ROI in the Age of AI: Powered by People, Enhanced by Agents+

In April, a startup launched an AI-powered application to deliver insights and intelligence to retail investors, small hedge funds, and family offices, for quantitative strategies. It calls its architecture LLMaaS (large language model-as-a-service). Should you have heard of LLMaaS? Not really. If you’ve ever tried to take an image of something from a moving vehicle, you get nothing but a blur. If you’re feeling like you can’t keep up with the advancements in AI, don’t feel bad. Just within the last two weeks, Anthropic released its most capable model yet with the rollout of Claude Opus 4; Open AI launched Codex, its cloud-based software engineering agent; and Google released Jules, an asynchronous coding agent poised to reshape how software development happens.

The rate of advancement in AI is unprecedented, outpacing the development of many past innovations. With the rapid pace of innovation driving real value, AI is at an inflection point.

Building on its momentum, the use of AI is shifting from experimental pilots and proofs of concept to practical applications that drive real-world outcomes. For investment firms, this evolution presents a compelling opportunity to streamline operations, enhance decision-making, and achieve greater efficiency—moving beyond the initial wave of hype. So, let’s stop the vehicle to capture a clear image of where AI is today and understand the steps buy-side firms can take to develop an implementation strategy to deploy the technology across their business functions.

RELATED READING: The Agents Are Coming to Finance

Embracing the change

If you work in the buy-side or the sell-side of capital markets, AI may sometimes seem like a bubble because everyone is talking about it. In the early excitement stage around generative AI’s large language models (LLM) circa 2023, people were both overly excited and nervous about things they didn’t feel quite ready for.

Sometimes in tech we over use the term “transformative” for solutions that fall a little short of earning that adjective. AI may prove to be the topper on the digital transformation cake that isn’t even fully baked yet. Many companies claim to use AI, but few have truly operationalized it in a way that delivers measurable ROI. We must distinguish between innovation theater and production-grade AI.

But the thing is, AI will change everything. The technology is transformative, and people shouldn't be afraid to participate in the progress. To be successful, firms must have a refined AI strategy backed by a strong message from leadership to really see the impact.

My colleague, Matt Katz put it perfectly in a recent conversation with Hedgeweek on Age of AI: The Latest on Artificial Intelligence in Hedge Funds Operations: “Efficiency is primary, but the way it manifests is through what I call ‘data centaurs’—humans augmented with technological capabilities. In this model, AI forms the supporting structure while humans maintain control, context, and accountability. This combination creates superpowers without replacing human judgment.”

Building an AI-ready organization

While there has been an urgent need for comprehensive investment lifecycle data technology for years now, the need to integrate AI agents into operations is just as urgent. It takes more than good models to run AI at scale—you need the full complementary technology stack: clean data, a well-defined information security posture, tools that allow for observability, and a knowledgeable team ready to open their arms to new AI innovation that enters the market.

The race to innovate calls for thoughtful planning, technical talent, ethical considerations, and investment. Organizations must also balance caution with vision. That starts with identifying the high-revenue potential or efficiency-gaining use cases.

The next step will be assessing whether the organization has the internal capacity—data scientists and AL- and ML-specialized engineers, technology stack, robust information security posture, culture of innovation—and refined software development lifecycle mechanisms to take experiments to production quickly. As firms work to develop a forward-thinking AI strategy, implementation should be viewed as a long-term investment that requires alignment across all levels.

Leadership that clearly communicates the importance of AI, sets the tone for innovation, and prioritizes adoption is more apt to create an environment where teams understand the strategy and have the right skills to keep up with AI’s fast pace of change. Often, this involves evaluating whether to build or buy a solution. Selecting relevant business use cases can further drive tangible operational efficiencies, revenue growth, or customer experience.

Constants and variables in the AI equation

Each AI use case has specific patterns. The first step is to convert your business problem into a workflow pattern and evaluate what you need for the unique agentic pattern. Asking: What technology should I use; Do we build or buy? And what will it take to build agents at scale?

