ESG research that answers any issuer question in minutes, every claim pinned to its source page
Issuer questions answered in minutes, every claim pinned to its source page.
See one case, screen by screen ↓ESG research is the work of finding out what an issuer actually does on environmental, social and governance matters, from its own reports and filings and from outside sources, before a vote, an engagement, a buy decision or a client report. ESG Research is software for asset managers that does this with eight AI agents: ask about any issuer in plain words and the answer comes back as a verdict with every claim cited to its source page, in the original language beside the English, for an analyst to decide and a second person to approve into the house view.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
ESG research on a single issuer still starts with a score and ends in the company's own documents. Ratings from six major providers correlate only 0.38 to 0.71, so before a proxy vote, an engagement meeting or a buy decision the ESG research analyst goes back to the annual report, the sustainability report, the supplier report, the assurance statement and the regulator filing, often in Portuguese, Japanese or German, with a general translator open beside them. Then come the NGO reports, the news and the public registers, with their own denominators and periods, compared by hand in a spreadsheet.
That takes 1 to 3 hours a question, and many small-cap, emerging-market and private issuers have no vendor coverage at all. The page behind each claim lives in the analyst's head or a footnote, the house view changes by email, and the person who approves it cannot see what it rests on. When that claim later appears in a client report and cannot be traced to a page, it is the kind of ESG statement the SEC has fined advisers $4 million to $19 million for.
Estimated split for one issuer question, by hand and with the solution; the 1 to 3 hours by hand is our estimate and the pilot measures it.
Eight agents read the issuer's own pages and outside sources at once, explain every disagreement and refuse what they cannot cite; an analyst saves, the ESG lead approves.
On October 14 an NGO, Instituto Cerrado Vivo, published Rastro Verde, a Portuguese report saying only 58% of Marquessa Foods' soy is verified deforestation-free. Quillfield Capital holds $412.6M of Marquessa Foods S.A. across 7 funds, the stewardship team meets the company on Thursday, and the house view, approved in June 2025, puts deforestation risk at Medium. It is Monday, October 19, 2026. Here is what happens next, screen by screen, in the working solution.
Lena's home page is an ask bar, four numbers and a short list. 412 answers this week, all cited; 23 disagreements open, 2 overdue; 3 house-view changes awaiting approval; 1,312 issuers no rating vendor covers. At the top of Needs you: “Marquessa Foods: an NGO report says 58% of its soy is verified; we say 71%”, due today at 5:00 p.m. Overnight the agents read 1,904 new documents.
Watch agent · Wed, Oct 14, 11:02 p.m.: “NGO report on Marquessa Foods soy filed and read (Portuguese, 48 pages).”
Lena asks in plain words: “Does Marquessa Foods have a deforestation-free soy commitment in Brazil, and is it audited?” The Planner builds a reading plan of 14 sources and 1,072 pages. The Issuer Reader reads 10 of the company's own documents, 8 in Portuguese and one a scanned annex, while the Outside Sources agent reads the NGO report, two vendor records and the federal embargo register in parallel. The verdict: “Committed, and partly audited.” 71% of 2025 volume, 2.84 of 4.00 million tonnes, was verified, and an outside auditor gave limited assurance on that figure; the 29% bought through indirect suppliers is not yet verified.
How the agents got here, step by step: plan 4 s, issuer documents 1 min 52 s, outside sources 1 min 07 s, disagreements 21 s, blind check 18 s.
Every claim sits beside a thumbnail of its page with the cited lines lit. Lena opens page 42 of the Relatório de Fornecedores de Soja 2025: the original on the left, the English on the right, the passage highlighted in both. The Issuer Reader read it at 0.95 and the Checker, reading blind on a second model family, agrees at 0.95. The page also shows where it is already used, and she can flag the translation or download the original file.
“Em 2025, 71% do volume de soja adquirido (2,84 milhões de toneladas de um total de 4,00 milhões) foi verificado como livre de desmatamento e conversão.”
Under the claims, the Contradiction agent lines up every source against every other. The NGO's 58% uses a different denominator: it divides the same 2.84 Mt by 4.89 Mt, which includes 0.89 Mt Marquessa traded for third parties. The CEO's “100% traceability” on the Q2 call is a different measure. A vendor record of November 2024 saying “no time-bound commitment” predates the March 3, 2025 policy, so a challenge to the vendor is drafted. And 2 farms supplying Marquessa's indirect suppliers appear on the federal embargo register, which the company does not mention. Lena decides the 58% with a reason: “Company figure stands; outside figure noted (different denominator).”
“The Contradiction agent explains each one; you decide.”
Claim 4, indirect suppliers at 29% of volume with only 34% of it traced to the farm, comes from scanned table C.3 on page 87 of the Relatório de Sustentabilidade 2025. The Reader read it at 0.86 and the Checker at 0.85, below the 0.90 the house view needs, so Save is blocked until a person confirms. Lena reads the original table, picks “Read the original table; value confirmed” and ticks that she read the page herself, not only the English line.
