This year, London Tech Week kicked off with a speech from Jessica Rusu, the Financial Conduct Authority (FCA)’s chief data, intelligence and information Officer. In it, she unveiled the ‘Supercharged Sandbox’ – a cornerstone of the government’s efforts to support firms in building early-stage AI proof-of-concepts.
In effect, this is a de-risking mechanism to boost the faltering adoption of AI in financial services. It’s designed to create demand-side readiness inside regulated institutions while startups that are developing AI solutions can learn what banks are looking for and iterate with clearer signals, thereby shortening the time to enterprise adoption.
It’s also an ecosystem play – one that recognises innovation in financial services rarely happens in-house. A 2024 survey from the Bank of England and FCA found that a third of all AI use cases in UK financial services are third-party implementations. Even as banks ramp up AI talent acquisition, their internal teams are typically focused on a few key use cases and governance over greenfield product builds. That leaves the exploratory work to external vendors.
The Supercharged Sandbox gives those vendors a way in under clear regulatory guardrails. Meanwhile, it offers banks a rare opportunity – a chance to experiment with AI without triggering the usual, institution-wide anxiety about risks in using the technology.
Solving for demand-side opacity
Too often, the barrier to innovation is access. As a startup founder, you’re always asking, “What does my customer want?” It’s all very good selling a solution you think will be helpful, but to be commercially successful, you must solve real-world problems that are not already being addressed, all without too much disruption of the customer’s day-to-day operations.
In that respect, product-market fit is the most crucial attribute a startup can possess. But it is not something you can design in a vacuum. As Paul Graham, founder of Y Combinator, said, “launching teaches you what you should have been building.”
On the other hand, banks often don’t have the tools or in-house expertise to build their own GenAI products or even more complex products from scratch. Even JPMorganChase, the largest bank in the world, which started building its AI research programme in 2019 with the hire of distinguished researcher Manuela Veloso, has admitted there is “probably not” any sense in it creating its own large language models. That is doubly true for smaller banks. Many firms do not have the infrastructure or availability in-house to even build their own prototypes or proofs-of-concept for basic AI use cases.
Overcoming institutional barriers
Even for banks with the requisite capabilities, however, the challenge does not end there. They must also overcome barriers rooted in an institutional mindset.
In highly regulated industries like financial services, there is strong inertia against technology adoption. These companies possess very talented employees but often lack the headroom or regulatory flexibility to run more than pilot programmes. Nearly a quarter of UK financial institutions marked data protection and privacy laws as a “large constraint” on AI deployment. Another third pointed to the regulatory burden associated with data privacy as their main blocker.
These regulations, though vital for safeguarding citizens’ personal information and assets, nonetheless make for slow-going innovation. This is precisely why Prime Minister Sir Keir Starmer has called for “sector champions” as part of the UK Government’s AI Opportunities Action Plan. Think of them as businesses that can blaze a trail and leave behind successful use cases for others to pick up and adapt to their own purposes.
In that respect, the Supercharged Sandbox scheme will help flip the typical dynamic on its head. Instead of banks having to list all the obstacles in their path, they can define what they need upfront, unlocking the tools and regulatory headroom needed to achieve them.
That is what makes this project interesting. It provides the market with a target to aim at and opens the floor for a broader set of players to respond.
Letting the ecosystem do the building
Despite banks like Capital One hiring tech talent “as aggressively as [they] can” for roles across the AI stack, I wouldn’t expect most banks to build many end-to-end solutions themselves. Their capital and resourcing are often earmarked for risk management. Even with NVIDIA infrastructure at hand, it will be difficult to dedicate teams to these types of experimental projects.
Instead, they’ll likely look to external vendors who can move quickly and slot into these sandboxed environments. Startups will be incentivised to find edge cases and iterate fast to deliver functionality that banks may not even know they needed. As the FCA’s chief data officer noted, the sandbox is specifically designed to help firms “who lack the capabilities” to test AI ideas, which underlines the need for partners who do have those capabilities and can slot into controlled, compliant environments.
A thriving ecosystem starts from the ground up. We saw this last week when NVIDIA announced partnerships with Mistral and other European stakeholders to deploy over 3,000 exaflops of compute for sovereign AI. The goal is a regionally embedded infrastructure that startups and enterprises can build on. In the UK, the sandbox plays a similar role, a policy tool to unlock local innovation by making it cheaper, safer and faster for vendors to engage with highly regulated institutions.
Creating space for discovery
Startups can’t solve for what they can’t see. One of the first steps to building successful AI solutions, as Paul Graham says, is to launch one, because the customer or user feedback will teach you what really matters. Not every bank knows exactly what it wants until it sees it.
That’s why the Supercharged Sandbox feels like a well-timed initiative: it encourages banks to discover what works well for them, and then leaves space for the rest of the ecosystem to finish the job.
Tom Henriksson is a general partner at OpenOcean
Read more: Sovereignty remains key to the UK’s AI ambitions
