Bank of America is accelerating its use of artificial intelligence across advisory, payments, and internal engineering workflows, offering one of the clearest case studies yet of how AI is moving from peripheral tooling into core banking operations, and, in turn, forcing a rethink of how those systems are tested, governed and validated.
Folowing thorough testing cycles, the bank is now deploying an internal AI-powered advisory platform to around 1,000 financial advisers, marking a shift from earlier chatbot-style deployments toward systems that directly support decision-making.
The platform, based on Salesforce’s Agentforce, is designed to help advisers handle client queries, prepare recommendations, and manage workflows in real time, reflecting a broader industry push to embed AI agents into day-to-day banking roles.
At the same time, Bank of America’s wider AI estate continues to expand rapidly, from customer-facing assistants to developer tooling and fraud systems, creating a complex, high-volume environment where QA, model validation and operational resilience are becoming inseparable.
AI at Bank of America is no longer confined to back-office automation or simple customer support. The bank’s virtual assistant Erica, first launched in 2018, has evolved into a central interface for both retail and business banking clients.
The scale is significant. Erica now handles interactions equivalent to tens of thousands of staff hours, with usage embedded across both consumer and business channels.
Holly O’Neill, president of consumer, retail and preferred lines of business, pointed out that “the two million daily consumer interactions with Erica save the bank the equivalent of 11,000 staffers’ daily work.”
She stressed that at the bank the entire infrastructure is subject to continous and rigorous testing.
Internally, the system has shifted from reactive chatbot to proactive decision-support engine, with deep integration into the bank’s architecture.
As Hari Gopalkrishnan, chief technology and information officer, explained: “So, if you came in via Erica but you wanted to then talk to someone in the call center, you could seamlessly go there.”
This progression is mirrored in the new advisory agent rollout, where AI is no longer just answering questions but shaping financial recommendations.
That transition materially changes the risk profile from a QA perspective: validating conversational accuracy is no longer enough when systems influence financial advice.
Digital scale and AI-driven engagement
The bank’s AI push is underpinned by massive digital adoption, which in turn creates a high-throughput testing environment for AI systems operating at scale.
Clients interacted with Bank of America approximately 30 billion times over the past year through digital logins and alerts, a 14% year-on-year increase.
Erica alone saw 20.6 million users generate nearly 700 million interactions in a single year, with total interactions surpassing 3.2 billion since launch.
“Our digital capabilities are a cornerstone of how we serve and deepen relationships with our clients and empower millions to manage their finances,” argued Nikki Katz, Head of Digital at Bank of America.
“We use innovative technology to meet and anticipate the needs of our clients,” Katz added.
“By understanding how their behaviors are changing, and what they’re trying to achieve, we can advance our digital and AI capabilities to deliver personalied insights and proactive advice.”
This level of engagement places sustained pressure on testing pipelines. Systems must operate reliably across billions of interactions, while maintaining accuracy, consistency and explainability in AI-driven outputs.
The move toward proactive AI, where systems initiate actions rather than respond to prompts, further complicates validation and monitoring.
AI integration
Beyond customer-facing tools, AI is now embedded across Bank of America’s internal engineering and operational stack.
Around 18,000 developers are using AI coding tools, improving productivity by approximately 20%, while AI assistants such as AskGPS are being deployed internally to provide instant access to business intelligence.
“We’ve combined thousands of vetted internal documents, so our associates have access to instant business intelligence,” disclosed Mark Monaco, head of enterprise payments and head of global transaction business.
The bank is also applying AI to core infrastructure components. “We’ve deployed a large language model to make OCR better,” he said. “It’s much more accurate. We’re getting much better data out of it.”
At a strategic level, executives see AI as fundamentally different from previous waves of automation.
Jeff Busconi, head of wealth management strategy, products and services, added that “this is really different from past cycles of innovation where the automation or the digitization was really focused on lower-cost high-volume areas like operations. Here, with AI, the opportunity is much broader.”
For QA teams, this proliferation of AI across multiple layers, front-end interfaces, internal tools, and core processing systems, introduces complex interdependencies that must be tested end-to-end.
Testing AI
Bank of America’s approach highlights how testing requirements evolve as systems shift from rules-based logic to probabilistic AI models.
