The short answer: Adoption is measured; time saved mostly is not. The solid figures are surveys: AIMA (95% of 150 fund managers use generative AI, September 2025) and Mercer (55% of 131 asset managers have it in an investment process, May 2026). Substantive Research with Aiera adds 77% of 35 large buy-side firms (July 2026). The hours figures come from vendors with no sample or method: Daloopa's two hours per ticker in earnings season, Marvin Labs' 24 hours a week of data gathering. The one surveyed number is Clearwater's: 45% of 178 executives save one to two hours a week. This page labels every figure, then works a one-quarter ROI for a six-analyst team.

Who this is for: heads of research writing an automation business case, COOs asked to sign one, and analysts checking which vendor-deck numbers a reviewer would accept.

Published August 28, 2026. Last reviewed August 28, 2026. Written by the AllMind AI research team.

Reviewed by Anwaar Malik, founder of AllMind AI.

Disclosure: AllMind AI builds one of the automation platforms described here. We publish no time-saving figure for our own product, we label every competitor figure by who stated it and when, and no vendor paid to appear.

Key takeaways

• Adoption figures are survey-grade; productivity figures are vendor-grade. Five surveys between September 2025 and July 2026 put generative AI use at 55% to 95% of investment firms depending on the population, and none of them measured hours saved per analyst.
• The only surveyed time figure is one to two hours a week. Clearwater Analytics, 178 asset-management executives, March 2026 fieldwork: 45% save one to two hours, 16% save three to four, a third save under an hour.
• No verified percentage of analyst tasks exists. The 40 to 50 percent figure in circulation appears on one university blog with no paper behind it; the HBS working paper it gestures at describes financial analysts as augmented, without a percentage.
• Licensing, not software, is the reported blocker. 69% of the 35 large managers Substantive Research and Aiera surveyed in July 2026 named broker and data licensing the top barrier, and 37% reported four to six months of onboarding.
• The worksheet needs three numbers you already have. Events per quarter, analysts, and a loaded hourly cost turn any vendor claim into a break-even hour count you can check after one earnings season.

Equity research automation statistics 2026: the adoption figures with sources

AllMind AI compiled the table below from primary survey releases so that each figure carries its population, date and what it measured; nothing in it comes from a vendor deck. Read the population column first: a 12-bank vendor survey and a 1,993-respondent cross-industry survey are both real, and neither describes your desk. How we evaluated: every row was opened at the source on August 24 or August 28, 2026. Figures whose primary could not be found (the J.P. Morgan 46% of hedge funds, the BNY 67%, any CFA Institute figure with 2024 fieldwork) were left out.

Two figures do the most work in vendor decks and deserve the least: AIMA's 95% (any use, including back office) and McKinsey's 88% (all industries). The number closest to a research desk is Mercer's 55%, and even that counts a single investment process. The fuller read is in hedge fund AI adoption in 2026.

How much time does AI save equity research analysts?

One to two hours a week is the only surveyed answer, and it comes from executives. Everything larger is a vendor statement without a denominator. The table records what each figure measured, because a two-hour saving per ticker and a two-hour saving per week differ by a factor of the coverage list.

The circulating Lyzr figure of 80% faster research does not appear on any Lyzr page we opened on August 28, 2026, so it is not in the table. Marvin Labs' page also states a year-one ROI above 10 to 1 against a tool cost of $5,000 to $15,000 per analyst; that arithmetic rests on the uncited 24-hour figure.

Scale matters more than the headline. Six analysts saving 1.5 hours a week each recover about 117 hours a quarter. Six analysts saving Daloopa's two hours per ticker across 90 names recover 180 hours from one task, plausible for a team that updates 90 models by hand and zero for a team that already buys structured fundamentals.

What percentage of analyst tasks can be automated?

