Banks spend $40B on AI but only 20% see sustained value

Financial institutions poured more than $40 billion into artificial intelligence initiatives last year, yet only one in five bank leaders report seeing widespread, sustained value from those investments, according to research from Accenture.

The disconnect stems from a fundamental implementation problem: banks are distributing AI tools to individual employees without redesigning the underlying work processes, says Mike Abbott, Accenture's global banking lead. "They're giving AI tools to individuals and saying, 'Hey, do your job a little bit better,'" Abbott explained. "Task-wise productivity – getting 10%, 15% for each person – does not add up to system-wide productivity."

Why it matters

As AI costs mount and enterprise software budgets face scrutiny, banks will need to demonstrate tangible returns or risk executive backlash. The shift from experimentation to optimization will likely accelerate in 2027, forcing financial institutions to justify model choices and prioritize use cases that deliver measurable business outcomes rather than incremental efficiency gains.

Where banks find AI value today

Accenture surveyed 212 retail banking and 110 capital markets executives across 20 countries between April and June. The research found banks currently focus more on expense reduction than revenue generation, with the most successful implementations concentrated in repeatable, scaled processes.

Those include call center operations, underwriting workflows, marketing content generation, and regulatory reporting automation. Yet even these applications haven't translated into the transformative productivity gains executives anticipated.

FOMO drives continued investment

Despite limited returns, a "fear of missing out" mentality continues to fuel AI spending across the banking sector. "There's a high level of confidence they're going to get something out of it, but it hasn't exactly been measurable just yet," Abbott said.

The competitive nature of banking means institutions feel compelled to invest even without clear ROI metrics. Abbott expects this dynamic to shift as leading banks demonstrate successful implementations: "Once one figures it out … the rest will figure it out. Banks are pretty good at copying each other."

The optimization phase begins

Sophisticated banks are already moving beyond indiscriminate spending toward strategic model selection. Rather than defaulting to large language models for every task, they're deploying smaller, open-source models for document processing and basic functions – approaches that reduce costs and hallucination risks.

Some institutions are also developing transactional foundational models using proprietary data, creating competitive advantages that can't be easily replicated.

Abbott predicts budget structures will evolve to reflect this new reality. Instead of separate line items for headcount and AI tools, leaders will receive combined budgets and authority to optimize the mix themselves.

Serial to parallel: the coming transformation

The real breakthrough will come when banks redesign processes from serial to parallel execution. Traditional Six Sigma methodologies optimized for limited human resources, creating sequential workflows with multiple handoffs and gates.

AI agents can execute these steps simultaneously. A mortgage application that currently moves through stages one at a time could be processed instantaneously across multiple dimensions. A single AI agent handling lost card inquiries could serve call centers, mobile apps, websites, and branches simultaneously rather than requiring separate implementations for each channel.

Software development represents perhaps the largest opportunity, as the traditional design-build-test-deploy sequence becomes parallelized. Abbott notes that elite banks are already demonstrating this capability, with mainstream adoption expected in 2027.

Accenture's findings and Abbott's analysis were first reported by Banking Dive.