The short answer: 95% of the fund managers AIMA surveyed use generative AI (150 managers, published September 16, 2025), and 55% of asset managers have AI inside at least one investment process (Mercer, 131 managers, February 2026 fieldwork). Only 5% let AI make or share a decision, and 8% can point to a measured return improvement. Permission is broad, process use is shallow, licensing and onboarding gate the largest firms, and analysts still out-score deep-research agents. Every figure below carries its source, sample size and date.

Who this is for: COOs and heads of research budgeting AI seats, allocators writing AI questions into due diligence, analysts reading about their own jobs, and anyone who needs a sourced statistic for a deck.

Published August 24, 2026. Last reviewed August 24, 2026. Written by the AllMind AI research team.

Reviewed by Anwaar Malik, founder of AllMind AI.

Disclosure: AllMind AI builds an AI research platform for institutional investors and appears below in the accuracy section and the pilot example. Every adoption and value figure comes from a third-party survey or paper, and no vendor paid to appear.

Key takeaways

• Permission is near-universal and process use is thin. 95% of 150 fund managers use generative AI (AIMA, September 16, 2025); 55% of 131 asset managers have it in an investment process (Mercer, May 21, 2026).
• Decision authority is rare and returns are unproven. 5% of asset managers grant AI decision authority and 8% report a measurable return improvement (Mercer, February 2026 fieldwork).
• Licensing gates the largest firms. 69% of 35 large buy-side firms cite broker and data licensing as the top barrier, and 37% say onboarding takes 4 to 6 months (Substantive Research and Aiera, July 16, 2026).
• Agents trail the headlines. 23% of 1,993 organizations are scaling AI agents (McKinsey, November 5, 2025), and 75% of 12 investment banks call this the first or second inning (Rogo, August 13, 2026).
• Analysts out-score agents and expect to keep their jobs. Analysts scored 2.84 against 2.31 for the best deep-research agent (JPMorganChase, April 22, 2026); 1% of 178 executives expect AI to replace their role (Clearwater Analytics, August 19, 2026).

AI adoption statistics asset management 2026: the numbers, dated

The table holds 29 statistics from 21 surveys and papers published between July 2025 and August 2026, each one sentence with its sample, publication date and a link to the primary document. Copy it whole. Cross-industry samples are marked.

How we chose these statistics, and what we left out

Every figure comes from a named survey or paper, opened and checked on August 24, 2026; nothing here is an observation from AllMind AI deployments. Survey figures are self-reported and benchmark figures are measured by the papers' authors, so the two are labeled and never averaged. Small samples are stated inline: AIMA's 18 investors, Rogo's 12 banks, Substantive Research's 35 firms.

Five figures readers ask about are left out on purpose.

• JPMorgan's 46% of hedge funds actively using AI, up from 18%: the primary document could not be retrieved, so it has no sample size or fieldwork date.
• BNY's 67% using machine learning or AI: the release date could not be verified.
• CFA Institute's daily and weekly AI-use percentages: February 2024 fieldwork, two model generations ago.
• 47% in production and 3 to 5% higher returns: a consultant blog with no source.
• A widely quoted 2024 paper on LLM financial statement analysis: withdrawn in February 2025 after a co-author identified inconsistencies in the data and analyses.

A vendor can also leave the market outright: Fintool did when Microsoft acquired it in April 2026, and the replacements are compared here. Vendor customer counts stay out of the adoption rows because they count one product's customers: Hebbia's over 40% of the largest asset managers by AUM (company-stated, October 2025), Rogo's 50,000+ professionals at 350+ institutions (rogo.com, August 2026), and AlphaSense's 7,000+ enterprise customers (June 3, 2026 funding release).

What percentage of hedge funds use AI?

Between 42% and 100%, depending on what counts as use; 95% is the figure most people mean, the share of 150 fund managers using generative AI in AIMA's survey, up from 86% in 2023 (AIMA, September 16, 2025). Use inside the investment process runs behind: 58% of the same managers expected to increase generative AI use in investment processes, against 20% two years earlier. The spread across surveys is a definition problem.

• Across every business function: 42% of emerging managers (AIMA and Marex, June 30, 2026).
• Moderate or large-scale use for investment research: 100% of 107 private equity, hedge fund and venture firms (Lowenstein Sandler, February 19, 2026).
• Significant integration across research and trading: nearly a third (Hedgeweek, 100+ managers, March 6, 2026).
• Not started: 54% of 150+ quants and analysts (Bloomberg Research Data Survey, January 22, 2026).