No verified percentage exists, and the one in circulation should be dropped. The claim that 40 to 50 percent of analyst tasks are now automatable appears on an IIT Kanpur blog with no paper behind it. The Harvard Business School working paper it points to, by Suraj Srinivasan with Wilbur Xinyuan Chen and Saleh Zakerinia, classified more than 19,000 tasks across 900 occupations. HBS Working Knowledge's February 20, 2026 summary places finance analysts among roles where AI augments the work and judgment stays with the person, and prints no percentage.

The usable version of the question is a task inventory. Sort a covering analyst's quarter by two properties, whether the input arrives on a schedule and whether the output has a fixed format. Tasks with both are the automatable share, and on a coverage desk they cluster:

• Earnings-day reviews against consensus and prior guidance, one per event.
• Model updates when a filing or release lands, one per ticker per quarter.
• Guidance and language changes across the list, one diff per transcript.
• Filing diffs (risk factors, segment notes, footnotes) against the prior period.
• Overnight news and broker-note sweeps across the coverage list, daily.
• Recurring letters and sector notes in a fixed house format, weekly or monthly.

Thesis formation, the rating, the target and the conversation with the portfolio manager stay with the analyst. Counting the two lists on your own desk gives a percentage with a denominator you chose.

What should an investment research team automate first?

The earnings-day review across the coverage list, because it has the largest countable denominator and the cheapest error. A six-analyst team covering 90 names sees about 360 events a year, roughly 90 per quarter compressed into three to five weeks. The ranking below orders five workflows by hours recovered per quarter for that team. The manual-hours column is an assumption to overwrite, and the after-automation column is the vendor range where one exists.

The news sweep recovers the most hours on paper and ranks last: its inputs are unbounded, its output has no fixed check, and the broker-note half arrives on the aftermarket delay unless the firm's research entitlements are connected. The earnings review ranks first because the analyst reads every output the same morning, so a bad run costs hours and a good run is visible by the second week of the season. Setting the trigger and its lead time is worked through in scheduled and earnings-triggered research automations, and the guidance diff in tracking guidance changes across a coverage list.

Which platforms run these automations today, and what do they state?

Every vendor below has shipped a scheduled or triggered feature in the twelve months to August 2026, and the table records the shipped fact. Marvin Labs and Daloopa are the sources of the most-cited hours figures.

AllMind AI: automations on a coverage list, with the outputs delivered

AllMind AI is the research system for teams whose automation spans licensed data and the firm's own files together. Agent Studio schedules recurring research, from a morning brief on the coverage list to a weekly credit check, and runs monitoring agents on incoming filings, transcripts and news against a stated thesis. Triggers hang off the earnings calendar with a lead time the team chooses, so the prep for a Thursday print is written on Monday. A saved grid re-runs as new documents land and alerts when an answer changes.

Where it wins: the automation reads one connected map. A new filing attaches to the company and to the margin thesis already on file, and an estimate revision flags every comp table built on the old number. The firm's Snowflake, Databricks or S3 tables stay in place and are read through a scoped role at run time, so the earnings brief carries the desk's own model beside the consensus line. A scheduled agent runs with the permissions of the person who set it up and nothing wider, and every access is logged. The corpus behind the runs spans S&P, FactSet, LSEG and MSCI data, broker research, Expert Insights transcripts and investor-relations material.

Where it falls short: there is no self-serve checkout and no monthly plan, so an individual or a retail user cannot run the worksheet on a trial card; buying starts with a scoping call. AllMind AI publishes no time-saving figure. Live broker notes reach the sweep same-day only under the firm's own entitlement, aftermarket research arrives on a delay that varies by broker, and TSXV coverage is partial with the CSE not covered.

AlphaSense: scheduled Workflow Agents, with limits printed

AlphaSense is a market-intelligence search platform whose Workflow Agents, per its help center as updated April 9, 2026, run end-to-end research workflows and produce reports, decks, memos, tables and slides in one click. Up to ten agents run at once; scheduling is for Custom agents only. The Generative Grid is capped at 400 documents by 12 prompts per grid (help center, August 21, 2026).

