AI that automates parsing of machine-readable filings shifts sell-side analysts from routine processing to private-information work, yielding quicker, bolder and more accurate earnings forecasts and expanded coverage; a Morgan Stanley rollout of AskResearchGPT provides quasi-experimental evidence consistent with this reallocation.

Main Finding

AI investments at sell-side investment banks change how analysts allocate effort: AI tools automate processing of public, machine-readable disclosures (like iXBRL 10-Ks), producing faster earnings forecasts, and the time/capacity saved is reallocated toward acquiring private information—yielding higher-quality and bolder forecasts, broader coverage, and more active information-seeking (e.g., greater participation in earnings calls). Results hold across robustness tests and a quasi-experiment around Morgan Stanley’s internal generative-AI launch (AskResearchGPT).

Key Points

• Two-step framework (informed by in-depth analyst interviews):
• AI facilitates automation-assisted processing of public, structured disclosures (improves extraction and timeliness).
• Freed time/capacity is reallocated to private information acquisition and synthesis, enhancing analysts’ human value.
• Empirical patterns consistent with the framework:
  • AI investments are associated with more timely earnings forecasts after 10-K filings, especially after filings became more machine-readable via iXBRL.
  • Analysts at AI-invested banks produce forecasts that are higher-quality and bolder, with stronger effects when private information is relatively important and accessible.
  • AI is associated with expansion of coverage into new firms and industries.
  • AI-linked analysts exhibit greater information-seeking behavior, notably increased participation in earnings conference calls.
• A within-firm quasi-experiment (AskResearchGPT at Morgan Stanley) yields results consistent with the cross-sectional analyses, supporting a causal interpretation.

Data & Methods

• Data sources (as described in the paper):
  • Measures of banks’ AI investments / adoption (internal disclosures, technology initiatives).
  • Analyst forecasts and metadata (timing, accuracy) from standard repositories (e.g., IBES or similar).
  • Filing dates and machine-readable tagging adoption (iXBRL implementation timing).
  • Conference call participation and transcripts / attendance proxies.
  • Coverage lists and counts by analyst/firm/industry.
  • In-depth interviews with sell-side analysts to structure hypotheses and mechanisms.
• Empirical strategy (high-level):
  • Event-study and difference-in-differences style analyses comparing analyst behavior/timeliness around 10-K filings, exploiting variation in bank-level AI investment and in the machine-readability of disclosures (iXBRL).
  • Tests linking AI exposure to forecast quality (accuracy) and boldness (deviation/dispersion measures), with heterogeneity when private information is more salient.
  • Analyses of coverage breadth and measures of information-seeking (e.g., conference call participation rates).
  • Quasi-experimental evidence using the rollout of AskResearchGPT at Morgan Stanley as a plausibly exogenous shock to analysts’ AI tooling.
  • Standard robustness checks and controls (analyst- and firm-level covariates, fixed effects, alternative specifications) to mitigate confounding.
• Outcome measures emphasized:
  • Forecast timeliness (timing relative to 10-Ks and other disclosure events).
  • Forecast quality (forecast error / absolute error).
  • Forecast boldness (aggressiveness relative to consensus or prior forecasts).
  • Coverage counts and entry into new industries.
  • Activity indicators for information-seeking (conference call participation, perhaps proprietary meeting measures).

Implications for AI Economics

• Complementarities between AI and human labor: AI automates routine public-information processing, increasing analyst productivity and shifting the marginal value of human effort toward activities that require human judgment and private information.
• Recomposition of analyst value-add: The sell-side analyst role may evolve from public-data processing toward more investigative, relationship-driven, and interpretive functions—potentially increasing the premium on networking, access to management, and sector expertise.
• Market outcomes and price discovery: Faster incorporation of public disclosures (due to AI) could reduce public-information frictions and shorten price discovery windows; simultaneous increases in private-information gathering could intensify information asymmetries across firms and investors depending on access to AI-enhanced sell-side research.
• Competitive dynamics and inequality: Banks that invest in AI may obtain a competitive edge in analyst productivity and coverage scope, potentially amplifying disparities in research quality across institutions and affecting which firms receive analyst attention.
• Policy and regulatory considerations: Greater automation in processing filings underscores the value of machine-readable disclosures (iXBRL) and raises questions about fair access to AI-enhanced research tools, potential concentration of informational advantages, and monitoring for misuse of private information.
• Directions for future research: Long-run effects on analyst careers/compensation, turnover, investor returns from AI-enhanced research, liquidity impacts, and whether firms alter disclosure/communication strategies in response to more rapid and AI-assisted analyst scrutiny.