Equity research automation that reads, models and notes every earnings release before the open, every figure tied to its page
Every result read, modeled and noted before the open, every figure tied to its page.
See one case, screen by screen ↓Equity research automation is software that takes the routine work of earnings season off a buy-side research team: catching each release, updating the analyst's model, checking the numbers and drafting the earnings note. In Equity Research Automation a crew of seven AI agents does this for every name in the coverage universe, in the analyst's own Excel model and the issuer's own language, and reports in a conversation on each company: what it caught, wrote, tied and drafted, every statement cited to its page. The analyst asks the crew questions in plain words, confirms or rejects what it proposes, approves estimate changes and signs; the Director of Research checks and releases the note to portfolio managers. It is AI for equity research in which the agents propose and people decide.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Buy-side equity research runs on the issuers' calendar: every quarter a team reads several hundred earnings releases in about five weeks, often four or five of one analyst's names on a single morning. For each one the analyst or associate catches the release, keys the numbers into the analyst's own Excel model, checks totals and history, compares the result with the house estimate, consensus estimates and the investment thesis, and writes an earnings note for the portfolio managers, ideally before the open and before the call. The note that matters most is the one most often late.
The costly errors are silent. A prior year recast after a divestiture makes growth read +0.8% instead of +4.6%; a segment table does not add up to the total; a figure in thousands is read as millions; a release published only in Japanese waits for a translation; a name on the restricted list gets a note. Research budgets are flat, so covering more names has to come from more reading per analyst hour, with the MNPI controls, the restricted list and the research records kept intact.
Estimated split for one analyst covering 38 names over a five-week season, by hand and with the solution. The analyst's time with the solution goes on decisions: confirming, approving, editing and signing.
Seven agents catch, read, re-tie, test and draft every release in the season; the analyst signs and the Director releases.
Thursday, November 5, 2026, week 4 of the Q3 season at Quillfield Capital, a global asset manager with research in New York, London and Singapore. Fourteen companies report at 7:00 a.m. ET; four of them are Priya Raman's. Here is how Marlowe Brands goes from the wire to a signed, checked earnings note in 29 minutes, in the working solution: the agents report in one conversation, Priya asks and decides, and every figure links to its page.
The season board shows Quillfield's whole coverage universe: 486 names, 14 analysts, 412 names signed so far. The Release Watcher, watching the newswire, EDGAR, exchange feeds and IR pages, catches 13 of the 14 releases due at 7:00 in a median of 7 seconds; Halden Industrial's IR page now says 7:30. Each row's status flips letter by letter as the crew works: Reading, Model refreshed, Ties checked, Note drafted.
Needs a person: “Kestrel held: tie failed · Halden delayed · Corvane restricted · 4 notes drafted, to sign.”
Priya covers US Consumer Staples, 38 names, and her calls start at 8:30. Her Today screen shows 5 names reporting today, 2 notes waiting for her, 31 of 38 names signed this season and her median of 1 h 41 m from release to signed note against a 2-hour target. Under “Needs you”: sign Marlowe Brands; Tessaro Foods withdrew guidance and the draft proposes Neutral → Underweight, which Nadia Kerr must co-sign; Gristmill Beverage is still being tied. Every line of what the agents did links to its source.
System strip: “Consensus feed 6 min behind · using the 06:45 snapshot.”
She opens Marlowe Brands (NYSE: MRLW) and reads the crew's conversation. The Watcher caught the Q3 FY2026 release 7 seconds after it posted and matched the 8-K (Item 2.02) on EDGAR at 07:01:40. The Extractor read the 14-page release and the 22-slide deck and wrote 64 values into her own model, MRLW_Model.xlsx, as draft v39: actuals only, forecast cells untouched. The Tie-out Reviewer re-read every value with a different AI model: 61 of 64 tie, 3 do not.
Tie-out Reviewer: “I will not let a note be drafted until these are explained.”
The Reviewer finds page 6, “Recast of prior-period results”: Pinecrest Pet was sold on September 30 and is now discontinued operations. Q3 FY25 net sales move from $3,188M in her model to $3,073M. It rebases 8 quarters, 24 cells, each cited to the recast table, and by 07:11 all 64 values tie. Model changes lays out the earnings model update: 64 actuals written, 24 history cells rebased, 5 estimates proposed, 0 forecast cells touched.
