\Recent advances in large language models and agentic AI have changed how executives and boards talk about strategy. This is clear in the large sums now being invested in the technology, with global AI spending expected to reach US$2.52 trillion in 2026.

Current trends show financial services will spend the most, and taken together, finance, software, information services and retail are expected to invest over $1 trillion in 2026.

Despite the high level of corporate interest, early rollouts have revealed a significant tension between the desire for innovation and the practical constraints of the enterprise environment. Aris Kossoras, a partner at KPMG with extensive experience in finance transformation, notes, “There’s still a significant gap of understanding in most functions within financial services organizations about the different types of automation and AI and where AI can really make the difference across enterprise-to-enterprise capabilities.” This knowledge gap often manifests as fragmented execution, where individual departments deploy isolated tools that lack a shared data framework. Without a common technical foundation, these efforts remain siloed. This leads to duplicated costs and prevents the organization from generating the high-level insights required for true firm-wide transformation.

Productivity versus transformation

There’s a difference between using AI to move faster and using it to work differently. Currently, most financial institutions focus on generic AI tools, such as Copilot and ChatGPT, to optimize administrative tasks. Their applications include summarizing meeting minutes and drafting presentations.

When properly implemented, these tools can provide capacity gains of 10 to 50 per cent per employee. However, as Kossoras observes, “This is not AI that redefines how we operate. It’s more about how work is done rather than how we operate.” To achieve true transformation, firms must shift toward “domain-specific AI” that can be embedded in core financial processes to reimagine them. This often involves moving away from massive, all-encompassing models toward small language models tuned to specific datasets relevant to the firm’s unique needs.

Lesson 1: Start with use cases, not data

When Kossoras speaks with financial leadership, the same issue keeps coming up. Their data isn’t ready for AI. However, the most successful organizations do not wait for a perfect information architecture before beginning their AI journey. Instead of launching monolithic, multi-year data programs that move like “oil tankers,” these firms prioritize specific business outcomes.

They start with a handful of clearly defined use cases such as liquidity reporting, client profitability and forecasting and work backward to identify which data actually needs to be fixed. In this model, data is treated as an asset or product in its own right, rather than a process. As Kossoras explains, leading firms “do not care about whether the solution is build or buy, they are instead fixated on the outcome and work backward to define the data products required to deliver the outcome.”

Lesson 2: Prioritize high-impact applications

Rather than spreading resources thinly across dozens of minor experiments, leading firms focus on a small number of applications that can fundamentally move the dial and scale easily. These high-impact solutions often require a multidisciplinary approach, weaving together technologies such as workflow automation, machine learning and changes to how teams operate to solve a complex problem.

A concrete example of this is found in global trader surveillance. Globally significant banks use AI to analyze trillions of internal and external communications in real time, matching voice and text data against trade records to identify potential fraud. This system is sophisticated enough to identify behavioural patterns that can suggest fraudulent activity months before it is actually solidified. By focusing on core capabilities like know-your-client, fraud, balance sheet optimization, liquidity and customer experience, banks prioritize outcomes tied to risk and control over incremental improvements.

Lesson 3: The organizational constraint

The primary barrier to AI integration is rarely a lack of technical capability. In most large financial institutions, the necessary infrastructure is already in place or is at least accessible through existing cloud partnerships and enterprise licenses. The challenge is organizational: how teams are structured and how processes are designed end-to-end.

Across finance functions, digital fluency is still catching up. Kossoras emphasizes that professionals must “work internally, themselves or through their organizations, to get a little bit up the scale on their level of data literacy and digital acumen.” Even where the technology exists and data can be proxied, the constraint tends to be human: “What keeps me up at night is people, then data.”

The speedboat mandate

The evolution of AI in finance is shifting from potential to execution. While the industry has spent the last year captivated by the novelty of large language models, the next phase will be defined by those who can move past incremental productivity and into core workflows while minimizing regulatory risk. The risk of replacing human oversight with unverified model outputs remains a significant concern for boards and regulators alike. Finding the balance between the reward of automation and the necessity of compliance is where the true competitive divide will emerge.

For the modern finance professional, the greatest hurdle is often a psychological one. There is a natural self-preservation instinct at play. Many fear that embracing these tools is a move toward self-replacement, and this tension often results in inertia that can stall even the best-funded initiatives. Kossoras suggests that the industry must look toward new models of employment where the value created by AI is shared. This could involve concepts such as licensing individual data or establishing policy frameworks to protect the workforce.

Ultimately, the mandate for the individual is to move from a defensive posture to one of active adaptation. Data will always be a work in progress, and technology will continue to iterate at a breakneck pace, but the ability to reimagine processes and increase digital fluency is a prerequisite for the new economy. As Kossoras aptly summarizes, your biggest challenge is your people. Success will belong to the practitioners who stop waiting for the perfect data environment and start launching the speedboats of targeted, high-value use cases to pull their organizations in the right direction.

Sebastien Davies, CFA, is a partner at Primal Capital, investing in digital assets and financial infrastructure. He has a background in institutional capital markets and digital asset integration and writes about how technology is reshaping financial systems, with an emphasis on real-world application rather than theory.