Expert networks and AI research tools converge as fund managers adopt both

Fund managers have stopped debating whether to use generative AI in research workflows. The numbers from two recent industry surveys make that clear: 95% of fund managers use generative AI in some part of their work, and 91% of asset managers say they are deploying or planning to deploy AI in investment research specifically.

That adoption rate is reshaping the expert network category, as InsightAgent argued in a recent post on the institutional research stack.

What's actually changing

The traditional expert network model (GLG, Guidepoint, Third Bridge, AlphaSense/Tegus) is built on two assets: a vetted roster of operators and ex-operators, and a compliance layer that lets buy-side analysts speak to them without tripping MNPI rules. The product is the call, plus the moderator, plus the legal cover.

AI-native research platforms are built on a different asset base: large transcript libraries, retrieval over filings and call archives, and increasingly, agentic workflows that draft questions and summarize answers. They are cheaper per-query and faster, but they don't put a human on the phone.

The convergence story is that each side is acquiring the other side's asset. Expert networks are layering AI search and summarization over their transcript archives. AI platforms are building or licensing expert rosters so analysts can escalate from a transcript search to a live call without switching vendors.

Why the survey numbers matter

Near-universal AI adoption among fund managers does not mean fund managers have replaced expert calls with AI. It means the first-pass work that used to justify a call (background reading, building the question list, finding the right industry primer) is now being done with AI tools. The expert call itself becomes higher-leverage because the analyst arrives prepared.

This is the workflow split that matters for vendor selection. Data collection and synthesis is moving to AI. Judgment, scenario building, and IC-defensible inference is staying with the analyst, sometimes informed by an expert call, sometimes not.

Where traditional networks still hold the floor

Three research questions still favor a human expert on a phone:

• Private-company diligence. Pre-IPO companies don't have transcripts, filings, or earnings calls. The information lives in the heads of ex-employees and channel partners. AI retrieval has nothing to retrieve.
• Cross-border verification. When public disclosures are thin or unreliable, a former operator with regional context is the verification layer. This is acute in China, where several networks have paused or reshaped their on-the-ground work.
• Tacit knowledge. How a sales cycle actually closes, what a procurement committee actually weighs, why a competitor's product is failing in deployment. None of this is in a 10-K.

Where AI tools win

Three other research questions favor the software:

• Public-company comp work. Reading 40 earnings transcripts to find every mention of pricing power is a retrieval problem.
• Industry primers. A new analyst ramping on semiconductors, biotech, or industrials needs structured context fast. Transcript search and LLM summarization deliver that in minutes.
• Hypothesis screening. Before committing to a $1,500 USD expert call, an analyst can use AI tools to test whether the question is even the right question.

For buy-side heads of research evaluating the stack heading into 2026, the practical question is not which vendor to pick. It is whether the team's workflow distinguishes between the research questions AI handles cleanly and the ones that still need a human on the phone, with documentation showing why.