AI is no longer an experiment in financial services. It’s already embedded in day-to-day operations.
Banks are rolling out AI-driven customer service, investing in automation, and exploring how generative and agentic systems can improve decision-making. Around 92% of global banks are already using AI in at least one core function, which suggests the industry is well on its way.
As these initiatives scale, however, a different reality starts to emerge.
What works in a pilot often becomes more complex, not to mention expensive, when rolled out across the business. In many cases, the issue is not the AI itself, but the environment it is being deployed into.
There is no ‘magic bullet’ for AI in financial services. Its impact is shaped by how well organisations can access, connect, and act on their data — increasingly through real-time approaches such as data streaming.
When AI meets reality
This is particularly visible in customer service.
AI-powered chatbots and virtual assistants are now widely used to handle routine queries, categorise transactions, and guide users through basic processes, and in the right conditions they can deliver real efficiency gains.
Scaling those systems, however, is a different challenge altogether. Despite the level of investment, many of these systems continue to rely on fragmented or outdated information, which sees the experience quickly starts to break down.
For example, customer context can be missing or inconsistent. Conversations don’t carry across channels. Employees are forced to double-check or correct outputs, removing much of the efficiency they were meant to deliver.
In a sector where trust matters, that’s difficult to ignore.
Legacy systems are still doing most of the damage
Frameworks like the EU AI Act are shaping how organisations deploy AI, particularly in high-impact use cases such as credit scoring, fraud detection, and customer risk assessment. Requirements around transparency, auditability, and data governance are raising the bar for how AI systems are built and monitored. And rightly so, given the importance of trust in financial services.
In practice, though, legacy infrastructure is often the bigger issue. When data is fragmented across systems or difficult to access in real time, it becomes far harder to evidence how decisions are made or ensure models are operating on complete, up-to-date information.
In many cases, it is these underlying data limitations — rather than regulation itself — that slow progress. With many banks reliant on systems built up over decades through mergers, upgrades, and workarounds, they struggle to create a single, reliable view of a customer or transaction — something AI depends on.
This also helps explain how FinTech startups can outmaneuver industry giants. Free of technical or technological baggage, they can build around modern data architectures from the start and bring new ideas to market more quickly.
AI can’t reach its potential, or in some cases even run, in a legacy environment. Sticking with the customer service example, one such example is personalisation.
Personalisation only works if the data is current
Personalisation has been a goal in financial services for years, but in many cases it has been closer to segmentation than true individualisation. If AI is going to change that, it needs a flow of data that can tell it exactly what’s happening in the moment.
Hyper-personalised experiences depend on real-time context — understanding a customer’s behaviour and situation as it evolves, often enabled by data streaming to ensure systems are working from the most up-to-date information. Without that, interactions quickly become disconnected.
A customer might start a query in an app, follow up by email, and then call support, only to repeat the same information each time. This is not a limitation of AI, but of how information moves between systems.
When that flow is in place, interactions become far more seamless, with context carrying across channels and experiences feeling more consistent.
AI should support people, not replace them
There is a lot of discussion about AI replacing human roles in financial services, but a fully automated model is unlikely.
AI is highly effective at handling repetitive, high-volume tasks, like triaging customer queries or processing transactions. But more complex interactions still require human judgement, empathy, and context — particularly in areas such as lending, investments, or financial advice.
Customer expectations also play a role. While digital channels continue to grow, many customers still value the option to speak to a person, particularly in more sensitive or high-stakes situations.
As a result, the most effective model is increasingly a hybrid one. AI handles routine processes and surfaces insights, while people focus on more complex and sensitive interactions — provided both are working from the same reliable, up-to-date information.
Getting the foundations right
As AI adoption continues, the conversation is starting to shift.
Early investment has focused on what AI can do. Attention is now turning to what is needed to make it work properly.
That starts with data. Organisations need to ensure data is connected, governed, and available when it is needed. Too often, this work is treated as something to address later, when in practice delaying it tends to create more complexity and cost over time.
There is also a cultural element to consider. In many organisations, established ways of working can slow progress just as much as technology. Addressing that is just as important as modernising systems, particularly when AI initiatives span multiple teams.
A question of readiness
Financial services organisations are being asked to move quickly while maintaining high levels of control and accountability. That is not easy.
The organisations that make progress will be those that focus less on AI as a standalone capability and more on the conditions that allow it to work. If those conditions are not in place, results will remain inconsistent.
As more AI projects move into production, that gap will become harder to ignore. The conversation will shift away from what AI could do and towards what organisations are actually able to support in practice.
In most cases, it comes down to a simple question: is the data ready?
Learn more at confluent.io
- Artificial Intelligence in FinTech
- Data & AI