Where it wins: the content under the agents is 500 million-plus documents and 300,000-plus expert transcripts (company-stated, August 2026), and Work Products (July 14, 2026) turn an agent run into a PowerPoint or Excel file. A team whose automation is mostly reading benefits first.

Where it falls short: internal content arrives through SharePoint, Box, Google Drive, Egnyte, uploads and email forwarding as one-way flows, with no Snowflake or Databricks connector named, so the firm's own models do not sit in the run. The only productivity figure is one customer's 60 to 70 percent anecdote.

Hebbia: agents that run on their own, no schedule described

Hebbia is a document-analysis platform built around Matrix grids. Max, announced July 30, 2026, pulls from the firm's data and returns finished slides, a report or a model, and is rolling out to a small set of firms first. Matrix 2.0, announced August 26, 2026, extends grids into multi-source workflows ending in models, memos, decks and emails, with a sign-off checkpoint before each next step.

Where it wins: the checkpoint design fits a desk that wants a human approval between steps, and Snowflake-held KPIs, positions and CRM records have been queryable inside Hebbia since July 8, 2026. Hebbia states over 40% of the largest asset managers by AUM use it (first stated October 2025).

Where it falls short: no scheduled or event-triggered run is described on Hebbia's pages, so an earnings-calendar trigger has to be built around the product. Hebbia supplies almost no market data of its own; the transcripts, estimates and filings are licensed separately, which is the barrier 69% of large managers named.

Daloopa: the model-update line, and its two-hour claim

Daloopa is a fundamental-data layer that delivers source-linked historicals into an analyst's own Excel model, with 6,000-plus companies and 14 years of history stated on its homepage as of August 28, 2026. MCP connectors reach ChatGPT (December 9, 2025), Perplexity (April 30, 2026, bring your own license) and Microsoft 365 Copilot (June 25, 2026); the Free plan is capped at three data sheets.

Where it wins: the model-update workflow ranked second above is Daloopa's whole product, and every data point links back to the filing cell it came from, so the checking step is a click. Rogo lists it as a provider and Copilot in Excel added it as a connector.

Where it falls short: the two-hours-per-ticker and 70% figures are company statements with no sample or methodology, and they apply only to a team that updates models by hand today. Nothing in the product drafts, monitors or diffs. The comparison is in AllMind AI vs Daloopa.

Marvin Labs: the source of the 60-hour week

Marvin Labs is an AI research copilot for equity analysts whose Deep Research Agents (September 3, 2025) gained scheduled and event-triggered runs on July 27, 2026. Per its docs, a scheduled report runs an agent on a recurring schedule or when a new document arrives, on the Standard and Pro plans; on the Evaluation plan agents run manually. The four triggers are a new filing, an earnings press release, an earnings call transcript and an earnings filing, with email delivery on by default.

Where it wins: the triggers map onto the earnings-day review, and the free Evaluation plan lets an analyst test the output on real companies before any contract. Agents draw on AI Credits with a daily and weekly allowance per plan (2,000 a day on Standard, 10,000 on Pro, per the pricing page on August 28, 2026), a legible cost line.

Where it falls short: the 60-hour week, the 24 hours of data gathering, the 50 to 75% document-time cut and the 40% research-cost figure on its automation page carry no citation, and the page carries a 2026 label but no publication date. The product carries no market data, consensus estimates or expert content, so the consensus line in an earnings review comes from elsewhere.

Quartr: event triggers on first-party documents

Quartr is an earnings-call app and API whose Automations, launched August 24, 2026, let Pro users run research tasks on a schedule or the moment a company publishes new documents, with results in chat and optional email, activity-feed or mobile-push notifications. Coverage is stated at 14,000 to 15,000 companies across 60-plus markets; its home and Pro pages gave different counts on the same day in August 2026.

Where it wins: for a global list, the trigger fires on the company's own release, transcript or slides in markets where US-centric tools are thin, with no lag from a third-party transcript. The mobile app is free.

