Fidelity International began its operations in India in 2001, initially establishing its presence in Gurugram, Haryana. The India Global Capability Centre (GCC) has evolved into one of the company's largest global capability hubs.
Rahul Jain heads the AI Platform function at Fidelity International and his role involves building core foundational Artificial Intelligence (AI) capabilities, used by teams across different departments, and functions. Rahuls’s team is not just building capabilities, but also helping teams adopt and use them to build their own solutions. Looking at the enablement side of AI.
After having served the company for nearly 23 years across roles spanning technology, delivery, and architecture, since the last nine years, Rahul started working as a data scientist, on classical machine learning, deep learning, and GenAI.
Positioning Fidelity International for the GenAI Phenomenon
AI became much more mainstream after ChatGPT happened, but in the pre-GenAI era the focus was primarily on machine learning.
Typically, a problem was identified and a machine learning solution was built around it. “My first implementation was focused on an optimization problem, which is a very specific category within machine learning. From there, I moved into more classical machine learning applications. At Fidelity, we established our AI Center of Excellence in 2019, well before the rise of GenAI. At that time, the focus was on how to use machine learning and deep learning to solve specific business problems,” said Jain.
A Responsible AI framework was established in 2022, which outlined Fidelity’s governance approach toward AI and machine learning. Naturally, the framework was revisited and evolved once GenAI became mainstream.
“By the time ChatGPT and GenAI gained popularity, we had already spent four to five years building AI capabilities. That positioned us well to leverage the opportunity when GenAI emerged,” said Jain.
The First GenAI Milestone
Over the last three years, the focus has shifted toward adopting GenAI while continuing to work on core machine learning problems. “We launched our own internal GenAI platform in early 2024, which saw strong adoption across the organization, with nearly 7,000–8,000 employees using it. That became an important milestone for us and helped shape many of the subsequent AI solutions we built.”
Moving Beyond AI Proofs of Concept
Fidelity International is engineering AI platforms that can move from isolated pilots to enterprise-wide deployment across functions while still maintaining governance, security, and cost efficiency.
The company looks at the approach in three parts. The first is the platform itself. Teams using AI capabilities need not worry about how models are accessed, how data processing happens, how retention policies work, how costs are tracked, or how request-per-minute and token-per-minute limits are managed. These concerns are centrally managed through the platform.
Most GenAI applications also follow familiar architectural patterns such as Retrieval-Augmented Generation (RAG) and Model Context Protocols (MCPs). “We have centralized those capabilities within the platform, making it easier for teams to consume them without reinventing the wheel repeatedly.”
The second part is governance. Fidelity has a cross-functional oversight group that ensures all AI solutions undergo reviews from risk, legal, compliance, security, and architecture perspectives. This gives us a comprehensive view of the risks associated with every solution.
The third part is behavioral. A thorough upskilling programme is undertaken in the company — not just of technology teams, but also control functions and business teams.
The company began with smaller implementations, which helped build organizational confidence that AI solutions could be deployed at scale. “That confidence has enabled us progress toward more sophisticated and complex AI workflows,” said Jain.
Fidelity’s AI Ecosystem and Vendor Strategy
Fidelity International has built an AI ecosystem composed of models, vendors, infrastructure, compute, and upskilling programs. From a vendor standpoint, there are two categories of providers: Established enterprise providers as well as niche companies with highly specialized implementations. That balance allows to combine enterprise-grade stability with cutting-edge innovation.
The first category includes vendors offering copilots and productivity-oriented AI solutions — tools such as note takers and similar applications that help scale AI usage for employees. The second category consists of vendors who assist to build advanced, pro-code AI solutions.
The company operates on a multi-provider platform. “We integrate with models from multiple leading providers, which gives us flexibility. Teams can choose Model A from one provider, Model B from another provider, and so on, depending on their needs.”
On the learning side, the efforts go beyond just training AI engineers or teaching prompt engineering. AI ways of working have been designed across the organization to create a baseline understanding of how employees should work with AI.
In terms of infrastructure, the AI platform is cloud-native by design and runs fully on the cloud. So far, the journey has largely focused on closed-source models, without any need to manage large-scale GPU infrastructure internally. There are situations where teams experiment with hosting specialized models themselves, and in those cases the teams rely on accelerated compute resources from cloud providers rather than maintaining racks of GPUs in the company’s data centers.
The AI Marketplace
Fidelity ensures that different business teams, from operations to investment research, use AI capabilities in a standardized and reusable way across multiple countries and functions. That is a very real challenge.
“Building the platform itself is only one part of the equation. The real objective is ensuring those capabilities are consumed within value-generating business applications. We therefore place equal emphasis on enablement. As part of the platform, Fidelity has an internal AI marketplace. Think of it as a centralized internal portal showcasing all platform capabilities. Employees can access it and explore the AI tools and capabilities available to them,” said Jain.
The marketplace also hosts accelerators, essentially demonstration solutions showing how platform capabilities can be applied to solve real business problems. For example, there could be accelerators for document analysis or quality analysis. This has helped democratize AI capabilities across the organization.
The marketplace has also been indigenously empowered with deep analytics capabilities to understand which accelerators are being accessed the most. That feedback helps refine the platform further.
“One important learning has been that business teams are highly interested in understanding what other teams are building. That cross-pollination of ideas has been extremely valuable. Teams also approach us directly with specific problem statements. In those cases, we provide support around design and implementation. Overall, our enablement strategy combines the marketplace, webinars, one-on-one support sessions, group workshops, and advanced AI training programs.”
Token Management and AI Economics
Today, companies are increasingly focusing on token-to-business-value economics. Since AI usage is fundamentally token-driven, companies must balance token management with ROI optimization.
Within Fidelity’s AI platform, all interactions happen through an AI gateway. Anyone accessing a model does so through this centralized gateway. This enables the company to apply very fine-grained controls at the user, team, and application levels.
“For example, if a team wants to build an application, initially a fixed credit budget might be allocated to them — say $100. That creates an upper limit on token consumption while teams develop their solutions. This is the first intervention: very granular cost and request controls. The second intervention, which is currently under experimentation, involves intelligent model routing. Depending on the prompt being fired, the platform can automatically determine the most appropriate model. Not every prompt requires a frontier model. Simpler prompts can be routed to more cost-efficient models. That capability can significantly improve economics.”
The third aspect depends heavily on the use case itself. Some research-oriented applications naturally consume large volumes of tokens and incur high costs. However, if the business value generated is substantial, that expenditure may still be justified. That evaluation cannot always be automated. It often requires contextual business assessment.
Finally, strong observability and monitoring are critical. Extensive observability features are built into the platform to quickly identify anomalies in token consumption or usage behavior.
Future Roadmap
Fidelity’s roadmap can broadly be divided into a few categories.
First is skilling and reskilling employees. As AI becomes more pervasive, the company is investing heavily in helping employees both build with AI and effectively use AI in their day-to-day work.
Second is the application of AI itself. The focus is on three broad outcome areas:
-
Better employee experience
-
Better customer experience
-
Better investment performance
Those remain the primary functional outcomes the company is targeting.
“Another major focus area is continuously enhancing our AI platform to make it safer, easier, and more scalable for teams to build AI solutions. That includes work around guardrails, observability, monitoring, onboarding new providers, and integrating new models,” said Jain.
