AI becomes the banker: Decision-making and the new testing burden. This three-part series explores how AI is reshaping banking end-to-end, and the growing testing, risk and assurance challenges facing QA teams as deployment moves from pilot to production. Today Part III. Please click here for the first instalment and part II. If the first phase of AI in banking transformed interfaces, and the second embedded intelligence into operational control, the third is pushing AI to the very edge of decision-making, where money moves, credit is granted and customer intent is interpreted in real time. This is the most consequential stage of AI adoption. Here, systems are no longer assisting processes; they are directly shaping financial outcomes. Voice assistants are replacing menus, messaging apps are becoming payment rails and AI models are determining who gets access to credit and at what cost. For QA and software testing teams, this represents a step-change in responsibility. Testing is no longer about validating workflows or even complex system interactions, it is about assuring outcomes in environments where inputs are unstructured, behaviour is probabilistic and errors carry financial, regulatory and reputational consequences. The challenge is compounded by scale. These systems are designed to operate continuously, across millions of users, often without human intervention. That makes observability, explainability and fail-safe design critical components of any testing strategy. "2024 was a pivotal year for AI in banking, as institutions worldwide moved from pilot projects to real deployments that tangibly improved the digital customer experience," stressed Alex Kreger, founder and CEO of UXDA Financial UX Design. He added that banks observed AI could "boost both customer satisfaction and operational performance - a true win-win." Voice and conversational banking go mainstream. Lunar, the bank that launched Europe's first GenAI-powered voice assistant for banking, uses a voice-native GPT-4 model to hold natural conversations and can handle both simple and complex requests, including interruptions and follow-ups. This type of interface introduces entirely new testing requirements, according to Kreger. QA teams must validate not just functional outcomes, but conversational flow, intent recognition across accents and languages, and the system's ability to recover gracefully when misunderstandings occur. Meanwhile at Nubank, AI has been embedded across operations through the use of OpenAI models. Kreger said the bank is enhancing "both customer experience and operational efficiency," with AI handling 55% of Tier 1 inquiries and reducing response times by 70%. The same bank also pushed AI into payments through Pix integration. Kreger said customers can send payments using "voice, text or even images through WhatsApp, with AI handling the interpretation and execution." For QA teams, this is a fundamental shift. Payment initiation is no longer tied to structured input fields but to natural language and even visual cues, requiring entirely new validation models to ensure accuracy, consent and security. "Institutions worldwide move from pilot projects to real deployments." - Alex Kreger At Upstart, AI-driven underwriting is redefining how credit decisions are made. Kreger said its models enabled "44% more loan approvals at 36% lower average APRs for the same risk level," compared with traditional approaches. He added that many loans are now approved instantly with no human intervention, demonstrating how AI can simultaneously increase efficiency and expand access to credit. However, this also introduces one of the most complex testing challenges in financial services. AI-driven credit decisions must be tested not only for performance, but for fairness, bias, transparency and regulatory compliance. Models must be explainable, auditable and resilient to drift over time. Predictive and proactive systems. At Commonwealth Bank of Australia, AI is being used to reduce fraud and scam losses through tools like NameCheck and CallerCheck. Kreger said the bank achieved "a 50% reduction in scam losses, a 30% drop in fraud cases and a 40% decrease in call center wait times." At Capital One, the Eno assistant has expanded into voice and chat-based banking, handling routine requests while providing proactive insights. Kreger said it reduced "call center contact volumes by 50%" while improving response speed and consistency. These cases highlight a key trend: AI is no longer reacting to fraud, it is anticipating it. That requires QA teams to test systems against evolving threat patterns, adversarial behaviour and complex real-world scenarios that cannot be fully replicated in controlled environments. ""Across the global banking industry, AI has shifted from pilot to profit generator." - Alex Kreger As AI moves into payments, credit and real-time decisioning, the testing burden increases sharply. Systems must not only perform correctly, they must behave responsibly under uncertainty. Kreger acknowledged the challenges, noting that "banks had to address customer trust and security concerns around AI," and that "incidents of AI errors, or 'hallucinations', were taken seriously, and institutions implemented guardrails and testing to maintain accuracy." He added that the human element remains essential, with leading banks combining AI with human oversight to ensure that customers can escalate when needed. A new QA mandate. The final set of case studies makes one thing clear: AI is no longer a feature within banking systems, it is becoming the system itself. From how customers interact with their bank, to how decisions are made and risks are managed, AI is now embedded across the entire value chain. "Across the global banking industry, AI has shifted from pilot to profit generator," Kreger said. "Intelligent tools now accelerate back-office work, raising employee productivity and shrinking response times, while smarter analytics empower advisors to deliver faster, better-informed guidance," he added. Kreger also pointed to the scale of impact, noting that instantaneous self-service resolves most issues in under two minutes, trims wait times by 40 percent and cuts repetitive contact by 55 percent, while AI-driven systems improve fraud detection and expand access to credit. The implication for QA and testing teams is profound. Assurance can no longer be retrospective. It must be continuous, embedded and aligned with both technological and regulatory expectations. As Kreger concluded: "The 'age of AI' in banking has truly begun," and the institutions that succeed will be those that can not only deploy AI at scale, but prove, consistently and rigorously, that it works as intended. WHY not become a QA Financial subscriber? It's entirely FREE Watch now. May 15, 2026 May 14, 2026 May 13, 2026 May 12, 2026

Lunar co-founder and CEO Ken Villum Klausen is stepping down after leading the Danish digital bank since its founding in 2015. He will be replaced by Søren Kyhl, a former Danske Bank and Saxo Bank executive, who takes over in June. Klausen said the timing "feels right", noting Lunar is now the biggest challenger bank across the Nordics with over one million customers and more than 40,000 business clients. The company has raised over €500 million in funding. Kyhl spent 10 years at Saxo Bank, most recently as deputy CEO, and over 13 years at Danske Bank. Last year, Lunar experienced several executive departures, including its co-founder and CTO, as it pursues long-term profitability whilst cutting management costs.