Artificial intelligence (AI) has moved beyond mere promise and is shaping the asset management industry, both in quantitative strategies and in research and decision support for discretionary managers.

Machine learning tools and generative models are already capturing data, monitoring risks, automating analyses, and increasing the productivity of teams that, until recently, worked manually.

To understand what already works, what's still hype , and how competition is likely to intensify, NeoFeed spoke with Ernest P. Chan, a global authority on quantitative trading and the application of AI in finance.

"Today, most managers have already worked with AI in some way and are already applying it. And the industry will become increasingly competitive as everyone adopts these tools," he says.

Chan is the founder and chief scientist of PredictNow.ai, a platform focused on adaptive portfolio optimization, and non-executive chairman of QTS Capital Management. He has also worked at Morgan Stanley, Credit Suisse, IBM Research, and the hedge fund Millennium.

Author of a trilogy that has become required reading in quantitative desks , Chan has just released, in partnership with Hamlet Jesse Medina Ruiz, the book Generative AI for Trading and Asset Management , in which he shows how the new wave of generative models can support everything from operational tasks to the construction of sophisticated market signals.

Chan came to Brazil last year at the invitation of Itaú Asset to participate in the Quant AI Challenge, a university competition that brings together more than 2,500 students. In this interview, he explains how AI is changing the sector, why alternative data still has limitations, and what the role of human managers will be in an increasingly automated market.

Below are the main excerpts:

Quantitative management has been around for a long time. When did AI start gaining traction in the industry?
I've been trying to use AI in finance since 1997, when I was at Morgan Stanley. But there was a long "AI winter," where the techniques simply didn't work consistently. For many years, talking about neural networks in the market generated almost complete rejection. I myself abandoned the topic for a while. Only from 2019 onwards did I start looking at AI seriously again. And then, yes, we began to see practical results. In other words, asset management also went through this cycle: a long phase where AI was more of an aspiration than a reality and, recently, a period of rapid maturation.

"Resource management has also gone through this cycle: a long phase in which AI was more of an aspiration than a reality, and, more recently, a period of rapid maturation."

Today, in which parts of the process does AI already provide a concrete advantage?
 The most obvious use is risk management. If the AI indicates high risk and is wrong, you only miss an opportunity. It doesn't break the fund. It's a domain where error is "more tolerable." Another strong use is as a research assistant. AI monitors news, investigations, legal events, relevant variations, and delivers this information in an organized way to the manager, something difficult to do manually. There is also its use in allocation optimization.

How does this work?
There's adaptive portfolio optimization. Instead of replicating a static allocation based on the past, AI adjusts the portfolio to the current context: competition between companies, technological changes, the macro environment. These aren't extreme bets, but continuous adjustments that improve the process. However, using AI to generate signals and decide trades required very robust results and, for a long time, was rare. Today we see success, especially in short-term horizons, such as high-frequency trading . It's still challenging, but it's no longer an anomaly.

In terms of adoption, where do we stand globally today?
I would say that most quantitative funds with more than $100 million already use AI in some way. Few are entirely "self-driving funds," but internal initiatives are practically universal. It's rare to find someone who says, "I don't use AI for anything."

And what about discretionary management?
AI has become a very strong support tool: summarizing information, generating ideas, monitoring events, and managing risk. But, in general, it doesn't make the final decision. In quantitative funds, however, it does. AI can already generate signals, define positions, and weigh the portfolio.

"Generally, it doesn't make the final decision. In quant funds, it does. AI can already generate signals, define positions, and weigh the portfolio."

Has this already spread to macro strategies and less liquid assets?
Yes. Many discretionary managers come to us to build macroeconomic forecasting models: inflation, activity, confidence indicators. Forecasting macroeconomics is, in a way, easier than forecasting stock prices, because its operation doesn't affect the outcome. AI functions as a macroeconomic "oracle" that informs human decisions.

What kind of new data has AI brought to the table?
The first leap was leveraging massive amounts of structured data that were previously difficult to use. For example, millions of credit card records. AI can transform this into useful information about consumption. Then comes text: central bank speeches, press releases, conference calls . AI reads, summarizes, and extracts information that generates prices. And the next level is unstructured data, such as images and weather, with satellites monitoring ports, crops, factories; hurricane maps and risk assessments for refineries. But this use is still restricted to the largest firms because it requires expensive research with uncertain returns.

More data always means better models?
 For a discretionary manager, yes, because any relevant data today helps. For quantitative models, not necessarily. The bottleneck is historical data and consistency. Many alternative datasets are expensive and have short series. If the provider changes the collection method mid-process, the data may become unusable for backtesting . So, despite the enthusiasm, a large portion of these datasets is still not suitable for robust quantitative models.

Can AI and new data help managers beat the index in liquid markets?
 They can, but it depends on the time horizon. If you say that "managers don't beat the market," try saying that to high-frequency firms like Hudson River Trading or XTX. They do outperform buy and hold in short-term windows. In longer time horizons, AI helps incrementally: better macro forecasts, better optimization, better risk management. This doesn't guarantee consistently outperforming indices, but it improves the process and the risk-adjusted return.

"If you say that 'asset managers don't beat the market,' try saying that to high-frequency trading firms. They do outperform buy and hold strategies in short-term windows."

Will we see fully autonomous trading over long time horizons as well?
I believe so. At high frequencies, this is already a reality. But there's the competition factor. If everyone has powerful models, the alpha disappears quickly. The market becomes so efficient that the advantage returns to those who manage risk best. And there's the risk of crowding. If everyone detects the same event – a hurricane in Texas, for example – and reacts the same way, the benefit diminishes. Even so, everyone will continue using AI, because when someone gives up, new opportunities arise.

Is building models in-house expensive? Is it worth it?
 Not really. You don't need to create algorithms from scratch. Google, OpenAI, and others already provide the foundations. The real cost lies in hiring people who know how to apply AI to financial data. And, often, it's more worthwhile to have someone from finance who has learned AI than a big tech engineer who has never dealt with markets. I don't recommend buying expensive, generic solutions. What really works is usually available for free. The difference comes from internal customization.

Does AI make the sector more asset-light and reduce costs?
 It increases productivity – especially for technology teams – but doesn't drastically reduce structural costs. Its biggest gain is in a better Sharpe ratio , not in operational expenses. For most funds, AI doesn't replace heavy infrastructure. It improves processes.

And what about jobs?
 Operational functions and mechanical IT tasks may decrease. AI generates code, automates routines, and reduces repetitive steps. But I don't see a drop in demand for portfolio managers. AI answers questions. Humans formulate questions. AI doesn't decide, on its own, which market to target or which strategy to create. That's still a human task.

Does this create a war for talent?
 Yes. But the curious thing is that, today, a big tech company pays AI specialists better than Wall Street. Therefore, finance is three to five years behind the technology sector in terms of the sophistication of its tools.

What advice would you give to Brazilian managers who are starting out with AI?
Start with low-risk, low-cost applications: data summarization, risk management, research. Test, learn, expand. Don't wait to find the perfect application before acting. AI is practice, not theory.

What will the industry look like in ten years?
AI will be present in all stages of management. Some funds will already have AI as the main decision-maker. Others, as an advanced co-pilot. Competition will become even more intense because everyone will have access to the same tools. The differentiating factor will continue to be human: whoever asks the right questions.