In fraud detection, for example, the bank “adopted a sophisticated AI system to monitor and analyse real-time transaction data across multiple channels,” moving toward behavioural pattern recognition rather than static rules.
This transition has direct implications for QA. Traditional validation approaches focused on deterministic outputs are no longer sufficient. Instead, testing must account for drift, bias, false positives, and explainability under real-world conditions.
The bank’s broader automation strategy reinforces this shift. Its robotic process automation programme is built around the principle that systems must be “testable, maintainable, and auditable,” with agile methodologies enabling rapid feedback and early identification of failure modes.
BofA stated that Agile methodologies for development, testing, and deployment of the robots would be the best approach to managing the programme.
In practice, this means testing must simulate complex workflows, edge cases, and exception handling across legacy and modern systems.
As AI agents and language models are introduced, those requirements extend further into adversarial testing, semantic validation and model behaviour under uncertainty.
Governance
As AI moves into client-facing and decision-support roles, governance and oversight are becoming central to deployment.
Bank of America has established an AI Council and applies a structured risk framework to new use cases, including checks for “bias, intellectual property, transparency, and explainability.”
The bank also enforces human-in-the-loop controls and validates whether systems “abide by those risk frameworks.”
For QA teams, this elevates testing artefacts into compliance evidence. Test coverage, failure scenarios, fallback logic and audit trails are no longer purely technical outputs but inputs into regulatory and risk reviews.
The need for explainability is particularly acute in advisory contexts, where AI-generated recommendations must be defensible.
As AI agents begin to influence financial decisions, the requirement to test not just outcomes but reasoning paths becomes critical.
High-performance testing
Looking ahead, Bank of America is positioning for further shifts in computational capability that will reshape testing environments.
“Quantum computing could be one of the biggest revolutions yet,” argued Haim Israel, equity strategist at BofA Global Research.
“It uses sub-atomic particles to store information and uses superpositions for complex calculations.”
The bank believes quantum systems could “master almost every aspect of banking,” including optimisation, analytics and fraud detection.
“A human would have to perform one sum every second for around 50 quintillion years to equal what a quantum computer can do in a single second,” Israel noted.
Meanwhile, Vanessa Cook, strategist at the Bank of America Institute, added that “the convergence of AI and quantum technologies can enable fundamental improvements in the physical world as well as the digital one.”
In parallel, high-performance computing and edge architectures are already reshaping AI testing environments. “HPC systems can perform quadrillions of calculations per second,” Israel explained, while Cook noted that “edge computing addresses latency, bandwidth, autonomy, and privacy requirements.”
For QA teams, this introduces new dimensions of validation, from distributed inferencing at the edge to large-scale simulation of AI models under real-world conditions.
Bank of America’s rollout of AI agents into advisory roles, combined with its broader AI and digital expansion, illustrates a shift already underway across financial services.
AI is no longer a layer on top of banking systems. It is becoming embedded within them, influencing decisions, workflows and client interactions at scale.
For software testing teams, the implication is clear: testing strategies must evolve alongside AI adoption. Systems must be validated not only for correctness, but for behaviour, resilience, explainability and governance compliance.
As Bank of America’s trajectory shows, the challenge is no longer whether AI can be deployed in banking, but how it can be tested, controlled and trusted once it becomes part of core operations.
NEXT MONTH IN LONDON
REGISTER TODAY – SIMPLY CLICK HERE
Why not become a QA Financial subscriber?
It’s entirely FREE
* Receive our weekly newsletter every Wednesday * Get priority invitations to our Forum events *
REGULATION & COMPLIANCE
Looking for more news on regulations and compliance requirements driving developments in software quality engineering at financial firms? Visit our dedicated Regulation & Compliance page here.
READ MORE
- Goldman puts AI coding to the test
- How to test AI models that banks do not control
- OpenAI, Filigran and SunTec: the latest vendor and product news
- Sygnum: Testing AI is ‘a measurement problem’
- Banks’ ‘code for all’ push raises testing risks
QA FINANCIAL PODCASTS
CLICK HERE TO LISTEN TO OUR EXCLUSIVE CONVERSATIONS
