Financial services firms are quickly deploying agentic AI, a type of artificial intelligence that can autonomously plan, reason, and act across complex workflows.
While 2025 was a year of pilots and exploration, 2026 is shaping up to be the year firms shift into production and seek tangible results.
Broadridge’s recently released 2026 Digital Transformation & Next-Gen Technology Study found that 26% of firms are already using agentic AI—signaling a shift toward practical applications, particularly in operations automation and customer experience.
But financial firms implementing agentic AI face significant barriers, including change management, integration with legacy systems, and governance and risk issues. Market research company Gartner predicts more than 40% of agentic AI projects will be canceled by the end of next year, often because the technology is being misapplied.
To overcome these obstacles, firms need a practical framework for where and how to deploy agentic AI. Success depends not just on the technology itself, but on making smart choices about use cases, implementation, governance, and other considerations. In our view, firms are most likely to generate real business value when they focus on five priorities:
1. Start with lower-risk, high-value use cases
Project selection is often a key determinant of success, and many firms are deliberately starting with lower-risk use cases.
But that doesn’t mean AI isn’t being used for important work.
One example of a low-risk but high-value use case is an agentic AI prototype Broadridge completed for a top Canadian bank. The system was designed to help automate the process of onboarding individual investors who move from one bank to another. This is paperwork-intensive and time-consuming. Depending on the number of accounts and the complexity of the investor’s holdings, onboarding can take weeks to complete.
The longer it takes to onboard an investor, the longer it takes for the bank to start realizing the investor’s trading revenue. Not only that, but some investors find the onboarding so onerous that they don’t complete the process. And when they drop out, the receiving bank loses that business.
Broadridge worked with the bank to develop an agentic AI solution to dramatically accelerate its client onboarding. The agentic solution checks transfer forms and account statements looking for missing fields, mismatched instructions, incorrect account details, and non-transferable assets before submission—preventing downstream rework and long delays. Based on this detailed document analysis, it summarizes the status for final approval by the operations associate.
The project has shown the potential for tremendous productivity improvements and revenue acceleration.
2. Treat agentic AI as a business transformation effort
Companies can spend too much time focusing on the technology itself and not enough on how agentic AI will reshape the way work gets done.
To capture the full value out of agentic AI, firms need to rethink workflows, operating models, and service design. As with previous generations of technology, simply automating current processes is a suboptimal approach.
“Many firms initially look at agentic AI as a way to automate individual service requests, such as onboarding a new user or updating entitlements,” said Roger Burkhardt, Head of AI at Broadridge. “But the bigger opportunity is to step back and redesign the process entirely—shifting from manual support models to intelligent self-service experiences that are faster for clients, more scalable for operations teams, and ultimately deliver a better user experience.”
Successful firms see agentic AI not just as a technology project but as an opportunity for business process redesign and optimization.
Roger Burkhardt, Head of AI at Broadridge
3. Build around legacy systems instead of waiting to modernize them
Firms often know how to build agentic AI systems quickly. Realizing value from them, however, requires integrating those capabilities with existing systems, workflows, and data stores—and that often exposes modernization gaps.
There is always an impedance mismatch between the rapid pace of AI development and the slower pace at which core systems can evolve. In many cases, that is the main constraint on capturing agentic AI’s full value. No firm wants to wait for some indeterminate, multiyear modernization program to get value out of AI.
One solution is to use application programming interfaces (APIs) to encapsulate existing systems—often in combination with Model Context Protocol (MCP), an emerging standard that allows AI models to securely connect to external tools, data sources, and software systems through a consistent interface. Together, APIs and MCP make it easier to connect AI to legacy environments, automate workflows across silos, reduce custom integration work, and move faster without major core-system rewrites.
“One of Broadridge’s roles is helping firms connect AI capabilities into the systems and workflows they already rely on,” said Burkhardt. “The goal is to accelerate adoption and results without requiring firms to first undertake large-scale system replacements.”
4. Put dedicated leadership, resources, and governance in place
Another key factor for success with agentic AI is educating business leaders at all levels about the technology’s capabilities and giving them hands-on experience. If business, product, and functional leaders—such as those in client services or human resources—don’t understand what’s possible or haven’t experienced it themselves, you’re not going to get very far.
Firms also need to provide employees with safe, controlled access to these capabilities so they can learn by doing. That does not mean giving everyone every tool. It means putting the right capabilities in the hands of the employees most likely to use them well—and reallocating access when adoption stalls. Interest and aptitude vary widely across the workforce.
“Successful AI adoption starts with people, not just technology,” said Burkhardt. “The firms making the most progress are identifying employees who can become early adopters and champions, giving them hands-on experience with the technology, and dedicating time for them to help rethink how the business operates. These leading firms are proactively defining new roles that make concrete the individual's opportunity for career growth.”
But here firms need to avoid another pitfall. One common mistake is asking staff to handle agentic AI projects as “side-of-desk” work—expecting employees to figure out how to transform the business while also managing their day jobs.
Without dedicated resources, firms are unlikely to achieve strong results.
You could team these dedicated people with an outside consultant—which can reduce the number of internal staff pulled away from day-to-day responsibilities—but external agentic AI experts alone are unlikely to know your environment well enough to produce a strong outcome.
Firms also need people who understand governance, especially in banking and financial services, where regulatory expectations and risk management are paramount. Agentic AI is inherently non-deterministic: the same prompt or task can produce different results. Some outputs may be acceptable; others may be inaccurate or problematic. Strong governance is essential to ensure appropriate oversight, controls, and risk management.
5. Measure success with practical business outcomes
Agentic AI projects can be expensive.
Costs can add up quickly. Model usage, orchestration, integration, testing, governance, and change management all carry real costs. Token usage alone can range from roughly $500 to $2,000 or more per developer per month. Firms need to manage that spend carefully. That’s part of the reason Broadridge encourages the use of a central team to provide these capabilities and to ensure they’re being used efficiently.
That said, firms also need to keep in mind that the AI costs are falling rapidly. Gartner says that by 2030 running predictions on a 1-trillion-parameter AI model will cost 90% less than it does today. A model that is expensive to build or run today may cost much less in 12 months as token costs decline. Another way to ride the cost curve down for a specific use case is to replace high-end models with mini models, which cost less and may be optimized for specific workflows.
Optimizing cost per interaction will become increasingly important.
You also need to track the return on your agentic AI investment. That starts with having a business goal—such as improving productivity by 40%—and having the metrics to measure your progress.
The ROI metrics for agentic AI are familiar: productivity, customer engagement, Net Promoter Scores, revenue growth, error reduction, and service-level improvements.
For many firms, however, the problem is that they don't know their current business metrics.
Successful firms are doing the hard work of establishing baseline metrics and being disciplined about where agentic AI can create value—whether by improving service levels, increasing throughput, reducing errors, or delivering the same service at lower cost.
Execution will separate the winners
What will separate leaders from laggards is no longer access to AI—it’s execution.
The firms that succeed will be those that move quickly from pilots to production, focus on high-value use cases, and invest in the leadership, governance, and operating models needed to scale.
Those that wait for perfect models, complete modernization, or clearer industry playbooks risk falling behind. In this environment, progress matters more than perfection. The advantage will go to firms that execute thoughtfully, avoid predictable pitfalls, and scale with discipline.