“On your old figures growth would read +0.8%; on the recast base it is +4.6%.”
Adjusted EPS is $1.42 against Quillfield's $1.31 (+8.4%) and consensus of $1.33; net sales are $3,214M against the house $3,180M. Gross margin of 41.8% is 120 bps short of her 43% thesis. The bridge sizes the $0.11 beat: SG&A +$0.13, tax rate +$0.05, volume and price +$0.02 and other +$0.01, less gross margin −$0.10. Three of her four thesis claims are on track; claim 2, “Gross margin back to 43% in Q3 FY2026 as resin costs ease”, is at risk.
Every row links to its page; consensus is labeled as the 06:45 snapshot while the feed runs behind.
The Note Drafter writes 7 sentences with 11 citations and proposes FY26 EPS $5.22 → $5.34 and FY27 $5.70 → $5.74 on guidance raised to $5.30–$5.40. The Compliance Sentinel clears the draft: every source public or licensed, MRLW not on the restricted or watch list, no expert-call content. Priya asks how much of the beat came from tax. The Variance agent answers from page 9: $0.05 of the $0.11, so adjusted EPS would be about $1.37 without it, still above the house $1.31.
Note Drafter: “Rating and price target unchanged; that is your call.”
In the Decisions panel she confirms the rebase, approves the five estimate changes into model v39 and moves the margin claim to Q2 FY27, and the note updates to say so. Then she signs: an attestation and a reason code. The signature is logged at 07:26, 26 minutes after the release posted.
“I have read the note and its sources; the views are my own and I hold no position in MRLW.”
Nadia sees the note signed by Priya at 07:26, ties 64 of 64, compliance clear, no rating or target change. She releases it to the portfolio managers with the reason code “Checked: ties, thesis, sources”. The checker can never be the maker, and a rating change or a price-target change over 10% would also need her co-sign.
On her screen: “Signed by Priya Raman at 07:26 · ties 64 of 64 · compliance clear · no rating or target change.”
Tom's morning digest covers his holdings in Quillfield US Core Equity Fund. Marlowe Brands, 1.2% of the fund: beat on costs and tax, margin thesis moves to Q2 FY27, Overweight and the $96 target unchanged, FY26 EPS $5.22 → $5.34, signed by Priya Raman and checked by Nadia Kerr at 07:29. Kestrel Semiconductor, 2.3% of the fund, shows as held: no note until the analyst decides.
Portfolio managers see research only once the analyst signs and a checker releases it.
The audit trail holds the whole morning, from the release caught at 07:00 to Nadia's release at 07:29: who, what, when, why and the evidence for every agent step and both signatures. It is hash-chained, kept 7 years against the 5 that SEC Rule 204-2 requires, and holds 18,406 events this season. An export pack gathers a note's sources, model changes, signatures and compliance checks for auditors and examiners.
“Chain verified.”
The same earnings morning, seen by five of the people who carry it: what earnings season looked like, and what it looks like now.
Seven agents catch, read, re-tie, test and draft every release in the season; the analyst signs and the Director releases.
Knows the report calendar and watches the newswire, EDGAR (8-K Item 2.02, 6-K, 10-Q), exchange disclosure feeds and 486 IR pages. It catches each release within seconds, confirms it is final, and flags a delay or a miss with what it checked.
Reads the release, deck and filing in the original language and writes reported actuals into the analyst's own model as a new version, every cell cited to its page.
Independently re-reads every value with a different AI model, re-adds segments and subtotals, compares history with the model and finds recast tables. When the issuer's recast explains a difference, it rebases the history, cites the page and asks the analyst to confirm.
Compares actuals and guidance with the house estimate, consensus estimates and each thesis claim, marks every claim on track, at risk or broken with the evidence, and sizes what drove the beat or miss.
Writes the earnings note in the house template with every sentence cited, proposes estimate and thesis changes, and asks the analyst what it cannot decide.
Checks every note before release: restricted and watch lists, public sources only, expert-call content cleared, disclosures present.
