Six artificial intelligence (AI) models recently competed directly against veteran equity analysts, creating SWOT (strengths, weaknesses, opportunities, threats) analysis, and the results were impressive.
In many cases, AI simply didn’t hold anything unique. It uncovered risks and strategic gaps that human experts have overlooked. This was not a theory. My colleague and I performed major tests on a large-scale language model (LLMS) against analyst consensus from three companies: Deutsche Telekom (Germany), Daiichi Sankyo (Japan), and Kirby Corporation (USA). Each was the most actively rated stock in the region as of February 2025. This is a type of “certain bet” that analysts overwhelmingly support.
If AI can identify weaknesses that humans are strong signals when looking at strengths alone, we intentionally chose market favorites. It suggests that AI can not only support analysts’ workflows, but it can challenge consensus thinking and change the way investment research is conducted.
The unpleasant truth about AI performance
With sophisticated prompts, a particular LLM must use sophisticated prompts beyond human analysts with the specificity and depth of analysis. Let’s sink it.
The machines have produced more detailed and comprehensive sweatshirts than the experts who have spent years in the industry. However, before eliminating the need for human analysts, there are important caveats. AI is excellent at data integration and pattern recognition, but it cannot read the CEO’s body language or detect the subtext of the management’s “carefully optimistic” guidance. As one portfolio manager told us, “On behalf of talking to management, there’s nothing that understands what they really think about their business.”
40% difference that changes everything
The most impressive discovery? Advanced prompts improve AI performance by up to 40%. The difference between providing detailed instructions “Give me a SWOT of Deutsche Telekom” is the difference between Wikipedia summary and facility grade studies. This is no longer an option. Engineering is as fast as Excel was in the 2000s. Investment experts who acquire this skill extract exponentially more value from AI tools. Those who can’t see their competitors will generate excellent analytics in some time.
Deliver the latest news stories and analysis directly to your inbox, starting and ending daily.
Model Hierarchy: Not all AIs are created equal
Six cutting-edge models were tested and ranked.
1. Google’s Gemini Advanced 2.5 (Deep Search Mode) – Clear Winner
2. Openai’sO1Pro – Close second with exceptional reasoning
3. ChatGpt4.5 – Solid, but especially behind the leader
4. Grok3 – Elon Musk challenger shows promise
5. DeepseekR1 – Chinese dark horse, fast but not refined
6. CHATGPT4O – Baseline for comparison
The Theansing-Optimised model (a model with “deep research” functionality) consistently outperformed standard versions such as the CHATGPT-4O. They provided more context, better fact checks, and fewer general statements.
Think of hiring senior and junior analysts. Both can do their job, but there are far fewer handhelds. Timing is also important. The best models took 10-15 minutes to generate a comprehensive SWOT, while the simpler models were delivered within 1 minute. There is a direct correlation between thinking time and output quality. This is something human analysts always know.
European AI deficit: strategic vulnerability
The unpleasant reality for European readers is: Of the models tested, five are Americans and one is Chinese. The absence of Europe from the AI ​​Leadership Committee is not only embarrassing. It’s strategically dangerous. When Deepseek emerged from China, it caused what was called the “Sputnik moment” of AI, with its competitive performance at some of the costs of the West.
The message was clear: AI leadership can change quickly, and incompetent people in the country risks technical dependence. For European fund managers, this means relying on foreign AI for critical analysis. Do you really understand that these models are aware of the European Central Bank News or German regulatory submissions and the US Federal Reserve Statement? The judges are out, but the risks are real.
A practical integrated playbook
Our study points to a clear four-step approach to how investment experts use these tools.
1. Hybrid, not exchange: use AI for heavy lifting – initial research, data integration, pattern identification. We reserve human judgment about anything that requires real insight into interpretation, strategy, and management thinking. Best workflow: AI draft, humans refine.
2. The Prompt Library is your new alpha source: Develop standardized prompts for common tasks. The cleverly crafted SWOT prompt is intellectual property. We share best practices internally, but adhere to the best prompts like trading strategies.
3. Model selection problem: Pay for inference optimization models for deep analysis. For an overview, the standard model is sufficient. Using GPT 4o for complex analysis is like bringing a knife to a shootout.
4. Continuous rating: New models launch almost every week. Our six-criteria assessment framework (structure, validity, specificity, depth, cross-checking, meta-assessment) provides a consistent way to assess whether the latest models really improve on their predecessors.
Beyond SWOT: The Expanding Frontier
Although we focused on SWOT analysis, the impact extends across the entire investment process. Some of these are listed below, but there are many more.
Revenue calls summary and analysis in minutes rather than hours
Identifying ESG Red Flags across the Portfolio
On the scale of regulatory submission analysis
Competitive Intelligence Collection
Market sentiment integration
Each application frees human analysts for valuable work. The question is not whether to employ AI. It’s fast enough to effectively integrate it.
An unpleasant question
Let’s deal with what many people think:
“Will AI replace analysts?” It’s not perfect, but it replaces analysts who don’t use AI. The combination of human and AI is superior to either one alone.
“Can you trust AI output?” I trust it, but please check. AI can hallucinate facts and miss context. Human surveillance is especially essential for investment decisions.
“Which model should I use?” For complex analysis, start with Gemini Advanced 2.5 or O1 Pro (or successor). But given the pace of change, we reassess each quarter.
“What if your competitors use AI better?” Then you’ll catch up while they find the Alpha. Staying on the sidelines while building AI advantages means making concessions in increasingly competitive situations.
The road ahead
The demon is off the bottle. LLM demonstrates that analytical work can be performed in seconds, which once took days. They bring speed, consistency and an enormous knowledge base. When used effectively, they are like having a tireless team of junior analysts who don’t sleep. But here’s the key. Success requires thoughtful integration rather than wholesale adoption.
Treat AI output in the same way as a junior analyst draft. A valuable input that requires advanced reviews. Master Prompt Engineering. Choose your model wisely. Maintain human surveillance. European experts have additional orders. We are promoting domestic AI development. Technology dependence in critical financial infrastructure is a strategic vulnerability that the region cannot afford.
Master the tools or overtake them
You can either intelligently embrace these tools or your competitors will leave you behind. The winners of this new landscape are those who combine the calculation power of AI with human insights, intuitions and relationship skills. The future of investment analysis is neither human nor AI. It’s humans and AI. Those who recognize this and act accordingly will flourish.
Someone who finds themselves outweighed themselves by people who have learned to work with them, not machines.
Hiring your next analyst may still need that coffee break. But they’ll know better how to prompt LLM, evaluate its output, and add human insights to convert the data to alpha. Because that’s the new standard for 2025. The tool is here. There is a framework. The winner will be someone who knows how to use them.
This content was taken from an article that first appeared in enterprising investors at https://blogs.cfainstitute.org/investor.
The complete research can be found here
The writer CFA is Chief Investment Officer at MHS Capinvest, employing AI tools with various market capitalizations to enhance the allocation of different market capitalizations, inventory selection, portfolio construction, risk management, and train teams of DAX listed companies with AI integration, and investment experts leverage tools like Chatgpt and Gemini to enhance performance.
