AI & the future of financial services

Connected to the account, a digital banking assistant in the app shows how much the energy bill has gone up this month, blocks a lost card or explains what a rate review means for the mortgage. At large banks, these assistants hold millions of conversations a month. More interesting is that the assistant notices what is out of the ordinary. It sees a payment taken twice and starts the refund, or flags a rising fixed cost and lines up a cheaper alternative that the customer only has to confirm. Complex questions and customers in payment difficulty go to an employee. Because the assistant handles the everyday tasks, staff have more time for conversations that need personal attention.

An AI agent is software that works through the steps of a task by itself, until the task is done. For a mortgage application, it retrieves the payslip and the valuation report, checks that the figures match, calculates the maximum loan and prepares a file for the underwriter. At an insurer, an agent organises a claim, asks for missing photos and proposes a payout. The first agents already support mortgage assessments; other uses are still in trials. An employee reviews the proposal and decides. In future, an agent could guide the whole process from application to payment, referring only the exceptions to a person.

A credit application used to be a snapshot: forms, payslips, annual accounts and days of waiting. AI turns assessment into an ongoing process. With the customer's consent, a model follows the cash flow in the current account, fixed costs paid on time and a business's turnover. A bank can then set a credit limit in advance, and the customer sees an offer in the app based on a prior assessment. First-time borrowers and entrepreneurs without a long credit history are judged on their current income and payment behaviour. On platforms, eligibility for business finance is already reassessed every day. In future, the credit limit could move with the cash flow all the time, so the credit is there the moment it is needed.

AI now does much of what used to take days of manual work. Image recognition estimates the damage to a car door or a ceiling with a leak from photos, a language model reads the repairer's quote and sets it against the policy terms, and a fraud model checks whether the same photo has been submitted before. A simple claim, such as a broken window or a damaged phone, is sometimes paid out within minutes. Doubtful cases, large losses and claims with signs of fraud go to a claims handler or loss adjuster, who has the final say. Soon the claim will start before the customer calls: a sensor in the house reports a leak, or a measured value triggers the payout automatically.

After a sharp rise in share prices, a robo-adviser automatically sells part of the shares to restore the agreed balance with bonds, in line with the investor's goal, time horizon and appetite for risk. Someone with a few hundred euros can invest with a fixed strategy, for a fraction of the cost of a personal manager. Professional investors, meanwhile, have AI run thousands of market scenarios: what an interest-rate shock, an energy crisis or a trade conflict means for every position in the portfolio, and which positions are vulnerable together. In time, that computing power could reach every saver, with an assistant that keeps checking whether the portfolio still fits the customer's life.

Older systems for checking money laundering and sanctions mostly rely on fixed rules and produce many alerts that turn out to be false on investigation. AI can recognise more complex patterns: a network of accounts passing money round in circles, or a company whose turnover differs sharply from its peers. That can reduce false alerts and bring real signals to light sooner. The analyst then spends time investigating rather than clicking alerts away. The first AI agents now also prepare the investigation itself: they gather transactions, public sources and earlier reports and put together a file. The analyst starts with a substantiated file, in which transactions and the parties involved are already linked.

Every year a bank or insurer receives hundreds of pages of new rules, guidelines and supervisory expectations. The question is always the same: what does this mean for our products, processes and contracts? AI compares a new rule with product terms, policies, process descriptions and existing contracts, and identifies the problems for each product. Then it works out the consequences: how many customers need different information, which policies need an amended clause, what a new capital requirement costs. Compliance receives a proposal for the change before the rule comes into force. Further ahead lies regulation published in a form that systems can read and test directly.

A digital pension assistant explains what pension rules and choices mean for one member's income. Linked to that member's data and a calculation module, it works through personal scenarios, which is especially valuable where pension systems are being rebuilt. A member asks what happens if he works one day a week less, or what becomes of the partner's pension after a divorce, and gets an answer in plain language, with his own figures. Pension providers are trialling such assistants now. Further ahead, the question itself changes. As more and more people reach a hundred, the assistant works out for each member how long their savings must last, and how they combine work, care and benefits.

Banks want to act as a finance director for every customer. Using income, fixed costs and previous years, AI forecasts how the balance will develop over the coming months. In October, a household gets a nudge that the annual energy bill and the car insurance both fall due in January, with a suggestion to set something aside now. An entrepreneur sees that a large client usually pays late and is offered temporary credit in advance. An assistant that makes proposals and, with the customer's consent, carries them out is now being trialled. The bank moves from the place where money is kept to the place where people think about money.