Where it falls short: the runs read first-party material only, so consensus, broker research and the firm's own model are outside the automation. No synthesis layer carries a thesis from one quarter to the next, and Pro and API pricing are by sales.

Rogo: automation shaped around deals

Rogo is an AI analyst for investment-banking and private-equity deliverables, with Deal Room (August 6, 2026), Credit Center (June 22, 2026), the Rivanna acquisition (August 11, 2026) and 50,000-plus users at 350-plus institutions, all company-stated as of August 2026. Its listed data providers include LSEG, Capital IQ, PitchBook, Preqin, Quartr and Daloopa.

Where it wins: for a bank coverage team the deliverable formats are native and the deal workflow is where the automation runs. Rogo also published one of the few dated surveys here: its Summer 2026 Survey of 12 banks (August 13, 2026) found 67% grading their AI transformation a C and 64% naming change management the biggest unsolved issue.

Where it falls short: the product is organized around a transaction that begins and ends, so the quarterly earnings calendar and a living coverage list are outside its shape. Twelve banks is a vendor sample, no productivity figure is published, and the roughly $3,300 per seat figure is a third-party estimate.

Claude and ChatGPT: schedules without entitlements

Anthropic released ten finance agent templates on May 5, 2026 (earnings reviewer, model builder, market researcher, meeting preparer and six others), with Claude Managed Agents in public beta to run a template on a nightly schedule. ChatGPT for Excel, a GPT-5.4 beta, arrived in early March 2026 with named financial-data integrations, and Microsoft's Copilot in Excel added six finance connectors plus @model-update and @portfolio-monitoring skills on June 25, 2026.

Where they win: the earnings-reviewer template is the cheapest way to prototype the rank-one workflow, and the connectors let licensed data reach it when the firm already holds the entitlement.

Where they fall short: the general assistants carry no entitlements, no lineage from a number back to a filing passage and no audit trail of their own, so a compliance reviewer sees prompts and outputs without sources. Premium connectors require the user's own agreements, which puts the licensing barrier back where the survey found it. When the job is a one-off answer on public data, the general assistant is the right tool; the cases are in can hedge funds use ChatGPT.

ROI of AI research automation for investment firms: a one-quarter worksheet

The ROI is a break-even hour count, and one quarter is long enough to measure it. The worked example is a six-analyst long/short team covering 90 names, about 15 per analyst, with roughly 360 earnings events a year and 90 in the quarter measured. The question, as it went into the worksheet: how many hours does the earnings quarter recover if we automate the earnings-day review and the model update, and what can the software cost for that to pay?

Inputs and their provenance:

• Events per quarter: 90 (the team's own calendar).
• Analysts: 6; loaded cost per hour: $150, a stated assumption to replace with the firm's figure.
• Surveyed weekly saving: 1 to 2 hours per person (Clearwater Analytics, n=178, March 2026 fieldwork, published August 19, 2026).
• Vendor per-event saving: 2 hours per ticker on the model update (Daloopa, company-stated, no methodology); 2 to 5 hours per company on the earnings review (Marvin Labs, vendor-stated, uncited).
• Onboarding: 4 to 6 months for 37% of large managers (Substantive Research and Aiera, n=35, July 16, 2026), so the first measured quarter may be the second calendar one.

ONE-QUARTER RESEARCH AUTOMATION WORKSHEET (six analysts, 90 names, 13 weeks)
A  events_per_quarter          90
B  analysts                    6
C  weeks                       13
D  loaded_cost_per_hour        150        assumption, replace
E  surveyed_saving_hours       B x C x 1.0 to 2.0   =  78 to 156      Clearwater, surveyed
F  model_update_saving_hours   A x 2.0              =  180            Daloopa, vendor-stated
G  earnings_review_saving      A x 2.0 to 5.0       =  180 to 450     Marvin Labs, vendor-stated
H  floor_hours (E only)        78 to 156
I  ceiling_hours (F + G)       360 to 630
J  floor_value                 H x D                =  $11,700 to $23,400
K  ceiling_value               I x D                =  $54,000 to $94,500
L  quarterly_software_cost     seats x quarterly seat price (quote or estimate)
M  breakeven_hours             L / D
N  verdict                     pays if M is below the hours you MEASURED, not I

