By Rishi Chohan, CEO, GFT USA
AI is at the top of every financial institution’s agenda, whether they’re a global bank, a mid-market asset manager, a P&C insurer, or a capital markets firm. What tends to get less attention, across all of them, is the infrastructure problem that determines whether AI initiatives succeed.
Most of the conversation focuses on which technologies to deploy and how to use them. But what these organizations should be discussing is whether the underlying data infrastructure can actually support AI’s needs. These considerations tend to come later, often after an initiative has already stalled.
The institutions furthest along started with the infrastructure, not the AI. Financial institutions cannot bolt AI onto systems that were not built for it and expect consistent results. What that foundational work looks like varies by sector because the infrastructure problem does not show up the same way across different types of firms. This piece looks at how it shows up in asset management, insurance, and capital markets, and what it takes to address it in each.
Asset managers are dealing with data they technically have but can’t use
Limited partners (LPs) are asking for more transparency and faster reporting, while companies are staying private longer, compressing the window for return. As a result, deal teams are expected to do more with the same or fewer resources.
The problem is that most firms are sitting on enormous amounts of information that they cannot effectively access. Deal memos, due diligence files, board decks and LP correspondence all live across systems that were never designed to communicate with each other. This means deal teams spend significant portions of their day tracking down information they know exists somewhere in the organization. But this is not a problem that AI can fix; it is a data infrastructure problem that has to be resolved before AI can do anything reliable with it.
The challenge also looks different depending on where a firm sits in the market. Larger managers with dedicated technology teams have the resources to go after this in a structured way. Middle-market firms often run a patchwork of tools, including a CRM that does not talk to the data room, and reporting is still done in Excel. They are earlier in their digital transformation, but the impact of getting there can be more immediate, because the current baseline is so manual.
What both types share is that the foundational work, centralizing data and structuring it so AI can actually use it, has to happen before the AI rollout, not alongside it, and not after the first initiative fails to deliver.
Insurers have the same legacy problem, with a customization layer that makes it harder.
Most P&C and life insurers have increased their investment in application modernization. However, the majority are still running legacy systems built over decades, each customized to fit their organization’s specific products and processes.
That customization is what makes insurance modernization difficult in practice. Every insurer has built its own version of the same core functions, underwriting, claims, policy administration, on an infrastructure that was never designed to support what AI needs. Addressing that requires understanding how the business actually works, not just what is in the technology stack.
The insurers making progress are starting with the underlying systems, not the AI. Deploying AI on top of an unmigrated legacy system does not get firms closer to the goal. Instead, it creates more outputs from a system that was already the problem.
What has shifted in recent years is that AI now has a role in the modernization process itself, not only as the end state. Legacy insurance systems are dense with business logic accumulated over decades of product changes and regulatory updates, most of which have never been formally documented. Extracting and understanding those rules has historically been one of the slowest parts of any modernization effort. AI tools can now analyze legacy codebases and surface that logic in ways that once required years of manual work. Modernization can move faster and with more confidence than it could even five years ago. This makes AI a part of how insurers are getting there, as well as the destination.
Insurers also have to navigate highly localized regulations that differ from one jurisdiction to the next. This means modernization must account for compliance at every stage, and generic approaches tend to break down in this context.
In capital markets, the same problem runs on a tighter clock
In capital markets, the cost of bad infrastructure shows up in real time. When a trading system cannot keep up, the problem is not only a slow report or a delayed workflow; it is a missed trade.
This is also what limits AI in capital markets environments where that foundational work has not been done. AI is only as fast as the data it can access, so when the underlying systems are fragmented, it cannot operate at the speed the business requires.
Added to this is the complexity that firms cannot phase in changes gradually or run old and new systems in parallel for long. The modernization work must occur while trading continues, making it both technically demanding and operationally unforgiving.
The firms that have done this work are not just better positioned for AI today. When stablecoin settlement and tokenization move into institutional practice, they will be able to move, while those still on legacy infrastructure will be rebuilding under pressure.
The specifics differ across financial services, from deal data fragmentation in asset management to bespoke legacy systems in insurance to latency constraints in capital markets. But the dynamic is consistent: the data underneath has to be right before AI on top can work.
What has also become clear is that AI has a role in the modernization process itself, not only as the end state. The same capabilities firms are working toward can help accelerate the foundational work, from analyzing legacy systems to surfacing business logic that has been buried for decades. The path and the destination are more connected than most modernization roadmaps acknowledge.
The firms making progress have treated the underlying data foundation as the first problem to solve, not something to address after an AI program falls short. That approach is slower and less visible than an AI rollout announcement. But it is the one that is producing results for firms today.
