Image: Caleb Luke Lin
IT IS AN article of faith among stockmarket professionals that their work involves a special human ingredient—judgment, intuition, perhaps even a bit of magic—that cannot be replicated by machines. A recent experiment suggests they may need to rethink that conviction.
Six leading artificial-intelligence (AI) models were pitted against seasoned equity analysts to produce “SWOT” analyses, evaluating the strengths, weaknesses, opportunities and threats of three companies. The results were arresting. In many instances, the AIs did not merely match the humans; they surpassed them, identifying risks and strategic gaps that the experts had missed.
The test examined the most highly rated stocks in America, Europe and Japan: Kirby Corporation, Deutsche Telekom and Daiichi Sankyo. These were not difficult cases on which to spot flaws, but market favourites, the sort that attract overwhelmingly positive coverage. The fact that AI could find weaknesses where humans saw only strengths is telling. It suggests the technology has the potential not only to assist with analytical tasks but to challenge consensus thinking.
With sophisticated prompting, some large language models (LLMs), the sort of AI that powers chatbots, exceeded human analysts in the specificity and depth of their assessments. They produced more detailed and comprehensive evaluations than professionals with years of industry experience. This capability stems from their ability to synthesise vast quantities of data and recognise patterns that might elude even the most diligent person.
Yet there are limits. An AI cannot observe a chief executive’s body language or detect the subtext in management’s “cautiously optimistic” guidance. As one portfolio manager notes, talking to company bosses remains essential to understand how they truly view their business.
The single biggest factor in the AI models’ performance was the quality of the prompts they were given. Advanced prompting improved their output by as much as 40%. The difference between a basic request for a SWOT analysis and a carefully crafted set of instructions was the difference between a superficial summary and institutional-grade research. Prompt engineering is becoming a core skill for investment professionals, as fundamental as spreadsheet proficiency was two decades ago. Those who master it will gain far more value from AI tools; those who do not will be left behind.
Not all AI models are created equal. The experiment evaluated six of the most advanced: Google’s Gemini Advanced 2.5, OpenAI’s o1 Pro, ChatGPT 4.5, Grok 3, DeepSeek R1 and ChatGPT 4o. The reasoning-optimised models, those with “deep research” capabilities, consistently outperformed the standard versions. They provided greater context, better fact-checking as well as fewer generic statements. The best models took ten to 15 minutes to produce comprehensive SWOTs, whereas simpler ones delivered results in under a minute. The correlation between thinking time and output quality is one that human analysts know well.
The way forward for investment professionals is a hybrid approach. AI should be used for heavy lifting: initial research, data synthesis and pattern identification. Human judgment should be reserved for interpretation, strategy and genuine insight into management thinking. The optimal workflow involves AI drafting and humans refining. Firms ought to develop libraries of standardised prompts for common tasks; a well-crafted prompt is a form of intellectual property. Model selection matters, too. For deep analysis, reasoning-optimised models are worth the cost. For quick summaries, standard versions suffice.
Although the experiment focused on SWOT analysis, the implications extend across the investment process. AI can summarise and analyse earnings calls in minutes rather than hours, identify ESG red flags across entire portfolios, parse regulatory filings at scale, gather competitive intelligence and synthesise market sentiment. Each application frees human analysts for higher-value work.
Will AI replace analysts? Not entirely, but it will replace analysts who do not use AI. The combination of human and machine will outperform either alone. Can the output be trusted? It should be verified. AI can invent facts or miss context, making human oversight essential. Which model is best? For complex analysis, Gemini Advanced 2.5 or o1 Pro are the leaders, but given the pace of change, the answer should be reassessed quarterly.
The genie is out of the bottle. LLMs can perform in seconds analytical work that once took days. They bring speed, consistency and vast knowledge. Used well, they are like a tireless team of junior analysts. But success is dependent on thoughtful integration, not wholesale adoption. AI output should be treated as a junior analyst’s draft—valuable input that requires senior review.
The future of investment analysis is not human versus AI, but human and AI. The winners will be those who combine the computational power of machines with human insight, intuition and relationship-building skills. The next analyst hired will still need to understand companies and industries. But they will also need to know how to prompt an LLM, evaluate its output as well as add the human spark that turns data into investment returns. The tools are here. Those who learn to use them will have the edge. ■