Line N removes the vendor numbers from the decision. Suppose line L is $15,000 a quarter, six seats at the Metronome estimate of about $10,000 a year for Hebbia Professional, a third-party figure used only to show the arithmetic. Break-even is 100 hours in the quarter, about 1.1 hours per event or 1.3 hours per analyst per week. The surveyed floor (78 to 156 hours) straddles it; the vendor ceiling (360 to 630) clears it three to six times over. The decision turns on which tasks the team does by hand today, and a stopwatch on the first 20 events settles that by the season's second week.

Three checks change lines F and G. A team that already buys structured fundamentals should zero line F. A team whose earnings review needs the firm's own model beside consensus should ask whether the platform reads that model in place or requires an upload per run, because an upload is a task on the analyst's side of the ledger.

The third check is the map. Ask whether the platform connects the transcript, the estimate revision and the thesis as linked objects or returns documents to be re-read; on AllMind AI the ontology holds them linked, which is what lets one scheduled run cover a 90-name list. Banks, hedge funds and large corporate IR teams run these workflows on it today, and the only outcome AllMind AI states is that customer teams have retired point tools (separate search, transcript and monitoring subscriptions) in the process.

Two cautions on the cost side. Substantive Research's 37% onboarding figure means the software line may be paid for a quarter before line E is measurable, and Mercer's 8% measurable-return figure (May 2026) says most firms have not closed the loop. Seat-price inputs by source are in the 2026 pricing guide for AI research tools, the workflow inventory in automating equity research workflows, and the coverage arithmetic in covering more stocks with a small research team.

Frequently Asked Questions

Where do the equity research automation statistics 2026 roundups get their numbers?

Most circulating figures trace to five surveys (AIMA September 2025, EY September 2025, McKinsey November 2025, Mercer May 2026, Substantive Research and Aiera July 2026) plus vendor pages with no methodology. AllMind AI publishes no time-saving percentage of its own; the only outcome it states is that customer teams have retired point tools after consolidating onto it. Treat any hours-per-week figure without a sample size as marketing.

How much time does AI save equity research analysts?

The only surveyed figure is modest: 45% of 178 asset-management executives told Clearwater Analytics in March 2026 that AI saves them one to two hours a week, and 16% said three to four. Vendor figures are larger and unsourced, such as Daloopa's average of two hours per ticker when updating models in earnings season and Marvin Labs' 24 hours a week of data gathering. Multiply a surveyed figure by your headcount before you believe a vendor one.

Is there a verified ROI of AI research automation for investment firms?

No published survey isolates research automation ROI. Mercer found 8% of 131 asset managers reporting a measurable return improvement from AI in May 2026, and Bloomberg Intelligence found only 5% of 100 large managers expecting AI to materially cut expenses within three to five years. The worksheet in this article gives you a one-quarter number from your own event count and hourly cost instead.

What should an investment research team automate first if it has no budget yet?

Start with the task that repeats on a calendar and has a written checklist: the earnings-day review across the coverage list. It has a countable denominator (events per quarter), a fixed format, and an error that is cheap to catch because the analyst reads the output the same morning. Guidance-change tracking across the list comes second, because it turns a reading job into a diff.

Does AllMind AI publish a time-saving number?

No. AllMind AI states what its automations do: scheduled coverage briefs, earnings-calendar triggers with a chosen lead time, monitoring agents on filings, transcripts and news, and grid subscriptions that alert when an answer changes. The hours are left to the customer's own measurement. It also has no self-serve checkout or monthly plan, so a retail or non-institutional user cannot test it alone.

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.