Financial services does not have an AI awareness problem. Most senior leaders in banking, insurance, wealth, lending and payments already know AI is going to change how their organisations operate. They have seen the demos, read the reports and probably sat through at least one presentation promising a step-change in productivity
The harder question is what happens next?
Launching an AI pilot is not the difficult part. Turning that pilot into something useful, measurable and safe enough to run inside the business is where many firms are now getting stuck.
According to the Bank of England and Financial Conduct Authority’s 2024 survey on AI in UK financial services, 75% of firms are already using AI, with another 10% planning to use it over the next three years. That is up from 58% in 2022, and firms expect the median number of AI use cases to more than double over the next three years.
So the issue is not adoption. It is execution.
KPMG’s 2026 Global Tech Report for Financial Services makes the same point from a different angle. Only 26% of financial services technology leaders say they are currently deploying AI use cases into production at scale, while 65% expect to be doing so within 12 months.
That gap tells us something important. The ambition is there. The activity is there. But the muscle required to move AI from experiment to operating capability is still developing.
Pilots are useful, but they are not the destination
There is nothing wrong with pilots. A focused proof of concept can help a business test an idea quickly, learn what is possible and avoid overcommitting too early.
The problem comes when pilots become a substitute for progress.
A demo might impress a steering committee, but it does not reduce cost-to-serve. A prototype might show technical promise, but it does not improve onboarding, speed up claims handling or take manual work out of a compliance process unless it is embedded properly.
That is usually where the hard work begins. The model may work, but the data is scattered across legacy systems. The workflow may look sensible on paper, but not match how teams actually operate. The automation may save time in theory, but compliance teams may struggle to explain or audit it. The business case may look compelling, but no one may have agreed who owns the process after launch.
This is why AI in financial services cannot be treated as a standalone technology project. It touches process, governance, data, risk, customer experience and operating model. Ignore those things and the pilot will stay exactly where it started: outside the real business.
Finance has to move differently
It is easy to borrow language from the technology world and talk about moving fast. But financial services has never had the luxury of simply moving fast and cleaning up later.
Banks, insurers, lenders, payments businesses and wealth platforms operate in a market where trust matters and mistakes have consequences. Any serious AI use case has to consider resilience, conduct, explainability, customer outcomes, data governance, cyber risk, model risk and third-party exposure.
The Financial Stability Board has warned about several vulnerabilities linked to AI adoption in financial services, including third-party dependencies, market correlations, cyber risks, and challenges around model risk and governance. (Financial Stability Board)
That does not mean firms should be slow. It means they need to be intentional.
Governance should not be bolted on at the end, when a team is already emotionally invested in launching something. It needs to be designed in from the beginning. That may sound like friction, but in practice it is what gives good ideas a chance of making it into production.
The best AI opportunities are usually unglamorous
There is a tendency to talk about AI in very large terms: transformation, disruption, reinvention. All of that may be true eventually, but it is not always the most useful starting point.
For most financial services firms, the best opportunities are much more practical.
Where are customers waiting too long? Where are teams checking the same documents again and again? Where is information being rekeyed between systems? Where are skilled people spending time on low-value work? Where are decisions slower than they need to be?
Those questions usually reveal better AI use cases than a blank-page innovation workshop.
Customer onboarding, KYC and KYB, fraud detection, claims handling, customer servicing, internal knowledge search, document review, compliance monitoring, advisor support, risk triage and operational reporting are not futuristic ideas. They are everyday pressure points.
And that is exactly why they matter.
The prize is not AI for its own sake. It is removing drag from the organisation. Faster onboarding can improve conversion. Better servicing can reduce cost-to-serve. Smarter compliance workflows can free specialist teams from repetitive review. Better internal knowledge tools can help people make faster, more confident decisions.
That is where AI becomes commercially useful: not as a headline initiative, but as part of how work actually gets done.
Klarna shows what focused execution can look like
Klarna is a useful example because the use case was specific, measurable and operational.
In 2024, it reported that its AI assistant handled 2.3 million customer conversations in its first month, equal to two-thirds of Klarna’s customer service chats. The company also said the assistant was doing the equivalent work of 700 full-time agents, reduced repeat enquiries by 25%, and helped customers resolve issues in under two minutes, compared with 11 minutes previously.
That does not mean every financial services firm should copy Klarna. Customer service is only one part of the picture, and every regulated business will have different constraints around risk, conduct and governance.
The useful lesson is narrower and more valuable. Klarna applied AI to a high-volume operational workflow, measured the impact against clear outcomes, and scaled where the business case was visible.
That is the shift more financial services firms need to make. Less AI theatre. More operational value.
Start with the problem, not the model
A common mistake is to start with the technology. A firm chooses a model, a platform, a vendor or a new capability, and then looks for somewhere to apply it.
That approach creates activity. It does not always create progress.
A better starting point is the business problem. Which process is too slow? Which customer journey is underperforming? Which compliance task is consuming too much specialist time? Which knowledge is trapped in documents, inboxes or legacy systems?
Once the problem is clear, AI becomes part of the solution rather than the headline. Sometimes the right answer will be an AI agent. Sometimes it will be workflow automation, better data architecture, a redesigned user experience or a simple internal tool that removes hours of manual work.
The question is not “where can we use AI?” The better question is “where would better automation, intelligence or decision support create measurable advantage?”
Why rapid prototyping matters
The traditional transformation model is starting to look too slow for the moment financial services firms are in. Long discovery phases, heavy documentation and months of governance before users touch anything may feel safe, but they also carry risk.
The risk is spending too long planning around assumptions that have never been tested.
Rapid prototyping offers a better route. Not reckless experimentation. Not “move fast and break things”. But a controlled way to test whether an idea has practical value before the organisation commits serious time and budget.
Within weeks, teams can understand whether users trust the solution, whether the data is good enough, whether the workflow makes sense, whether the compliance questions are manageable and whether the business case holds up.
That matters because the cost of scaling the wrong thing is high. A prototype gives leaders evidence before commitment. It separates ideas that sound exciting in a meeting from the ones that can actually survive inside the business.
Moving beyond the pilot
The next phase of AI in financial services will not be won by the firms with the most experiments. It will be won by the firms that can identify the right use cases, test them quickly, govern them properly and scale them with discipline.
That means asking better questions. Where are we experimenting but not seeing measurable value? Which bottlenecks would improve margin, customer experience or risk control if solved? Are our AI initiatives connected to real workflows, or are they isolated innovation projects? Do we have the data, governance and integration foundations to scale what works?
These questions matter because the market is moving from curiosity to execution. AI adoption is already widespread, expectations for scaled deployment are rising quickly, and the pressure to improve productivity, resilience and customer experience is not going away.
For financial services leaders, the opportunity is clear. The winners will not be the firms that talk most confidently about AI. They will be the ones that turn it into trusted, useful and measurable operating advantage.
And that starts by moving beyond the pilot.