Almost every fund has someone using a general assistant, about a third have AI wired into research and trading, and a quant desk is as likely as not to have no program. 60% of the 18 investors on AIMA's panel were more likely to back a fund with a meaningful generative AI budget, a small sample worth stating as such. Firm-type detail is in how hedge funds and asset managers are using AI in 2026; the stacks funds run are in the best AI research systems for hedge funds.

How many asset managers use AI in 2026, and for what?

55% of asset managers have AI in at least one investment process, 27% are piloting it, and 5% let it make or share a decision (Mercer, 131 managers, February 2026 fieldwork). At the largest firms, 77% of 35 global asset managers have organization-wide generative AI deployments (Substantive Research and Aiera, July 16, 2026). The "for what" is mostly operations and research support: 69% of Mercer's respondents cite operational efficiency, 55% faster or higher-quality insights, and 8% a measured return improvement.

Where the AI goes, by survey.

• Research feeds. 77% of the largest firms called broker research the most valuable machine-readable input, and 69% name licensing the top barrier (Substantive Research and Aiera, July 16, 2026).
• Alternative data. 63% of 56 buy-side firms plan to spend more (Coalition Greenwich, December 17, 2025).
• Agents. 78% of 100 wealth and asset managers are exploring agentic AI (EY, September 16, 2025).

Which platforms they buy has its own page, top AI platforms for asset managers in 2026. The survey signal on platforms is early: just over a quarter of the largest firms are evaluating or have implemented a vertically integrated finance AI platform, and 44% see such platforms as potential long-term partners. A procurement view is in AI vendors for institutional investment teams.

What the accuracy benchmarks say

Professional analysts still out-score deep-research agents, and agent factuality varies by a third from product to product. In JPMorganChase's Deep FinResearch Bench, analysts scored 2.84 against 2.31 for the best agent, with factuality from 86.0% for OpenAI to 53.2% for Grok (arXiv 2604.21006, April 22, 2026). On the Vals Finance Agent leaderboard the best model answered 64.37% of 537 expert questions over SEC filings on June 4, 2026, leaving roughly one answer in three to check.

Three findings recur: report quality trails analysts even for the best agent, factuality is the widest spread in the set, and accuracy drops hardest when work crosses entities or periods. The Mercury study found forecast errors rose most for analysts covering more firms or integrating more distinct sources, which is the work an analyst is paid for.

AllMind AI, read against the benchmarks

AllMind AI is an AI research system for institutional investors, launched publicly on July 13, 2025, a financial ontology with agents on top, and the one product on this page we make.

Where it wins: on the two failure modes above. Against the factuality spread, each figure in an output opens the passage of the filing, transcript or broker note it came from, with the calculation behind a derived number shown line by line. A verification pass re-checks each figure against its source before a report goes out.

Against the cross-entity drop, the ontology holds companies, suppliers, customers, estimates, filings and the firm's own research as linked entities. An agent walks from a supplier's guidance to a customer's estimate revision along maintained relationships for as long as the workflow runs, minutes or days. Expert-call transcripts come through Expert Insights, which is included in the subscription; broker research needs the firm's own entitlement while it is embargoed and arrives on a delay after that. Agents inherit each user's entitlements and cannot widen them, and the log records each question and each export, the audit trail the 69% licensing figure implies.

Where it falls short: none of the six benchmarks graded AllMind AI, so there is no published score to set beside the 64.37% or the 2.31, and the only test that counts is a pilot on the firm's own questions. Depth has a setup cost too: the internal half of the system means connecting warehouses, dashboards and document stores, which starts with a scoping call about permissions and cannot be done from a signup page. That fits the 37% of large firms reporting 4 to 6 months to onboard.

Is AI paying off for asset managers yet?

As hours, yes; as returns, rarely, and the two should be tracked separately. 8% of 131 asset managers report a measurable improvement in investment returns from AI (Mercer, May 21, 2026). 5% of 100 CIOs and CTOs at large managers expect AI to materially reduce expenses within three to five years (Bloomberg Intelligence, February 25, 2026). Across industries, 5% of 1,250+ companies achieve AI value at scale and 60% report no material value (BCG, September 2025).

The hours are real and modest: 45% of 178 senior asset management executives say AI saves them one to two hours a week and 16% say three to four (Clearwater Analytics, March 2026 fieldwork, published August 19, 2026). Agents explain part of the gap: 62% of 1,993 organizations are experimenting with AI agents and 23% are scaling them (McKinsey, November 5, 2025).

