Agentic AI in Finance: How Data Leaders Are Scaling Safely

Across Europe, data leaders in financial services find themselves walking a tightrope – eager to implement and scale AI tools, yet constrained by compliance, risk management, and the challenge of proving tangible value. According to our CDO Insights 2025 survey, more than 97% of global data leaders say they struggle to clearly demonstrate generative AI’s business value. And, while 87% plan to accelerate investment in AI, 67% admit they’ve transitioned fewer than half of their AI pilots into full-scale deployment.

One of the biggest stumbling blocks is securing leadership buy-in. Over a third (35%) say securing support and demonstrating value is a key challenge impeding the rollout of AI. What this means is that many remain stuck in a holding pattern, hesitant to commit to wider rollouts without measurable proof points.

This hesitancy stands in sharp contrast to the technology’s potential. McKinsey estimates that AI and analytics could deliver up to $1 trillion in additional annual value to global banking, while generative AI alone could contribute up to $340 billion to operating profit. It’s an opportunity too significant to ignore – but one that must be approached in a way that safeguards compliance, builds trust, and generates proven returns.

The path forward

Despite significant headwinds, there are organisations across Europe, and the rest of the world, that are progressing their AI rollout, exploring how they can reap the rewards of AI agents. Those that are moving the needle aren’t doing so by diving headfirst into complex, long-tail deployments. Instead, they’re adopting a measured approach: starting small, building confidence, proving value and scaling up only once the technology proves its efficacy.

The most successful AI rollouts don’t happen overnight. They begin with small, high-impact moves that build trust and deliver results. Here are three steps to get started.

1. Use AI to clean data before scaling

Even with compliance sign-off, AI systems are only as strong as the data they’re built on. Poor data quality will undermine accuracy, efficiency, and trust. In fact, 43% of data leaders say data issues are their biggest barrier to scaling generative AI.

Encouragingly, AI itself can help fix these data problems. In financial services, for example, some firms are using AI tools to cleanse accounts receivable data, removing duplicates, correcting outdated entries, and resolving mismatched records. Once the data is aligned and reliable, companies can automate follow-ups, improve cash flow, and operate with greater confidence in their AI-driven insights. This is also a top investment priority. 86% of data leaders plan to increase data management spending, with nearly half citing making data fit for AI as their primary motivator.

2. Begin with focused executor agents

Deploying narrow purpose “executor” agents is one of the fastest ways to generate measurable wins. These agents are designed to handle very specific, well-defined tasks, such as compiling meeting summaries, processing standard transactions, or categorising incoming customer queries.

Because executor agents are straightforward to monitor, they produce outputs that are clearly trackable and easier to validate for accuracy. This not only reduces operational risk, but also provides early proof points for stakeholders, helping secure buy-in for wider adoption.

Once success has been demonstrated with single-task agents, organisations can introduce more complex agentic structures, such as planners and orchestrators, to handle multi-step workflows.

3. Streamline compliance reporting through automation

Compliance is a heavily resource intensive area in financial services. Regulatory reporting often requires gathering and reconciling data from multiple sources, a process that can consume hundreds of hours and rely on a small pool of trained specialists. AI excels here, providing an excellent starting point to test and scale the tech.

Once underlying data is cleaned and structured, AI can take over some of the heavy lifting. For example, generating BCBS 239-compliant reports can be partially automated using metadata mapping combined with agentic AI models. These systems can produce accurate first drafts that are then reviewed by compliance officers, reducing turnaround times while maintaining quality control.

The potential here is significant. McKinsey highlights one global bank that achieved productivity gains of 200% to 2,000% in know-your-customer (KYC) processes by adopting an “AI agent factory” approach. They retained human oversight but automated the most time-consuming steps.

Lessons from the data journey of a multinational bank

One Dutch multinational bank recognised the importance of building the data foundations for AI success. It realised the importance of data management, making that a priority. It invested in the right organisational processes to enable delivery at scale, making deliberate choices to empower teams. And it gave teams clear direction and strong cross-functional collaboration to succeed. This combination of trusted data, empowered teams, and clear strategic direction is what enables AI to deliver business value — not just technology outcomes.

Building momentum without losing control

With 76% of financial services firms planning to roll out agentic AI solutions in the next 12 months, momentum is building. However, it’s clear that the most successful organisations are not rushing into full-scale transformation. They're deploying AI strategically, focusing on small, well-contained use cases that deliver measurable value and improve operational efficiency. They’re also embedding governance into every stage, ensuring compliance teams are involved early and often.

By adopting this incremental approach, firms can accelerate AI adoption without sacrificing trust or regulatory alignment, turning “starting small” from a perceived limitation into a deliberate, proven growth strategy. In AI adoption, speed matters, but safety and scalability matter more. Financial service institutions that start small, prove value, and scale with confidence will be the ones best placed to unlock AI’s trillion-dollar potential.