“I read the original page myself, not only the English line.”
In the conversation beside the answer, Lena asks: “Draft three engagement questions for Thursday.” The Note Writer returns three, each cited: when the 1.16 Mt bought through indirect suppliers will be verified to the farm, and by whom; whether the two embargoed farms were blocked, and when; and, with the 85% bond target missed, what the plan is to reach 100% by 2027. One click sends them to the engagement tracker, and the embargo finding is flagged onto Thursday's agenda for Tomás Ferraz.
Follow-ups answer with citations and act: “Why does the NGO say 58%?” opens the disagreement; “Show only what the auditor covered” highlights the two claims the assurance covers.
Lena saves the answer into research note RN-2026-1187. The Note Writer drafts the summary with every citation kept and proposes the house view move from Medium to Elevated: 71% verified, 29% indirect not verified, the bond target missed with a 25 basis-point step-up, the NGO's 58% explained by a wider denominator, and the embargoed farms. The note already holds two earlier answers, among them how much revenue is exposed to the EU deforestation regulation: 18% of 2025 net revenue, from European customers of soy meal and oil. Her own view sits beside the draft, marked as hers rather than the agent's, and she submits it to Helen Achterberg, due October 21.
“A house-view change cannot go to the checker with a material disagreement open.”
Helen sees the current view beside the proposed one, the 5 citations, each opening its page, and the checks before she signs: citations resolve 5 of 5; every claim at 0.90 or above or confirmed by the analyst; 3 of 4 disagreements decided, explained or flagged; Lena Duarte drafted, Helen Achterberg approves; 7 holding PMs, stewardship and fund ESG oversight will be told. Approve stays locked until she ticks that she read the cited pages and the decided disagreements herself. She approves with “Evidence supports the change”.
When Lena tries to approve her own change: “You drafted this change. The maker of a house-view change can never be its checker.”
The approved view is written to the research management system and the portfolio system's ESG field, and the seven portfolio managers who hold the name are told. On the issuer page Ravi sees the house view now Elevated, approved today at 10:04 a.m., the issuer's own record beside the three vendors' views, and who holds it: $138.2M in his fund, $71.9M in the SFDR Article 8 Global Sustainable Equity Fund and $21.7M of the 5.875% 2031 sustainability-linked notes, whose coupon steps up 25 basis points from November 15, 2026.
Portfolio managers ask and read; they cannot change the house view.
Lena runs one portfolio question over the 64 issuers that buy Brazilian soy, in 23 funds, with the same plan and the same checks: 17 committed and audited, 15 committed with a verified share but no audit, 9 committed with no figure, 14 with no commitment, and 9 for which the agents found no source they could cite. Each row carries its verdict, its key line and its citation, and she can save, send to stewardship, re-run or export in bulk.
“For 9 the agents found no source they could cite, and say so rather than guess.”
The same issuer question, seen by five of the people at Quillfield Capital who ask it, approve it, act on it and test it — what their week looked like, and what it looks like now.
Eight agents read the issuer's own pages and outside sources at once, explain every disagreement and refuse what they cannot cite; an analyst saves, the ESG lead approves.
Turns a plain question into a reading plan: the issuer, the topics, the sources and the period. Asks back when the issuer or the period is unclear.
Reads the issuer's own documents in their original language, scans and tables included, and returns the sentence, the page and an English line.
Reads NGO reports, news, public registers such as embargo, sanctions and labor-inspection lists, regulator filings and the licensed vendor records.
Compares every claim with every other source and explains why they differ: denominator, period, measure, scope, a stale vendor record or a claim no document supports.
Re-reads every cited passage blind on a second model family, scores confidence and removes any statement without a page.
Drafts the research note entry, the proposed house-view change and engagement questions from saved answers, with every citation intact.
Re-runs saved questions when new documents arrive and tells the owner and the holding PMs when an answer changes.
Files new documents from the document store, regulator filings and issuer websites, detects the language and flags poor scans, locked files and duplicates.
One drafts and decides each claim; the other approves the house view. The agents propose; a named person decides.
Beside every answer is a conversation with the agents. Ask about a claim, or tell them what to do next: draft questions, save, compare, watch. Every reply cites its page.
Why does the NGO say 58%?
Different denominator. The NGO counts the 4.89 Mt Marquessa traded in 2025, including 0.89 Mt it originated for third parties. The company counts its own purchases of 4.00 Mt. Both use the same 2.84 Mt verified: 2.84 / 4.89 = 58%; 2.84 / 4.00 = 71%. By the source-precedence rule the audited company figure leads; the NGO figure should be noted.
Show only what the auditor covered
The limited assurance covers one indicator: verified volume from direct suppliers, the 71%. It does not cover the commitment, the indirect 29% or the bond outcome, which the bond's own verifier checked.
Which funds hold Marquessa, and does the Article 8 fund care?