Of the 12 investment banks in Rogo's survey, 67% graded their own AI transformation a C, 64% named change management the biggest unsolved issue, and about 40% said return on investment was too early to quantify (Rogo, August 13, 2026). Value shows up when an agent finishes a workflow that used to take a person a day, with an audit trail a reviewer can open; that layer is described in AI agents for investment research.

A 30-person asset manager's 90-day pilot, worked through

The question asked: a long-only manager with 30 investment professionals wants to know how many seats to put into a 90-day pilot of an AI research platform, and what to measure so the renewal is decided on evidence. Three statistics settle it and a fourth sets the calendar.

Seats: 12 of 30. Deloitte separates firms by whether AI reaches over 40% of the workforce, a bar 43% of high-expertise financial services firms clear against 19% of the rest (Deloitte, November 4, 2025). 40% of 30 is 12 seats, including at least two portfolio managers.

Hours: set the bar at 18 a week. 45% of Clearwater's 178 executives saved one to two hours a week, 16% saved three to four, and about a third saved 31 to 59 minutes (Clearwater Analytics, August 19, 2026). At the midpoints (0.45 times 1.5 hours, 0.16 times 3.5, 0.33 times 0.75) that is about 1.5 hours per seat, or 18 hours a week across 12 seats, roughly half an analyst; a cohort below 18 at day 90 is underperforming the survey.

Process, and no return attribution. 55% of Mercer's 131 managers have AI in at least one investment process and 8% can measure a return improvement (Mercer, May 21, 2026). Leave returns out of a 90-day scorecard; measure whether AI sits inside a recurring process step, an earnings review the day after the close or a model update on filing day, for at least 6 of the 12 seats.

The calendar: start the data connection in week one. 37% of the 35 largest firms say approving and onboarding a platform takes 4 to 6 months (Substantive Research and Aiera, July 16, 2026). So pick one deep workflow for the pilot: a quarterly review that joins the firm's own model and last memo, read from the warehouse where they already sit, to the filing, the transcript and the broker notes.

That workflow is the reason to buy a research system over a chat seat, and it is where AllMind AI runs an agent for hours across internal and external data. Score it on traceability, the share of figures that open a source passage and the count the verification pass corrected, against the 53.2% to 86.0% factuality range above.

Pricing is by quote for AllMind AI, AlphaSense and Hebbia alike, so the cost side of the case is the seats and subscriptions the pilot displaces; third-party estimates are collected in the 2026 pricing guide to AI research tools.

Will AI replace equity research analysts?

Not on the 2026 evidence, and the people in the seats expect the opposite. 61% of 178 senior asset management executives said AI increases their job security, 28% expected it to reduce it, and 1% expected it to replace their role (Clearwater Analytics, March 2026 fieldwork, published August 19, 2026). What is changing is the task content of the job and the error profile of the output.

The JPMorganChase benchmark put analysts at 2.84 against 2.31 for the best deep-research agent across 100 reports, and its authors called for finance-specialized agents (arXiv 2604.21006, April 22, 2026). The Mercury study is the uncomfortable one. Across 52,428 US reports from 409 analyst teams at 24 brokerages, reports written with AI assistance cited 40% more distinct sources and covered 34% more topics, and their forecast errors rose by roughly $0.44, 59% of the sample average (arXiv 2512.19705, December 12, 2025). Errors rose most for analysts covering more firms or integrating more sources.

Read together, the replacement question inverts: AI widened what analysts read and, unsupervised, worsened what they forecast, so the analyst's judgment became the limit on quality. A widely read CFA Institute post by Michael Schopf, CFA, reached a similar conclusion from six models and three companies' SWOT analyses: AI will displace analysts who do not adapt (CFA Institute, June 23, 2025). The 2026 data adds the sample size and one specific: the displacement risk sits with the analyst who stops checking.

Can AI (ChatGPT, Claude, etc.) replace junior analysts?

For the extraction share of a junior analyst's week, partly, and it already does; for the seat, no.