With agent payments, a customer allows an AI agent to pay on their behalf. That gives banks and payment companies a new question: how do you know the customer has authorised the agent to make that payment? The answer is a digital mandate. The customer records what the agent may buy, from whom and up to what amount, and the bank or payment network checks every payment against it. Payment networks support the first real agent payments in online shops and are testing further uses, including business purchasing. The customer who presses the button makes way for agents that decide on their behalf, within limits the customer has set.

At some tills, a palm or face scan is now enough to draw on a linked account, with no phone or bank card. AI compares the scan with the stored template and helps check that it comes from a person who is actually present, not from a photo, a mask or a fabricated image. This is called liveness detection. On the phone, paying with a fingerprint or face scan is already normal for many people; at the till it is moving from trial to daily practice, in shops, stadiums and staff canteens. The body becomes the bank card, and the customer no longer needs to carry anything.

At the moment a customer pays, before the money changes accounts, AI compares the transaction with the customer's usual behaviour: the amount, the time, the device, the recipient and even typing speed. If a payment differs sharply, the system stops it or asks for extra confirmation, within milliseconds. AI learns from millions of earlier transactions and also recognises connections outside the set rules, sometimes at the very first attempt with a new type of fraud. What comes next is a shared picture. Banks and telecoms companies share fraud signals in real time, so a fraudster stopped at one bank is recognised at the next.

An AI-first bank or insurer uses AI for the work people need not do: checking documents, doing the calculations and also the predictive work behind every application, claim or customer question. An AI agent prepares a proposal. An employee discusses with the customer what a change in income, job or family means. People form a judgement, make choices and build trust; AI calculates how things might develop. One scenario for 2040: most applications are handled without any human step, while the number of real conversations with customers actually grows.

When AI checks documents, filters alerts and prepares files, the work shifts. The middle office needs less manual work, and new specialisms appear. Model validators test whether a credit or pricing model does what it promises, and people are needed to set up AI agents, watch their limits and step in when something goes wrong. Further ahead come roles that barely exist today. The agent banker manages the relationship with the agents of large business clients. The climate risk designer builds policies and loans around a changing climate. And a custodian of tokens and smart contracts looks after money that carries its own rules. On the customer side, demand grows for advisers who handle life's complicated questions.

Before the meeting, AI gathers the customer's details and works through scenarios: with and without extra repayments, with a fixed or a variable rate. During the meeting, AI calculates new options as soon as the customer mentions a different choice, and shows what they mean for income and outgoings. The adviser can keep their attention on the people at the table: on the doubts of a couple buying their first home, or on an entrepreneur whose pension is tied up in the business. Soon the customer will bring their own financial AI agent to the meeting, and adviser, customer and two agents will work out the options together.

Leaders in finance take the initiative on AI and new technology while protecting customers' trust. They describe the organisation's future clearly enough for employees to explain it to others. They show the curiosity and commitment they want to see in their people, and get them involved in building it. Employees get room to bring in ideas and help decide how the organisation develops. That is how leaders and their teams build the bank, the insurer and the pension fund of tomorrow.

In a future academy, employees practise with the AI and new technology that are changing their work. Organisations build such an academy themselves because they do not want to wait for training programmes to catch up. Those who have worked there for years learn to judge what AI and AI agents do. Newcomers practise straight away with the systems the organisation uses. Experienced colleagues pass on their knowledge to the generation that will work with agents.

Start-ups, tech companies, scientists and partners on different continents develop the service the customer sees together. Organisations also work with parties they meet in the market as competitors, for example to recognise patterns that are not visible in their own data. With scientists they test what AI can do reliably, and with governments they work on shared facilities, such as a digital identity. New services reach the market faster as a result. The quality of that ecosystem increasingly determines the quality of the service.

An organisation runs on processes: assessing an application, carrying out an instruction, running a check, handling a complaint. When AI agents can carry those processes from start to finish, the question of how many people an organisation needs changes. Digital players already serve millions of customers with far fewer employees per customer than traditional organisations. The scenario for 2040: an organisation with a small team and thousands of agents. The people design the products, decide how much risk the organisation takes and hold the conversations in which trust is at stake. Agents do the routine work and check each other's results; people judge the exceptions and keep oversight.

In the financial sector, leadership has always meant accountability too: who decided this, and why? When AI agents assess files, check payments and speak to customers, a leader will manage a team of which part is not human. A team leader decides which decisions an agent may take alone, looks at the outcomes of thousands of cases rather than ten files, and remains answerable for every decision the team makes. At the same time, the human part of the team needs more attention. People who see their profession changing want to know where their judgement still counts. Leaders of people and agents make that visible every day.