Source: Shanthikumar & Yoo (2026) — “Beyond Automation: AI and the Human Value of Sell-Side Analysts”
The question of what happens to professional roles when AI automates part of their work is usually posed as a replacement question. This paper reframes it as a reallocation question — and the reallocation finding is more interesting than a simple displacement story.
The researchers study investment bank AI investments and their effect on sell-side analyst output. The result is not analyst displacement. It’s analyst enhancement — but with a specific structure.
AI investment leads to higher-quality earnings forecasts and more timely forecasts following 10-K filings. The improvement is particularly strong after the implementation of iXBRL, which made filings more machine-readable, suggesting that the AI gains come from processing structured, publicly available information more efficiently. AI is handling the routine data extraction and synthesis that analysts previously did manually.
Here’s where it gets interesting. The time freed by automation doesn’t sit idle — it gets reallocated to private information acquisition. Analysts at AI-invested banks issue bolder earnings forecasts (which implies they have more private information to back up non-consensus calls). They expand coverage to new firms and industries. They participate more actively in earnings conference calls and engage in more strategic behaviour that facilitates management access.
The net picture: AI does the public-information legwork, and analysts use the freed capacity to do the higher-value private-information work that AI can’t do. The human comparative advantage — relationship building, management access, industry intuition, the judgment required to identify which private signals matter — is preserved and in some ways amplified because the analyst can now spend more time on it.
The additional evidence from Morgan Stanley’s AskResearchGPT rollout is consistent with this pattern. An in-house generative AI designed for research produces the same qualitative effects: better public-information processing, more capacity for private-information work.
This is the framework I’d use to think about AI in most professional service contexts. The question isn’t “will AI replace this job?” but “which parts of this job does AI do better, and does freeing up human capacity for the remaining parts create more or less value overall?” In this case: more.