Today, the industry focus is shifting from GenAI to AI agents. Foundational models or LLMs like those from OpenAI, AWS Bedrock, or Anthropic are the underlying technology. Firms no longer need to "train models", as these foundational models are trained on large datasets. The vast majority of use cases can be addressed by using these models "as-is" to speed up time to market and save money.

An agentic framework can enable LLMs to take action, and different agents can be built using the same underlying model and framework but perform different tasks or patterns such as knowledge retrieval versus enriching queries.

Still, AI is not a replacement for having a robust platform that supports middle- and back-office functions. In some cases, AI enhances the user experience or acts as user itself. Think of it this way: you often still need a physical vehicle to get from point A to point B. What AI does is make the experience more efficient in your journey.

When it comes to choosing use cases, hedge funds and other buy-side firms all have the same basic, daily demands. They need to accurately and efficiently execute trades, reconcile the books, optimize portfolios, and report to investors, among other things. Just a few decades ago, investment managers did most of these tasks using manual, paper processes and spreadsheets on their 640 kilobytes RAM desktop computers.

The fundamentals of investing have remained (mostly) constant, while the tools and technology have changed dramatically.

For an asset or hedge fund manager, the business of driving risk-adjusted returns in capital markets—a feat that includes sound treasury management; airtight compliance, proactive risk management, and often, daring allocation strategies—is the domain that will stay constant. Can AI help an analyst or back-office line employee execute these fundamentals? Absolutely. But can AI do these functions on their own?

Rise of domain-specific AI agents

The industry is in the midst of a fundamental shift in which AI is maturing from being a "helper" that answers queries to being a "doer" that performs tasks. For example, Arcesium built and launched an AI copilot last year in our Aquata Data Platform. Our copilot is answering queries, generating SQLs from conversational requests for data, and helping establish data quality rules. Now, we’re taking it a step further testing domain-specific AI agents trained to do specific tasks for funds. We are building agents in Aquata to do things for hedge funds, such as unstructured data management, governance, and creating insights from data.

A key challenge is ensuring the AI applications provide measurable value. Simply building something is not enough. Many financial institutions are chasing value with their AI initiatives. Sometime soon we fully expect AI agents to simplify the work of middle- and back-office professionals’ in functions like reconciliation that auto-assign root causes to exceptions and actioning the anomaly.

AI is helping businesses cut operational costs through automation, predictive maintenance, and decision augmentation. Firms can get even more value out of their investment by identifying the right transformative applications for AI and by focusing on the KPIs that matter most. Margin improvement, customer retention, or delivery speed and innovation are just some of the use cases that come to mind.

AI amplifies calls for data transformation

Garbage in, garbage out goes for AI agents as much as any data management operation. Before they even have an AI agent to execute their tasks, PMs, analysts, accountants, and risk management professionals all rely on clean, consistent, actionable investment data for positions, cash, PnL attribution, margin and financing attribution, and more.

AI tools may look very different in five years than they do now, but in today’s models and frameworks, technologists are laying the foundation building blocks of the future agents. Companies that ensure they have good data and the right tools are setting themselves up for continued success down the line.

Human-AI collaboration will drive your AI success

AI is already transforming the world of finance—but human and AI collaboration is what will lead to the best outcomes.

The World Economic Forum proposed that “agentic AI advice systems will also integrate collaboration with human experts as part of their workflows, strengthening the credibility of the human-AI compound systems.”

Operational AI must augment—not replace—humans. The best ROI will come from human-AI collaboration.   

AI isn’t a hammer for every nail—without a solid operational foundation, all it does is amplify noise. The smartest deployment of AI starts with building a solid foundation of technology and teams, as well as choosing the right use cases, not forcing it into the wrong ones.

AI is no longer hype, it’s becoming transformative. Firms must embrace the tools and teach their teams to do the same or risk being left behind.