The more Wall Street relies on AI, the more it starts to think alike. In this article, Alex Tsepaev, Chief Strategy Officer at B2PRIME Group, shares his insights on AI.
There is an old saying in markets: when everyone agrees, someone is wrong. And following this wisdom, asset managers have always competed on the quality of their insights, seeking to make their research as original and as close to the truth as possible. Now, a new force is reshaping this perception, and, as I see it, not necessarily in the industry’s favour.
The thing is, current surveys found that 55% of asset managers have integrated AI into at least one investment process, and 91% plan to increase their use over the next twelve months. This is, by many measures, a genuine efficiency gain.
However, AI is far from smooth sailing, as its adoption raises a question that the industry has barely begun to grapple with. If investment firms increasingly rely on similar models, trained on similar datasets, following similar prompts, will their portfolios start to look the same?
What AI is already doing well
To begin with, let’s be clear about where AI earns its place. The most measurable results the technology brings in financial analysis so far are operational. Data show that 73% of surveyed firms use AI to automate routine tasks, and 68% use it as an analytical co-pilot for generating basic insights.
This basic analytics includes, for instance, processing earnings call reports or scanning regulatory filings. AI proved very effective at these routine tasks, and managers confirm it can now do in minutes what used to take analysts thousands of hours.
Research from Harvard Business School goes further, as it revealed that AI can already replicate a large share, namely 71%, of mutual fund managers’ investment behaviour. But here is the twist: this can only work for managers following relatively consistent stock selection. That said, AI can anticipate their next trades only when the trading pattern is repetitive.
Although the boundary is not fixed, the direction of travel is clear. AI models will grow more capable, and the ceiling on what AI can replicate is rising. But are there any limits?
The risks firms underestimate
Given that only 5% of firms grant AI autonomous or semi-autonomous decision-making authority, it seems that, at least for now, AI’s potential is growing quite slowly.
The thing is, the technology so far cannot replace human judgment, especially when it comes to investments. Most of the AI models have a serious bias when training on the same data. For example, a model that used the last two decades of equity information may systematically underweight tail risks or encode assumptions that are irrelevant.
Moreover, the data it is trained on is indeed common to many models, which results in analyses produced by different analysts (that use AI, of course) clustering. Many firms have already reported this issue, but this number is definitely growing.
The other serious reason why AI should be prevented from going mainstream in finance is poor explainability. The reasoning that AI offers is usually opaque, and the technology tends to incorrectly flag potentially good investments or counterparties. In the end, this means that a human analyst needs to intervene almost every time to avoid costly mistakes.
For that reason, the Treasury Select Committee has specifically called on the FCA to publish guidance on senior manager accountability for AI-driven harm. Even if AI makes decisions in your firm, the accountability is still on a human.
The Financial Policy Committee has also flagged risks that AI poses — the same convergence of opinions. The Committee notes that correlated behaviour across institutions can lead to another economic shock, as it diverts markets from diversification.
Human judgement results in premium
All in all, AI can make individual firms more efficient, but markets as a whole less so, at least for now. The same technology that accelerates analysis and cuts costs for individual firms may, in aggregate, erode the very disagreement that makes the markets actually work efficiently.
Yes, AI can replace your repetitive and routine tasks, but when the strategy becomes unique, it struggles to anticipate. This way, managers who invest in distinctive ways tend to greatly outperform their more replicable peers — there is already a 20% rise in postings for roles needing analytical or creative work.
This is, in a sense, a relief to those who were anxious about the AI’s first results in financial services. From the advent of technology to now, the fear has been that machines would make human analysts redundant. But the evidence shows the opposite: AI will make average human judgment redundant, but exceptional judgment will become more valuable than ever.
