Picture a family with pensions scattered across former employers, investments spread across different providers, and a home insurance policy coming up for renewal. They may be paying more in fees than they realise, missing opportunities to grow their savings, or simply lacking a clear picture of their finances.
The first wave of consumer AI tools could explain what pension consolidation means or how investment fees affect long-term returns. The next wave of AI tools is likely agentic: it will help this family do something about it. It will identify pensions that could be consolidated, compare investment costs and opportunities, and check whether their insurance still offers the right coverage. And when the family decides to act, it will help them take the next step.
This is closer than it sounds. In September, Meta launched Muse in the US, a personal agent that connects to a user's email, calendar, payment methods and accounts and acts on their behalf. Within ten days it was the top app on the US App Store. Financial markets took note too: on 22 September, bank and insurance shares fell as investors weighed how quickly agents could change the way people choose financial products. Consumers are ready for this: in a recent EY survey across 23 countries, almost half said they had used AI to guide savings or investment decisions in the previous six months.
More recently, we've seen banks in the US raise concerns about agentic access - users authorising agents to perform tasks on their behalf.
Much of the technology for this already exists. What remains challenging is access to the financial information these agents need.
Much of the technology for this already exists. What remains challenging is access to the financial information these agents need. A family's pensions, investments and insurance policies may sit with several different providers, each with its own systems and processes. Financial data access already exists across Europe, but it remains fragmented. As AI agents begin acting on behalf of consumers, questions around responsibility, permission and supervision become even more important.
To tell a family whether they could be getting more from their savings, an AI agent needs to know where their investments and pensions are held, what they're worth, what they cost in fees and which pensions can be transferred. To assess whether they're overpaying for insurance, it needs the actual policy, including its coverage and price. Without that data, even the most capable model is guessing, and over a working life, a wrong decision about pensions or investments can cost a family thousands of euros lost to unnecessary fees or missed investment returns.
Where no clear framework exists, people improvise. They upload PDFs and screenshots, paste statements into chat windows, or hand over their online banking and insurance credentials so a tool can sign in on their behalf.
Each of these workarounds creates its own risk. Documents are often out of date by the time they are shared. Unmanaged sharing of login credentials can expose customers to fraud and may breach their provider's terms of use. Once the data has left, it can be difficult for anyone to tell which service holds it, for what purpose, or how to withdraw access.
For banks and insurers, improvised access can show up as unidentified traffic on their systems, with no clear answer on who is liable when something goes wrong. It also means a growing layer of AI services now sits between them and their customers, with no formal relationship in either direction.
Data access rules were first designed to get financial institutions to open up, but consumers are now sharing their data themselves, with whichever tool asks. Financial data sharing has been established in many European markets for years. With AI, however, the trend is accelerating.
What remains open is whether data moves through channels that customers, institutions and supervisors can see and trust.
The principles for doing this safely are well established, and Europe has already applied several of them to payments under PSD2. In short:
A framework like this also works in institutions' favour. Data holders and service providers can operate under clearer technical and commercial arrangements, and responsibilities can be established when something goes wrong. That is much harder when an agent simply signs in with a customer's password outside a managed data-sharing process.
These principles were first designed for a person clicking "share" in an app. They matter even more when an agent may act many times on a single instruction. Being able to set limits and withdraw access is what makes delegation safe enough to use at all.
When a customer asks an agent to review their pensions or investments, does that permission extend to initiating a transfer, or only to recommending one? Rules and technical standards will need to address questions like this, including how permission is scoped for multi-step tasks and how to make clear which service is acting on whose behalf.
The core principles can handle this, provided the industry leaves room for new ways of using data.
The financial industry has spent years building digital products around individual accounts, policies and transactions. AI is changing the starting point. Instead of asking customers to navigate between providers and piece together their financial situation themselves, an agent could help them understand and act on their finances as a whole.
That creates an opportunity for banks, insurers, wealth platforms and fintechs alike. Better access to financial information can help them develop services that are more relevant to each customer, identify opportunities that would otherwise be missed and remove friction from complex financial decisions.
But the quality of those services will depend on the quality of the data behind them.
An agent cannot identify unnecessary pension fees if it cannot see what a customer is paying. It cannot help someone understand their investments if it only has part of the picture. And it cannot responsibly act on someone's behalf without clear permission and accountability.
The industry therefore needs to think beyond what AI models can do and focus equally on the infrastructure that allows them to do it safely. That means making financial data accessible across providers and markets, establishing clear responsibilities and ensuring that information is consented, reliable and structured.
The next generation of financial services won't be defined by AI capabilities alone. It will be defined by how well those capabilities connect to the financial lives of the people they're meant to serve.
Because even the most capable AI is only as good as the data it can access.