Introduction
The customer is a Tier-1 global bank, a systemically important institution running online and mobile banking for millions of customers, where a single broken flow can mean a customer who can’t move money, a compliance gap, or a headline.
At an institution this size, software defects are one of the largest hidden costs in the business: a bug caught in production costs 10-100× more to fix than one caught in development, and for a bank of this scale the aggregate cost of defects runs into the hundreds of millions of dollars a year. Meanwhile the engineering organisation had begun using generative AI to write code faster than ever, but quality assurance hadn’t kept pace. Every accelerated release widened the gap between how fast the bank could ship and how fast it could verify.
The Challenge
E-banking isn’t a single flow. Login and authentication, payments and transfers, standing orders, statements, card management, onboarding: each spans web and mobile, multiple device types, and strict regulatory requirements. Multiply those paths together and the number of journeys that must work on every release climbs far faster than any manual QA team can follow.
The bank had already invested in scripted automation with Playwright. But writing Playwright tests from scratch takes time, and maintaining them takes even more. As the product evolved, every UI change broke tests, and engineers spent their days repairing a suite rather than expanding coverage. The result was familiar: automation that was fragile, coverage that lagged the product, and a QA effort that had quietly become a release bottleneck — the exact opposite of what an AI-accelerated engineering org needs.
The Solution
Turning a fragile script suite into a self-healing platform
The bank deployed Klarent’s autonomous QA agents against its live e-banking environment. Rather than replacing the Playwright investment, Klarent supercharged it, taking the same end-to-end flows and running them through agents that generate, execute, and repair tests automatically, cutting test creation from hours to seconds.
Autonomous navigation of real banking journeys.
Klarent’s agents worked through the bank’s e-banking flows exactly as a customer would (authenticating, navigating menus, completing transactions), and reported back on what worked and what didn’t, with screenshots and a full step trail attached to every run. In the initial engagement, the agents autonomously covered 80 of the bank’s highest-risk revenue and compliance journeys across web and mobile.
The agents successfully navigated autonomously through our e-banking flows.
QA Lead
Tier-1 Global Bank
Tests written in plain language.
Because each test is described in natural language rather than code, scenarios are readable and reviewable by anyone, not just automation specialists. That removes the specialist bottleneck that caps how fast coverage can grow, and it’s what makes the platform’s ~100× efficiency over manual testing possible.
Self-healing that ends the maintenance treadmill.
When the product changed, the agents adapted the tests automatically instead of failing. Across the entire engagement the bank carried a zero maintenance backlog, and the recurring cost that had made its Playwright suite so expensive to own simply disappeared.
Writing Playwright tests from scratch takes time, but maintaining them takes even more. Klarent freed our team from the maintenance burden. We now have great coverage that heals itself when the product changes.
QA Lead
Tier-1 Global Bank
Enterprise-grade from day one.
The deployment ran in a hybrid model with local execution, tenant isolation, encryption, audit logs, and Swiss/EU data residency, built to sit inside the bank’s existing security and regulatory perimeter.
The Impact
From a fragile bottleneck to a P&L lever
The engagement changed the economics of QA at the bank. Manual QA effort on the covered journeys fell by roughly 90%, and because the suite is self-healing, that reduction wasn’t traded for a maintenance burden later — the bank ended with zero maintenance backlog. Test creation that used to take hours now takes seconds.
Just as important as the effort savings is the change in posture. QA moved from an on-request, manual gate (run when there was time before a release) to continuous coverage that keeps pace with how fast the bank now writes code. Once a test exists, it costs no human effort to run again, so the team can validate the critical surface before every release, after every hotfix, or on a schedule, as often as they like. Every incident caught before it reaches production is a cost, potentially in the millions, that never lands.
For a bank of this scale, that combination is what turns QA from a cost line into a P&L lever: Klarent’s modelled projections at full enterprise rollout put the opportunity at up to ~$10M a year in hard savings, a reduction from roughly 100k to 5-8k annual QA engineering hours, and tens of millions more in avoided production incidents.
Savings and engineering-hour figures above are Klarent’s modelled projections at full enterprise rollout, based on the bank’s environment and the results of the initial engagement — they are forward-looking, not yet-realised results.
What’s Next
Building on the results, the bank is expanding coverage from its initial set of flows to all documented critical journeys, integrating Klarent deeply into its engineering workflows and CI/CD pipelines, and extending autonomous QA to more applications across the organisation. As coverage scales, so does the return, toward the ~90% reduction in QA engineering hours and the multi-million-dollar annual savings modelled above.
The end state is a governance layer that moves at the speed of AI: the bank keeps shipping quickly, while quality stops being the thing that slows innovation down, and starts being a measurable advantage.
Note: Customer details have been anonymised at the customer’s request.
