Beyond automation: Is AI forcing banks to confront QA culture gaps?

Michiel Willems
April 17, 2026

Artificial intelligence is rapidly embedding itself into the software testing stacks of banks and financial institutions, promising faster test creation, broader coverage and more efficient pipelines.

But without addressing underlying cultural and organisational weaknesses, firms risk amplifying the very problems they are attempting to solve, according to Mudit Singh, co-founder and head of growth at TestMu AI.

“AI speeds up test creation and finds edge cases faster while handling the repetitive work that drains a team’s energy,” Singh said.

“What it doesn’t do is fix misaligned incentives, clarify who owns quality or rebuild trust in a test suite that people have learned to ignore,” he added.

As financial institutions scale AI across development and testing environment, often under increasing regulatory pressure around resilience and operational risk, the warning reflects a growing concern: weak QA foundations do not disappear with automation, but instead become more visible and more persistent.

“When your QA culture is weak, AI exaggerates the problems you already have,” Singh stressed. “You get more test artifacts, more noise in your pipeline and the same production issues with every release.”

Legacy weaknesses in testing

Within banking environments characterised by legacy systems, complex release cycles and regulatory scrutiny, Singh argued that AI’s reliance on existing data and patterns presents a structural limitation.

“AI learns from your existing patterns and simply produces legacy code at machine speed rather than improving the tests,” he said. “You end up curating a massive pile of low-value tests that need constant updates.”

This dynamic raises particular concerns in a regulatory context where the effectiveness of testing, not just its scale, is increasingly under scrutiny.

“Developers rush to meet deadlines. QA gets squeezed at the end.”
– Mudit Singh

Singh also highlighted persistent gaps in how organisations measure and prioritise quality outcomes.

“Fifty percent of organisations do not track the cost of bugs that reach production,” he noted. “Quality gets cut first because upper management has a bird’s-eye view and pushes for speed. Developers rush to meet deadlines. QA gets squeezed at the end.”

In such environments, AI risks reinforcing reactive testing practices rather than enabling more proactive quality engineering.

“Viewing QA as downstream cleanup means you will deploy AI as a cleaner,” Singh says, “which can be expensive, as you’re still creating reactive tests.”

Organisational maturity

For institutions investing in AI-driven QA tooling, Singh emphasised that the technology’s effectiveness remains dependent on existing engineering and testing foundations.

“AI doesn’t replace these capabilities, which means it’s building on top of whatever foundation you already have,” he said, referencing maintainable test suites, reliable automation and consistent release standards.

He described AI as comparable to a high-output but imperfect contributor within a team.

“Think of AI like bringing in a high-output teammate who produces fast but has imperfect judgment,” Singh explained. “In a team with clear standards and quick feedback, the speed of output becomes beneficial.”

“But in a team that argues about what ‘done’ means every sprint, AI becomes another stream of work that nobody can trust,” he added.

For banks operating under strict release governance and regulatory expectations, ambiguity around release readiness can translate directly into operational and compliance risk.

Test strategies under pressure

Singh pointed to the need for clearer governance structures and ownership models as a prerequisite for effective AI adoption in QA.

“Strong QA culture runs on shared release criteria that spell out what must be true before shipping, what can wait and how you handle risk when things get tight,” he wrote in a recent Forbes Council analysis.

He also highlighted the importance of embedding quality ownership across delivery teams.

“When developers ‘throw code over the wall’ and QA ‘catches bugs,’ AI just increases throughput in the catching phase,” Singh said. “End-to-end ownership means quality gets built and verified inside the delivery team.”

Test suite management emerges as another critical pressure point as AI accelerates output.

“Without governance, AI generates more tests faster than you can maintain them,” he warns. “Coverage goes up on metrics, confidence drops in your gut, and people start ignoring failures.”

“That’s when automation becomes performance art.”

For financial institutions already grappling with test suite sprawl and automation reliability issues, this scenario risks undermining confidence in testing outcomes at a time when regulators are demanding stronger assurance.

AI dependency

Singh also cautioned against over-reliance on AI tools without corresponding investment in internal expertise.

“If your plan is ‘buy AI and the problem goes away,’ then you’ll end up dependent on a tool you can’t govern or debug when it produces nonsense,” he warned.

The challenge is compounded by ongoing skills gaps in AI-enabled testing, which risk limiting organisations’ ability to validate and control AI-generated outputs.

“End-to-end ownership means quality gets built and verified inside the delivery team.”
– Mudit Singh

Despite the risks, Singh maintained that AI can deliver tangible benefits when applied within clearly defined boundaries.

“AI works when it reduces busywork and tightens feedback loops,” he pointed out.

“The World Quality Report notes that 71% of respondents use AI-driven tools for repetitive tasks like test data generation and test analysis. This works because humans stay in control of intent and risk.”

In practice, this includes using AI to support, rather than replace, engineering judgment and domain expertise.

“Generate candidate test cases and review them against a risk model,” Singh recommended. “Ask AI to suggest edge cases and validate them with domain knowledge.”

“Use AI to assist triage by clustering failures, and let engineers drive root cause analysis,” he concluded.

As banks continue to balance accelerating software delivery with tightening regulatory oversight, Singh’s assessment underscores a broader industry reality: AI may enhance QA capabilities, but without strong cultural and organisational foundations, it risks scaling inefficiency rather than improving quality.