Executive Summary
AI has sparked extensive discussion in recent years about its ability to automate the work of financial analysts. To clarify this question from a practical standpoint, Opimas dedicated this report to examining several day-to-day responsibilities of financial analysts to determine where AI could help and where significant limitations remain.
Opimas evaluated which tasks AI systems can handle effectively, from competitor analysis to financial modelling, and which ones are still challenging. The methodology was as follows: the same prompt was given to each model, and each model's output was evaluated. The results were mixed, depending on the task, the industry, and the time at which the question was posed to the model. Among all the AI models tested, the weakest performer was Perplexity AI, which produced many errors and hallucinations; ChatGPT, Claude, and Grok followed. Gemini finished second and the specialized provider Brightwave achieved the best overall score (see Figure 1 and Figure 11).
Figure 1– Accuracy and completeness of answers varied greatly between tasks and AI systems.
Source: Opimas Analysis
Each AI system demonstrated distinct technical strengths. Gemini was able to extract multiple UK Companies House files with ease, but ChatGPT and Claude were not. For market sizing, Grok provided an initial bottom-up analysis, whereas the other AI models relied solely on third-party sources. Despite these specific perks, every system produced hallucinations or inconsistent results. This lack of reliability is a significant hurdle for the adoption of AI tools in the financial industry, where absolute precision is non-negotiable.
Looking beyond general-purpose AI systems and their varying performance, a similar pattern emerges within the specialized Generative AI providers for financial research. They face intense competition and growing commoditization making it difficult to differentiate themselves given their similar offerings and client bases. Some specialize in certain sectors, Rogo and Finster AI have integrations with PitchBook and are more focused on private markets, whereas LinqAlpha and Samaya are more oriented toward public markets. Nevertheless, these vendors are not immune to the garbage-in, garbage-out challenge. Because they rely on external data providers and the web, the AI models can still produce erroneous outputs, as the Opimas team observed during demonstrations. Although these vendors' auditability improves source traceability, their AI technology still lacks the judgment to distinguish accurate data from conflicting or incorrect information.
Based on these findings, Opimas concludes that AI tools are not yet mature enough to replace junior financial analysts. They can provide meaningful support, but only with careful supervision and precise instructions.