Investment industry leaders see AI as an almost inevitable technology, even as they face pressure to cut costs, improve performance, and meet rising expectations from clients, boards, partners, and regulators. Although every firm wants leverage, starting points differ by segment:

Hedge funds have two advantages for early exploration: high volumes of data and a fast pace; and strong frameworks for testing and managing risk and errors.
Private markets teams operate with big positions that demand depth and accuracy.
Institutional managers carry the heaviest weight because they are responsible to institutional investors with strict investment mandates, and they have a lot less discretion about how they do what they do.
On the sell side, many firms are committing significant resources and spend, but their organizational scale slows them down.

As we look across the industry, we are seeing how these different conditions shape where front-office AI gets traction and where teams need more confidence and clarity. And despite the differences, there’s one common requirement: the strength of their data foundations. Strong foundations determine how far each segment advances toward agentic workflows.

Hedge fund experimentation

Hedge funds are likely to feel the impact of AI earliest because their whole investment model depends on speed. Their teams already work with large volumes of data, from traditional market data sources and tick data to alternative data sources they can find, like earnings reports, maritime data, or foot traffic. The value comes from moving fast, because the faster they can process data and extract useful insights, the bigger the competitive advantage. AI pushes that pace even further.

Their momentum also benefits from a culture that expects experimentation and tolerates risk. AI is just the next leading edge for doing that. Front offices use that mindset to explore new workflows without slowing down investment activity. An AIMA survey found that more than eight in 10 already give their teams access to generative AI tools.ⁱ

For example, quant and global macro portfolio managers and researchers are familiar with building various algorithms, supported by a risk structure, so they can try those algorithms out, do what works, double down on it, and throw out what doesn’t. They can apply the same risk tolerance framework to AI, where they set parameters, evaluate output, and throw out experiments that aren’t helping.

However, long-short strategies may tilt toward more conservative approaches. Especially when they have highly concentrated portfolios, the tolerance for error is much lower, even as other parts of the hedge fund landscape pursue more automated approaches. They can almost seem more like private market firms where a small number of large positions raise the risk for outsized consequences from hallucinations or AI errors.

The unstructured data frontier

Private markets teams use AI differently because their front-office reality centers on documents rather than structured data. Large investment decisions and monitoring activities depend on covenants, financial statements, board materials, and deal documents that arrive in uneven formats.

This challenge has been a problem for years and decades. So far, every attempt to standardize inputs has struggled. Many teams tried pushing portfolio companies or their general partners to structure their data in a certain template, but that strategy never quite landed. New AI models might finally shift the data-collection model. Instead of waiting years to build clean historical sets or enforcing templates that vary across sponsors and management teams, AI moves faster. We’ve already seen some AI models accelerate the process for both new deals and years’ worth of historical data.

Teams now get value much more quickly. When AI handles more extraction and interpretation, the advantage no longer sits only with the groups that spent years building proprietary datasets. For example, ratings agencies are using AI to support data extraction.ⁱⁱ It puts better data in the hands of a lot more people, because the bottleneck around historical data collection starts to disappear.

That said, even with these potential efficiency gains, private markets still require strong controls. AI-driven extraction and analysis depend on the right guardrails and structure, and a human in the loop to get the right data out and instill confidence that the data is accurate. The stakes are high: If you’re buying a $50 million private equity investment, even small mistakes can create enormous problems.

The challenge of scale and scrutiny

Institutional asset managers approach AI with the most caution because their responsibility runs through the entire investment chain. They serve pension plans, endowments, sovereign funds, and other asset owners who expect consistency, oversight, and clear explanations.

When investment teams carry big institutional money with layers of review, they have the lowest risk tolerance. When those teams evaluate AI for front-office use, they face regulatory pressure. They can’t just say, “Oh, the AI did it, not my fault.” They need to rely heavily on data lineage and provenance to show where every figure comes from and how it fits within a portfolio’s constraints.

Workload also expands as portfolios do. Multi-asset teams cover more markets and instruments, so they look for ways to keep their process broad and stable. Having a strategy built to track or modestly outperform an index leaves little space for untested ideas, which shapes how they use AI.

But AI still does play a growing role in this environment. Mercer finds that 91% of managers are currently (54%) or planning to (37%) use AI within their investment strategy or asset class research.ⁱⁱⁱMany firms use it to synthesize research, summarize market commentary, and check alignment between holdings and client policies.

These controlled tasks have review layers that preserve auditability. The front office might also explore early idea-screening workflows, where the model surfaces signals, but human teams validate the reasoning. This discipline creates stability, even as AI becomes part of their daily work.

Big budgets, slow impact

Sell side institutions sit in a different place on the AI curve from the buy side. Many large banks, for example, commit enormous resources and spend to build corporate, centralized AI teams. But the challenge is scale. These are big organizations, and it takes time to get those types of things moving across thousands of users, desks, and systems.

One common complaint is that early progress feels narrow. Teams are solving small problems today rather than materially changing the way they’re operating. Much of the visible activity happens within internal use cases, like helping write employee reviews, drafting emails, or supporting knowledge management. In 2025, EY found that more than three-quarters of banks have actively launched or soft-launched GenAI applications.^iv

These uses create efficiency, yet they sit far from the more sophisticated front-office workflows often discussed publicly. The potential is there, but the road to front-office transformation is a lot earlier in the lifecycle. As with the rest of the industry, clarity, auditability, and workflow maturity will shape how quickly the sell side can move.

Enabling the next stage

Adoption still feels early because progress in front-office AI depends on the strength of the underlying data. Agents are still only as effective as the data they’re using and what they’re learning from, which means firms need clean access as well as lineage that shows how each output comes together. Provenance also becomes part of the workflow, because teams rely on clarity when the stakes rise.

Even so, momentum is building to evolve from generative AI to more agentic approaches. Teams that treat data foundations as a strategic capability move faster, explore more ambitious workflows, and avoid getting left behind as AI evolves.