Financial Services AI Spending Is Rising, but ROI Is Still Elusive

Financial institutions are committing ever larger budgets to AI, but investment and experimentation have not reliably produced measurable value. Quinlan & Associates’ core message is straightforward: the sector has moved beyond the question of whether to adopt AI; it now has to prove which deployments improve revenue, cost, risk or customer outcomes.

The scale of investment—and the ROI gap

Quinlan & Associates estimates that financial services spent $76 billion on AI-related initiatives in 2025 and projects sector spending will exceed $130 billion by 2030. 14 At the same time, 95% of traditional financial institutions are reportedly either exploring AI or using it in business operations, while 71% say they have difficulty achieving or demonstrating clear ROI. 3

That tension matters. Broad adoption can show organizational intent, but it is not evidence that individual programs are creating durable value. Quinlan-linked findings also report that 43% of AI initiatives are cancelled, underscoring how often activity fails to reach a sustainable production outcome. 15

Why AI initiatives fail to translate into returns

Quinlan identifies three connected obstacles to value realization:

1. Technology-first strategy. Organizations can begin with an available tool rather than a defined business problem, customer journey, workflow bottleneck and value target. That makes it harder to select use cases with a credible path to impact. 6
2. Governance introduced too late. When risk, compliance and control are treated as an approval step after design has begun, teams may discover constraints only after substantial time and money have been committed. 6
3. Operating models not built for scale. Fragmented data, weak implementation foundations and insufficient capability to govern, deploy and scale AI can leave promising work stuck between pilot and enterprise adoption. 6

These are execution constraints rather than simple technology constraints. An institution can license models, run proofs of concept and report many AI experiments without redesigning the workflows and accountability needed to capture a financial result.

Governance is part of the ROI mechanism

For financial services, trustworthy AI governance is not separate from commercial performance. It helps determine whether a use case can be approved, deployed, monitored and trusted at the level of autonomy its business case requires.

The World Economic Forum’s financial-services AI playbook recommends embedding risk management and responsible-AI practices at every layer: define risk appetite early, validate models rigorously, and keep AI actions explainable, traceable and controllable. 1 These practices are especially important as firms move from assistive systems toward systems that can take multi-step actions. Clear boundaries, escalation paths and human oversight can reduce the operational uncertainty that prevents valuable use cases from scaling.

Industry evidence also reinforces the importance of foundations. The IIF-EY survey identifies data quality and data availability as leading challenges to deploying AI in production, alongside the need for skilled staff and governance safeguards. 4

How the findings compare with BCG and NVIDIA

The broader research landscape supports Quinlan’s diagnosis, although it uses different samples and measures.

BCG’s survey of more than 280 finance executives found a median reported AI and generative-AI ROI of 10%, below the 20% many respondents targeted; nearly one-third reported only limited gains. 43 In a separate study of more than 1,250 companies, BCG found that only 5% were achieving AI value at scale, while 60% reported little material value despite substantial investment.

NVIDIA’s 2025 financial-services survey presents the other side of the picture: some deployed use cases are generating tangible benefits. Nearly 70% of respondents said AI had driven revenue increases of at least 5%, and more than 60% said it had reduced annual costs by at least 5%. Trading and portfolio optimization and customer experience were the leading generative-AI use cases by reported ROI.

Taken together, the studies do not conflict. AI can generate returns, but deployment is not itself the outcome. The difference lies in whether a firm has selected a high-value use case, changed the relevant workflow, built reliable data and controls, and measured the operational and financial result.

The shift from adoption to execution

The clearest signal from Quinlan’s report is the mismatch between spending, engagement and realized value: a $76 billion 2025 investment base, widespread institutional AI activity and persistent difficulty proving returns. 14 3

For financial-services leaders, the practical response is to treat AI as a business-transformation program rather than a portfolio of disconnected pilots. Start with a measurable business objective; establish risk and governance requirements before development; assign ownership for outcomes; and track lifecycle metrics after deployment. BCG similarly argues that firms creating significant value focus on a small set of initiatives, scale them quickly, change core processes, upskill teams and systematically measure operational and financial returns.

The next competitive divide is therefore unlikely to be between institutions that have AI and those that do not. It will be between those that can operate AI safely and measurably in core workflows—and those that continue to accumulate experiments without a repeatable route to value.