AI Tools Boost Financial Analysts' Research Breadth—But Paradoxically Hurt Forecast Accuracy

Researchers investigating how generative AI reshapes financial analysts' work have uncovered a paradox: while AI dramatically expands the scope and speed of research reports, it does not reliably improve the quality of investment forecasts. The tension points to a fundamental constraint—human attention—as a newly critical bottleneck in the information-driven financial sector.

The Productivity Gains Are Real#

When FACTSET, a major financial-data vendor, integrated generative AI into its platform in 2023, analysts began producing markedly more comprehensive reports. Compared to earlier work:

• Reports cite 26% more distinct information sources, pulling from a wider pool of data and research
• Coverage spans 24% broader topical ground, expanding the range of factors analysts examine
• Analytical techniques increase by 21%, reflecting more sophisticated methodologies
• Reports also arrive faster, improving timeliness of insights

These are substantial improvements in information production. AI systems appeared to liberate analysts from routine data-gathering and synthesis work, enabling them to cast a wider analytical net.

But Forecast Accuracy Stumbled#

Paradoxically, this richer information landscape did not translate into better predictions. When analysts had to process greater volumes and variety of material—a direct result of AI's expanded information supply—the accuracy of their forecasts relative to actual market outcomes declined noticeably.

The puzzling part: a machine-learning model trained on the same observable inputs that analysts saw showed no such deterioration. This suggests the problem was not that the underlying data became less predictive, but rather that human analysts, overwhelmed by choice and complexity, struggled to synthesize and act on the available evidence effectively.

A Human Processing Problem, Not Bad Data#

The finding isolates the constraint to cognition rather than information quality. Generative AI has relaxed the bottleneck on data acquisition and report-writing speed, but it has inadvertently exposed a different, older limitation: the finite capacity of human attention and decision-making under information load. Placebo tests using competing data vendors ruled out broad platform trends, reinforcing the conclusion that the FACTSET GenAI integration itself drove the observed shift in analyst performance.