As AI compresses software development timelines across banking and financial services, the pressure is moving sharply onto validation, with QA teams facing a harder task: proving that faster releases can still be trusted in highly regulated environments.
That shift is especially acute for banks because AI is accelerating digital transformation while leaving the cost of failure untouched, according to Rohit Raghuvansi, chief technology officer at Leapwork, whose clients include financial giants such as BNP Paribas, Blackstone, the London Stock Exchange, Bain Capital, Investc, Paypal and Credit Suisse.
“It is a very significant shift for banks, probably more than for many other industries because a lot of digital transformation will be accelerated by AI,” Raghuvansi told QA Financial.
Historically, he noted, “one of the biggest costs in banking transformation was simply creating and migrating software,” as large teams spent years modernising legacy estates, rebuilding workflows and moving applications into newer environments. Now, however, “AI changes that equation.”
“It lowers the cost and time needed to create, translate, migrate, and modernize software,” Raghuvansi explained. “So the problem is no longer only, ‘Can we build or migrate this?’ The harder question becomes, ‘Can we trust what we are about to release into a highly regulated, highly interconnected environment?’”
That question lands at a moment when AI is already a priority across enterprise software teams, but validation infrastructure is lagging behind.
Leapwork’s own research found that 88% of software development professionals say AI is a priority for their organisation, yet only 12.6% use AI across key testing activities today, while 59% of testing remains manual.
Validation moves to the centre
For banks, Raghuvansi argued, this is not just a tooling problem but a release-risk problem.
“That matters a lot in banking because release cycles are accelerating, but the tolerance for failure has not changed,” he said.
“In fact, in some ways it has become even lower. A defect in banking is not just a bug. It can become a customer impact issue, an operational disruption, a compliance problem, or a reputational problem.”
Raghuvansi clarified: “So yes, I do think the bottleneck is shifting from creation to validation, and in banking that shift is especially pronounced because AI is speeding up change, while the cost of getting trust wrong remains extremely high.”
“One of the biggest costs in banking transformation was simply creating and migrating software. Now AI changes that equation.”
– Rohit Raghuvansi
In his view, the economics of AI-driven development make that unavoidable. “AI is compressing the cost of software creation. It is not compressing the cost of release failure. That is why continuous validation becomes even more strategic for banks.”
That position builds on concerns Leapwork has raised previously about fragile automation and the risks of unstable validation in regulated environments.
Earlier this year, chief executive Kenneth Ziegler said that “in banking, you cannot introduce variability into the validation layer,” adding that “AI can accelerate testing, but trust is earned through consistency.”
Why testing still lags development
Raghuvansi said one reason validation continues to trail development is structural.
“Validation has historically been organized as a downstream activity. Development happens first, then validation starts,” he elaborated. “That means validation has been cyclic, dependency-heavy, and often forced to wait for development completion before it can really move.”
That leaves testing vulnerable whenever release pressure rises. “In practice, development was always the long pole in the tent,” Raghuvansi stressed.
“It took the most time, so when release pressure increased, validation was often the part that got compressed.”
He also pointed to fragmentation across enterprise estates, a familiar issue for banks juggling legacy systems, cloud services, APIs and newer AI-driven tooling. “Validation work is often spread across multiple tools, teams, and environments,” Raghuvansi continued.
“Test creation happens in one place, execution in another, defect analysis in another, performance testing somewhere else, and release confidence often lives in spreadsheets, meetings, or individual judgment. That creates a real efficiency bottleneck.”
Too much effort, he added, still goes into low-value maintenance work rather than real assurance.
“Too much validation effort still goes into low-value mechanics: creating scripts, maintaining artifacts, fixing brittle automation, and moving data between systems.”
That problem is reflected in the company’s study, which found that 71% of teams say test creation slows them down the most, 56% cite test maintenance, and 45% say updating tests after a change in a critical system takes three or more days.
From test execution to release trust
Leapwork is framing its latest platform launch around the idea of “continuous validation,” rather than simple continuous testing. Raghuvansi drew a clear distinction between the two.
“Traditional continuous testing is an important step, but it is still a narrower idea,” Raghuvansi said. “It often focuses on automating test execution continuously through the delivery pipeline.”
By contrast, “continuous validation is broader,” he pointed out. “It is about continuously proving that software can be trusted as it changes.”
“That includes functional correctness, but it also includes system behavior, resilience, performance, downstream outcomes, governance, and auditability.”
“QA needs to move from only validating exact outputs to validating behavior, boundaries, and acceptable outcomes.”
