Journey stages
By: Joy Macknight
With financial institutions across the world grappling with the complexity of rolling out artificial intelligence (AI), seeking efficiency gains within strict governance frameworks, it should come as no surprise that a session entitled ‘Building AI in a bank without burning millions’ was filled to capacity at Money 20/20 in Amsterdam earlier this month.
Situated in the context of rising AI costs, CEO Andy Ellis and Chief Technology Officer Wayne Freeman outlined NatWest Boxed’s approach and lessons learned.
A fully owned subsidiary of NatWest Group, NatWest Boxed is a cloud-native banking-as-a-service (BaaS) provider offering an embedded finance platform to the likes of motoring association The AA, financial services firm Saga Money, and supermarket chain Sainsbury’s, three of the largest British consumer brands.
NatWest Boxed’s foundations were laid as the underlying technology platform for NatWest’s digital business bank Mettle, before being spun out as a separate entity in 2022.
With 360 engineers out of 800 employees, NatWest Boxed has built its own technology stack over the past seven years. “We decided to organise ourselves in a modern way in terms of our set up, as well as how we deploy and ship code,” said Ellis.
A closely tracked metric is the number of production releases in a month. Last month recorded 8,500 releases, which is unusual for a bank. “Unsurprisingly, we have played a key role in driving the AI agenda across the bank, given the nature of AI and our strong technology team,” Freeman said.
According to Ellis, NatWest Boxed has spent a few million pounds to date in understanding how to roll out AI effectively and steadily. His main challenges were knowing where to start, choosing which models to use, getting everybody up to speed, and enabling innovation.
“We identified 38 AI use cases, even after shortlisting. My plea to the team was to provide tangible results – I need to see more P [profit] or at least no more L [loss],” he said.
NatWest Boxed has already had some AI success, particularly in the engineering space. For example, even with integrating 100 new engineers in the past year, its productivity has increased by 25% in terms of the amount of code and number of releases.
Ellis added: “We have focused on culture, governance and return on investment to deliver results.”
Journey stages
The first step for NatWest Boxed was to turn on its existing IT providers’ AI tools to connect immediately. “We started off small to allow time for the team to become accustomed to having AI within our environment. Then we looked for the biggest bottleneck in the bank,” explained Ellis.
NatWest Boxed’s engineering function is change and run, so the person that makes changes to the code also runs the code. With this set up, its high-end engineers were spending 40% of their time on the run, effectively searching for vulnerability in their code. “It’s boring and expensive, so we focused our AI efforts on that,” said Ellis.
For example, it had an average of 2000 vulnerabilities, which had to be gone through manually and closed down. NatWest Boxed implemented an AI agent to find and propose new code, which was then reviewed by an engineer to determine if it was the right code. It was a “brilliant initiative”, according to Ellis.
In phase two, NatWest Boxed aimed to get all 800 staff fluent and engaged in AI. “That’s incredibly difficult and I still don’t think we’ve cracked it,” said Ellis. “Some people just don’t want to learn about AI, so it’s an ongoing battle. One of the things we do, quite simply, is provide self-service courses on the best use of AI every Friday afternoon.”
As part of phase three, which is just beginning, the team is evaluating developments in AI over the past nine months, as well as how NatWest Boxed has changed as an organisation. “The first change is that we’re no longer running AI from the side – we’re put it in the middle of our prioritisation, so product owners and engineers aren’t making a choice between feature building and AI,” said Ellis. “Over the next three months, we’re going to accelerate the agentic software development lifecycle. We think the time is right.”
According to Freeman, the AI journey is made more difficult because, although the rough direction of travel is widely understood, the destination changes every week. “It’s hard to keep everyone aligned with developments. You can’t top down everything, because this affects people’s jobs, livelihoods, and careers. It needs to be bottom up, top down and side to side,” he said. “We’ve now realised the people complexity piece is where the real challenges are.”
In light of constantly changing technology, Freeman believes governance is key for banks. “We’re trying to be light touch, so we don’t close off innovation, but rigorous at the same time. Our approach is to step into it and try to get it right, with the understanding that we don’t know exactly where we’re heading,” he said.
Cost concerns
Undoubtedly, the cost of AI is becoming an issue, particularly in light of Microsoft and GitHub moving Copilot from flat-rate allowances to usage-based billing.
“Until now, we’ve taken an empowered, federalised, decentralised approach. But if AI is starting to become meaningfully more expensive, we’ll need to limit or decide how much freedom we can allow,” said Ellis.
Freeman has built a FinOps function in the team, so every unit of cloud or any other spend is categorised. They also measure productivity gain. “For example, with the security vulnerability remediation, we know how long each task takes and that we fix 10,000 vulnerabilities across many thousands of files, which equates to 70,000 engineer hours,” he said.
Separately, a specific team is tasked with understanding the AI token spend across the organisation. “Interestingly, 5% of the engineers are spending almost 50% of the tokens, while others are using a fraction or none. I’m more concerned about those that aren’t using any. The cajoling is finished – it’s now mandated,” said Freeman. He added that the team is managing spend tightly and monitoring it day by day.
Ellis believes that the NatWest Boxed team is putting its best foot forward in uncharted territory with some success. “I am confident that we will make massive strides this year, and, looking back, we’ll consider 2026 as the breakthrough year in terms of real clarity. We’re set up to do that,” he said.
Article Info
Related Articles
USD dominates the global currency and trade finance market: SWIFT’s Global Currency Tracker June 2026
The US dollar (USD) remains the undisputed leader in global payments, commanding a dominant 59.10%...
Digital Trade +4Trust, trust … and then more trust – the missing infrastructure of digital trade
By: Dave Meynell, TradeLC Advisory For more than two decades, the trade finance industry has...
Risk Management +3Rebalancing priorities: sustainability in an era of global uncertainty
By: Tim Staheli At the EBRD 2026 Annual Meeting and Business Forum in Riga, the...
Stay Ahead of the Curve
Get exclusive insights, expert analysis, and breaking news on liquidity and risk management, delivered to your inbox
Article Info
Author
Stay Updated
Get the latest insights on trade finance, treasury management, and global payments delivered to your inbox.
Join 25,000+ professionals. Unsubscribe anytime.


