initiatives on their agendas - a level of executive sponsorship unmatched by any other technology wave in decades.

The benefits of AI powered banking are two-fold: sharper, more human-centred customer experiences, and radically leaner operations. Here are three areas AI is impacting the banking sector.

1 Re-imagining the customer experience

Banking chatbots have matured from the basic FAQ widgets of the past to the sophisticated, conversational advisers in use today.

This technology can execute transactions and escalate complex cases, saving institutions an estimated $7.3 billion in annual service costs, according to tech analysts Juniper Research, and freeing up resources that banks can redeploy to higher-value client work.

Large language model (LLM) agents already handle document queries and policy explanations at near-human levels of comprehension. Research prototypes - such as the Customer Analysis Pipeline Retrieval-Augmented Generation (CAPRAG) hybrid AI agent developed by researchers at Tunisia's engineering school ESSAI - show how banks can blend different data (vector and graph) from external information to improve their results. 

Meanwhile, machine-learning algorithms continuously analyse individuals’ spending patterns, lifestyle signals and financial goal trajectories to determine 'next-best actions'. Recent studies on AI-based personalisation have revealed impressive prediction accuracy above 88 per cent for recommending credit-risk-aware products, such as a loan guarantee.

In open-banking markets, the scope of these insights is widening. By aggregating data from multiple accounts, banks can enable richer, self-driven ways to help people manage their money, such as automated bill-splitting; just-in-time savings sweeps, where an AI agent automatically moves money into savings or investment accounts at the optimal moment for consumers; and tax-loss harvesting, which helps reduce investors' tax liability, long before the consumer even makes a request.

2 Quiet revolutions in the back office

JPMorgan’s COIN platform analyses commercial loan agreements in a matter of seconds - work that previously took lawyers an estimated 360,000 hours a year to complete. Similar natural-language models now generate regulatory reports, reconcile payments and draft marketing copy, slashing turnaround times from days to minutes.

Meanwhile, according to research firm PYMNTS seven in 10 financial institutions already lean on AI capabilities to police faster-payments fraud and synthetic ID schemes, often using third-party platforms that monitor billions of signals in the cloud. Even financial regulators are adopting these tools - Germany’s financial regulator BaFin reported that AI added to its market-abuse alert system last year has “substantially improved hit rates”, raising the odds of catching offenders.

3 Governance and ethics: holding the trust line

As models move deeper into functions such as credit approvals, portfolio advice and surveillance, then bias, explainability, and privacy become existential.

The forthcoming European Union AI Act will designate many financial-risk models as 'high risk', meaning they will require rigorous documentation, fairness testing and human-override channels before the 2026 enforcement date.

Firms that embed model cards, counterfactual explanations and privacy-preserving learning methods such as federated or synthetic-data pipelines into their development lifecycle, so the reasoning behind AI algorithms' decisions are transparent and understood, will be in a stronger position for global compliance.

What comes next for banking's AI strategy?

In the next couple of years we will see 'agentic finance'. That is 'level-3 autonomous finance, ie systems which automatically move idle cash to best performing accounts, refinance debt when interest rates dip and negotiate utility contracts. Frameworks outlining the six levels of autonomous finance suggest mainstream adoption of self-driving money - ie AI agents automatically managing people's money will be within 24 months.

We will also see embedded AI and open finance. Secure APIs are dramatically shortening the distance between data, model and moment. Early results from Citizens Bank’s open-banking platform show a 95 per cent drop in traditional screen-scraping, which is when a bank's customers share their login credentials with a third-party app that then mimics user behavior to pull financial data onto the external platform. This comes with security risks and it can misread the data, but a secure API overcomes this and can pave the way for real-time credit scoring via external apps.

As the computing footprint for LLMs shrinks, ‘small’ frontier models will run directly on secure enclaves on smartphones or other devices. This will keep biometric spending signatures, such as face ID, local while still supporting federated learning updates with the cloud.

Then there is the rise of 'continuous assurance tooling' - expect AI-for-AI. These are advanced AI validation models that watch and check AI production systems for drift, hallucinations and unfair impact. Regulators are likely to mandate such controls as a condition for using generative AI in regulated financial advice.

Human-in-the-loop evolution - The most successful fintechs will treat AI as a teammate, not a replacement. Job roles will shift from rote processing to model stewardship. This will entail curating data, auditing outputs and designing empathetic intervention pathways - a skillset already highlighted by BaFin’s experience and echoed in global surveys of banking sector leaders.

AI models are no longer a laboratory curiosity; it has become the new operating system of finance. Institutions that harness its power responsibly - balancing radical automation with transparent oversight - will define the next era of customer trust and operational excellence.

At the Gillmore Centre, our research agenda centres on these twin pillars: unlocking AI’s generative potential while engineering the guardrails to ensure finance remains fair, explainable, and human-centric.

The next 18 months will separate early experimenters from AI-native leaders; the next five years will decide the competitive map of global financial services. The time to scale, audit and govern is now.

This article is republished from Finextra.

Further reading:

What does the bitcoin inflation rate really mean?

Your next holiday may hold the key to your future equity investments

AI bias: What can the Claude chatbot leak teach investors?

Why leaders should seize the day in adopting AI

Ram Gopal is Professor of Information Systems Management and Director of the Gillmore Centre for Financial Technology. 

Learn more about AI in banking and finance on the MSc Financial Technology course.