$412.6M across 7 funds; the largest is the Global Consumer Equity Fund ($138.2M). The Global Sustainable Equity Fund ($71.9M) is an SFDR Article 8 fund whose binding element excludes companies with a severe deforestation controversy and no remediation plan. Marquessa has a plan, so this is not a breach today; it was added to that fund's oversight list.
Does Marquessa pay a living wage to farm workers in its soy supply chain?
Not stated in any source. No Marquessa document, outside source or vendor record states a living-wage figure or commitment for farm workers in its soy chain. The supplier code requires the legal minimum wage only, which is not a living wage. The answer lists the eight sources the agents read.
13 screens from the working solution, on its sample data. Pick one to see it large.
How every issuer question is read, checked and approved, set by the administrator; threshold, rule and role changes take effect only when the Head of Sustainable Investing approves them.
| Setting | Default | Choose from |
|---|---|---|
| Show a claim from | 0.80 | Slider from 0.60 to 0.95; below it a claim is not shown at all |
| Enter the house view from | 0.90 | Slider from 0.80 to 0.99; below it a person must confirm the claim against the original page |
| Refuse uncited claims | Always on | Locked on; every statement must carry a page, or it is removed |
| Checker on a second model family | On | Locked on; the Checker reads blind on a different model from the Reader |
| Source precedence | Audited issuer figure → unaudited issuer figure → regulator → NGO → vendor | The Contradiction agent proposes by this order; a person decides |
| Stale vendor records | Older than 12 months | Or older than the issuer's latest document on the topic; flagged, and a challenge drafted |
| Disagreement and approval SLAs | Material 1 business day · minor 3 · house view 2 | Overdue disagreements escalate to the Head; overdue approvals to the CIO office |
| Retention | 7 years | 7 years · 10 years; legal hold suspends deletion |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate (1 to 3 hours by hand per issuer question) · “typical” = published figures (Berg, Kölbel and Rigobon, “Aggregate Confusion: The Divergence of ESG Ratings” — correlations of 0.38 to 0.71 across six providers; SEC ESG enforcement actions, 2022 to 2024 — penalties of $4 million to $19 million). Regulations are named as context only. Quillfield Capital and the people, issuers, funds, NGOs, auditors and documents named on this page are fictional — characters and sample data in the working solution; data vendors are not named.
ESG research is the work of finding out what an issuer actually does on environmental, social and governance matters, from its own reports and filings and from outside sources, before a vote, an engagement, a buy decision or a client report. ESG Research is software for asset managers that does this with eight AI agents: ask about any issuer in plain words and the answer comes back as a verdict with every claim cited to its source page, in the original language beside the English, for an analyst to decide and a second person to approve into the house view.
Ratings from six major providers correlate only 0.38 to 0.71, because they measure, weight and date things differently. ESG Research does not add another score: the Contradiction agent lines up the issuer's own figure against NGO reports, vendor records and registers and says why they differ — a different denominator, period, measure or scope, a stale record or an unsupported claim — and the analyst decides with a reason.
Yes. The Issuer Reader reads the issuer's own documents in their original language, scanned tables included, and shows each cited sentence beside its English line. In the working solution it reads Portuguese supplier and sustainability reports, a Japanese annual securities report and a German climate transition plan, and an analyst can flag any translation for a second reading.
No. Refusing uncited claims is locked on: any statement without a page is removed before the answer is shown, and a question no source answers gets “Not stated in any source” with the list of where the agents looked. The Checker re-reads every cited passage blind on a second model family, and a claim below 0.90 waits for a person.
Every claim that reaches the house view is tied to its document, page and passage, checked by a second model and approved by a second person, and the whole path is on an audit trail kept for 7 years. Compliance testing can export an evidence pack of the answers, original pages and approvals behind any statement, in the context of the SEC Marketing Rule, SFDR Article 8 disclosures and the FCA anti-greenwashing rule.
Yes. A portfolio question runs the same plan and checks across many issuers at once: in the working solution, 64 soy buyers in 23 funds, with a verdict and a citation per issuer and 9 marked “Cannot cite”. Results can be saved to notes, sent to stewardship, re-run or exported in bulk, and saved questions re-run on new documents and alert the owner when an answer changes.
Yes. The analyst confirms, edits or rejects each claim and decides each disagreement with a reason; only the Head of Sustainable Investing approves a house-view change, and the analyst who drafted it can never approve it. The agents never approve a house view, change a holding or send anything outside the firm.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your coverage list, house views, research notes and the systems your documents and holdings live in — and it goes live once every quality gate has passed, including calibration on a golden set of questions your analysts have answered. We will walk you through it on your own issuers first.
Holdings move from estimated to reported ESG data, each figure cited to its page.
1-min demoPortfolio Company ESG Data CollectionEach company confirms prefilled values, and no outlier reaches an LP unexplained.
1-min demoESG Risk MonitoringControversies in holdings scored within hours, and no fund's screen breached unseen.
1-min demoWe'll run ESG Research on a sample of your own issuers, their documents and the questions your team asked last quarter.