Why AI in Financial Technology Is Harder Than You Think

At Quod, we didn’t set out to AI everything overnight. We targeted real friction points: bottlenecks, trapped expertise, inefficient workflows. Our method is fast, pragmatic, and rooted in measurable outcomes—not grand visions.

Our first attempts at knowledge management exposed the reality. Training models by labeling Jira tickets quickly failed—the complexity of interconnected knowledge overwhelmed simple approaches. We pivoted, mapping task relationships through graph databases and building specialized LLMs wrapped in multi-agent systems. Progress was real, but early constraints like limited context windows and basic RAG methods held us back.

Today, with better tooling, we’re aiming for an 80%+ accuracy rate. When achieved, it will transform how expertise is shared across the company—moving knowledge out of silos and into everyone’s hands.

In test automation, AI didn’t just generate new scripts; it refactored legacy ones, cleaning technical debt and scaling quality sustainably. In our codebase modernization work, we carefully introduced AI to assist with refactoring critical systems, like cache access layers—always with human developers guiding the process, ensuring robustness and scalability.

Across every project, the lesson was clear: AI doesn’t replace expertise—it accelerates it, but only when the humans around it shape the process intentionally.

Today, we are running over 25 AI initiatives in parallel, prioritizing them through a clear framework: impact, team ownership, difficulty, and whether external or internal LLMs can be used.

High-impact, low-difficulty projects—especially those leveraging external technologies—are prioritized for fast wins. If a project doesn’t deliver meaningful benefits quickly, we don’t mourn it. We shelve it, learn from it, and wait for the technology to evolve. When the time is right, we pick it up again.

This “benefit or fail fast” mindset allows us to innovate aggressively without overextending. It’s not about betting everything on one breakthrough—like trading, it’s about running dozens of disciplined experiments in parallel, knowing a few will unlock disproportionate returns.

The biggest trap in AI is waiting for perfect outcomes. We don’t. If AI can move something forward—even modestly—we double down. If not, we move on. Every project aims to chip away at inefficiencies: faster onboarding, sharper test coverage, smarter support workflows.

In three years, we expect these incremental gains to deliver at least a 50% overall productivity improvement—though we believe it could approach 80%. Double output, same cost base!

One of our biggest advantages has been starting with a system that was already modular, auditable, and compliant. We didn’t have to retrofit brittle architectures to make space for AI. Our infrastructure, data pipelines, and compliance frameworks were designed for traceability from the outset, making it easier to safely embed AI deeper into operations.

Without strong foundations, no model—no matter how powerful—can help. But with the right system in place, AI compounds value at every layer. For those with less modular systems there is perhaps only so much you can do to retrofit. How long before you get left behind if you can’t use this kind of technology?

The real transformation with AI isn’t a flash of magic. It’s a thousand disciplined steps: cleaner data, smarter workflows, faster feedback loops. It’s a willingness to rethink processes, to insert AI where it adds value today—and the discipline to abandon efforts when it doesn’t.

At Quod Financial, we’re not chasing AI hype. We’re building a system where AI amplifies human intelligence, unlocks trapped knowledge, and quietly compounds value day after day.

It’s not easy. But it’s absolutely worth it.