– Rohit Raghuvansi
For banks, where releases are judged not only on whether tests passed but also on whether controls held up across connected systems, that broader definition matters.
“A bank is not just asking, ‘Did the test pass?’” Raghuvansi stated. “It is asking: Did the system behave correctly across multiple applications and workflows? Did the release preserve expected controls? Did performance remain acceptable under real conditions? Did the right downstream state changes occur? Can this be explained and audited? Is this safe enough to release into a regulated environment?”
His conclusion is blunt: “Continuous testing is about continuous execution of checks. Continuous validation is about continuous proof of release trust.”
QA mindset
That challenge becomes harder still as firms deploy more non-deterministic AI systems into production.
“This is one of the most important mindset shifts,” Raghuvansi remarked. “QA teams cannot approach AI-driven systems in the same way they approached deterministic software.”
In conventional environments, he said, testing tends to follow a binary logic of “same input, same output, pass or fail.” With AI-driven systems, especially those interpreting intent or generating outputs, “that logic becomes less absolute.”
Instead, “QA needs to move from only validating exact outputs to validating behavior, boundaries, and acceptable outcomes.”
That means focusing less on whether a model produces a single expected answer and more on questions such as whether “the system interpret[ed] the request appropriately,” whether it “choose[s] the right action or workflow,” and whether outcomes remain within policy, compliance and governance constraints.
“This is where the role of QA becomes more strategic,” Raghuvansi said. “The job shifts from script writer to system validator.”
Raghuvansi summed up the challenge in one line: “AI is probabilistic. Enterprise trust cannot be.”
That concern is echoed by the company’s survey data, which found that 54% of respondents cite accuracy and quality concerns as the primary barrier to broader AI adoption in testing.
“QA teams cannot approach AI-driven systems in the same way they approached deterministic software.”
– Rohit Raghuvansi
As enterprise systems become more interconnected and more agentic, Raghuvansi said QA teams are being forced to validate full business workflows rather than individual screens or functions.
“Yes, absolutely. That is one of the biggest shifts happening,” he firmly stated.
“In a more connected and agentic world, it is no longer enough to validate one isolated application screen or one narrow function,” he added.
“QA teams increasingly need to validate end-to-end workflows, system interactions, orchestration paths, and business outcomes.”
That is especially relevant in banking, where a single AI-assisted workflow may span customer interfaces, internal decisioning layers, back-end systems and audit controls. “The real question is, ‘Did the whole chain behave correctly, safely, and reliably?’” Raghuvansi said.
That means validation is moving “from application-level testing to system-level trust,” he added.
Validation becomes a board-level issue
The organisational consequences, he suggested, are now unavoidable. Banks can no longer afford to treat validation as a downstream QA function.
“Banks need to stop thinking about validation as the last mile of delivery and start treating it as a core capability for software trust,” Raghuvansi said.
He argued that as investment shifts toward software, automation and AI tooling, validation platforms must increasingly be seen as a software category in their own right.
“If more capital is going into software creation, software acceleration, and autonomous workflows, then more capital also has to go into software trust.”
In his words, “continuous validation” is becoming “one of the most important areas of software investment.”
That reflects a broader budget shift described in Leapwork’s release, which says QA and validation are rising from 25–35% of software budgets to 35–40% in the AI era.
Raghuvansi said leadership sponsorship will be essential if banks are to keep up. “Validation needs stronger ownership from CIO, CTO, risk, and quality leadership,” he shared.
“In organisations, software trust is not just a QA concern. It is an operational resilience concern, a governance concern, and increasingly a strategic investment concern.”
His bottom line is that AI is not removing the need for control. It is intensifying it.
“Validation has to move from being a downstream QA activity to being an enterprise trust capability,” Raghuvansi concluded, “and from being treated as supporting tooling to being treated as a top software investment category.”
WHY not become a QA Financial subscriber?
It’s entirely FREE
* Receive our weekly newsletter every Wednesday * Get priority invitations to our Forum events *
READ MORE
- Trust, not speed: Why AI governance is now a testing battleground for banks
- NatWest’s AI trade finance overhaul opens new chapter for QA teams
- Banking UAT moves beyond sign-off as QA takes centre stage in system rollouts
- Citi ramps up AI-driven testing in race to modernise legacy systems
- Lloyds, HSBC and NatWest get OpenAI access amid mounting concerns
WATCH NOW
QA FINANCIAL PODCASTS
CLICK HERE TO LISTEN TO OUR EXCLUSIVE CONVERSATIONS