• The best model on the Vals Finance Agent leaderboard answered 64.37% of 537 expert questions over SEC filings (Claude Opus 4.7, June 4, 2026), up from 46.8% in August 2025.
• The best system on BigFinanceBench scored 58.8% across 928 expert tasks (arXiv 2606.03829, June 2, 2026).
• In Rogo's survey of 12 investment banks, about 40% of firms anchor their AI return on junior time savings, and the work still flows down the hierarchy (August 13, 2026).
• OpenAI's listing for an investment-banking subject-matter expert, at up to $205,000, was reported by The Register on July 8, 2026; Anthropic shipped ten finance agent templates for Claude, an earnings reviewer and a model builder among them, on May 5, 2026.

What the junior does that none of these grade: reconciling a filing figure to the firm's own model, noticing the footnote that moved, and chasing the answer investor relations would not put in writing. That is where the role moves over the next 12 to 18 months.

When a general assistant is the right answer: junior work that never touches entitled content or the firm's own systems, such as summarizing public filings for an internal note, is served by an enterprise assistant seat. Perplexity's Enterprise Pro is $40 per seat per month by third-party reports of vendor pricing as of July 2026. The line moves the day a compliance reviewer, entitled content or a client report enters the workflow, as worked through in ChatGPT, Claude and Perplexity against institutional research platforms.

Three predictions, dated August 24, 2026

1. Process use passes 70% by the 2027 surveys. Mercer's 55% with AI in at least one investment process crosses 70%, while decision authority stays under 15%, from 5% today.
2. Measured return improvement stays under 20%. The 8% who can show it (Mercer, May 2026) does not reach 20% in 2027, because the first gains land as hours and coverage breadth.
3. Replacement expectation stays in single digits while the job description moves. Clearwater's 1% stays under 10% in 2027, and junior postings shift toward verification and model ownership within 12 to 18 months.

Frequently Asked Questions

Can AI (ChatGPT, Claude, etc.) replace junior analysts?

Not the seat, on the 2026 evidence, but a growing share of the extraction and formatting inside it. The best model answered 64.37% of 537 expert questions over SEC filings on the Vals leaderboard of June 4, 2026, against a November 2023 FinanceBench run where GPT-4-Turbo with retrieval got 81% of 150 questions wrong or refused. In Rogo's August 2026 survey of 12 investment banks, about 40% of firms anchor their AI return on junior time savings, and junior roles are shifting from extraction toward verification.

How many asset managers use AI in their investment process in 2026?

55% of the 131 asset managers Mercer surveyed in February 2026 had integrated AI into at least one investment process, and 27% were running pilots. 91% planned to increase AI use within 12 months, while 5% granted AI autonomous or semi-autonomous decision authority. Among the 35 largest buy-side firms polled by Substantive Research and Aiera in July 2026, 77% had organization-wide generative AI deployments. EY's September 2025 survey of 100 wealth and asset managers put the share with generative AI scaled to multiple use cases at 95%.

Do asset managers see measurable returns from AI yet?

Rarely, as of mid-2026. 8% of asset managers in Mercer's May 2026 report saw a measurable improvement in investment returns, while 69% cited operational efficiency gains. 5% of 100 large-manager CIOs and CTOs surveyed by Bloomberg Intelligence in February 2026 expected AI to materially reduce expenses within three to five years. The measurable gain today is time: 45% of 178 executives told Clearwater Analytics in March 2026 that AI saves them one to two hours a week, and 27% of the 100 firms EY surveyed in 2025 reported substantial impact.

How accurate is AI at equity research compared with human analysts?

Below analysts on report quality, and uneven on facts. In JPMorganChase's Deep FinResearch Bench of April 2026, professional analysts scored 2.84 against 2.31 for the best deep-research agent, and agent factuality ranged from 53.2% to 86.0%. A study of 52,428 sell-side reports found analysts writing with AI assistance cited 40% more sources and raised forecast error by roughly $0.44, about 59% of the sample average. Accuracy also falls 18.60% and 14.35% when tasks move from one document to longitudinal and cross-entity analysis (Fin-RATE, 2026).

Which AI adoption statistics should you not cite?

Several widely repeated figures have no usable source: the CFA Institute daily and weekly AI-use percentages come from February 2024 fieldwork. The BNY figure of 67% using machine learning has no verifiable release date, and the JPMorgan figure of 46% of hedge funds using AI has no retrievable primary document. The 47% in production and 3 to 5% higher returns figures trace to a consultant blog with no source. Use the AIMA, Mercer, Substantive Research and Clearwater Analytics surveys instead, which publish sample sizes and dates.

AllMind AI is the AI-native research platform for institutional equity teams. If you want proof on your own work, send us the workflow you want tested.