Builds the season calendar and each analyst's load, and proposes who takes which names when an analyst is out or a morning overloads.
Maker, checker, restricted names. The agents propose; a named person decides.
Analysts ask about a release, a tie or a thesis claim in plain words, or give the agents a task. Answers cite the page they came from; when the answer is not in the sources, the crew says so instead of guessing.
How much of the beat came from tax?
$0.05 of the $0.11 beat came from the lower tax rate: 21.4% this quarter against 23.0% in your model. Excluding it, adjusted EPS would be about $1.37, still $0.06 above our $1.31.
Why did gross margin miss?
Gross margin was 41.8%, down 80 bps on the recast prior year; management blames resin and packaging costs and promotion. Your claim of 43% in Q3 is at risk. I would ask about Q4 resin contracts on the 8:30 call.
Why is Kestrel held?
Kestrel's end-market lines add to $1,212M but the total says $1,216M; last quarter a $3M “Other” line closed the gap. I cannot tell which is right from the release, so the note is held. The 10-Q is due Friday; Ben can wait for it or accept with the difference noted.
Add a question for the 8:30 call
Added to your 8:30 call list: “Q4 resin contract pricing, and when does promotion normalize?” I will read the transcript when it posts and draft an addendum if the answer moves the margin claim.
13 screens from the working solution, on its sample data. Pick one to see it large.
Tie tolerances, variance alerts, service levels and the sources a note may cite are settings. Every change is proposed by the administrator, tested on past releases first and approved by the Director of Research.
| Setting | Default | Choose from |
|---|---|---|
| Tie tolerance, reported values | Exact to the release's rounding | A value that does not tie holds the note |
| Tie tolerance, derived ratios | ±0.05% | ±0.03% proposed by the administrator, waiting for the Director |
| Extraction confidence for straight-through | 97% | Below it, a value goes to the research associate |
| Variance alerts | EPS ±5% · revenue ±2% · margins ±50 bps | Set per line; changes approved by the Director |
| Price-target change needing the Director | Over 10% | Rating changes always need the Director's co-sign |
| Target: release to signed note (pre-open) | 2 hours | Each note in the sign-off queue shows the time left to the target |
| Escalate an unsigned note to the Director | After 4 hours | Signing can then be handed to a regional head |
| Sources a note may cite | Public filings, releases, licensed data | No expert-call content without clearance |
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 · “typical” = published figures (Substantive Research, research budgets at the 44 largest managers in H1 2025, via Hedge Fund Alpha; a model-update vendor's claim of 15–45 minutes saved per company update). Rules named for context: SEC Form 8-K Item 2.02, Advisers Act Section 204A and SEC Rule 204-2. People, firms, funds and companies named on this page are fictional: characters and sample data in the working solution.
Equity research automation is software that takes the routine work of earnings season off a buy-side research team: catching each release, updating the analyst's model, checking the numbers and drafting the earnings note. In Equity Research Automation a crew of seven AI agents does this for every name in the coverage universe, in the analyst's own Excel model and the issuer's own language, and reports in a conversation on each company: what it caught, wrote, tied and drafted, every statement cited to its page. The analyst asks the crew questions in plain words, confirms or rejects what it proposes, approves estimate changes and signs; the Director of Research checks and releases the note to portfolio managers. It is AI for equity research in which the agents propose and people decide.
The Extractor reads the release, deck and filing and writes reported actuals into the analyst's own Excel model as a new version, every cell cited to its page. It never touches forecast cells; estimate changes are proposed in a separate block for the analyst to approve, edit or reject. Values read below 97% confidence go to the research associate instead of being guessed.
An independent Tie-out Reviewer re-reads every value with a different AI model, re-adds segments and subtotals, and compares history with the model. When the issuer recasts prior periods, as Marlowe Brands did on page 6 after a divestiture, it rebases the history to the recast table with the page cited and asks the analyst to confirm. A difference it cannot explain holds the note.
It drafts one. The Note Drafter writes the note in the house template with every sentence cited and proposes estimate and thesis changes, but it never changes a rating or price target and never sends a note. The analyst edits and signs it, and a second person checks and releases it to portfolio managers.
To support the firm's MNPI compliance, the Compliance Sentinel checks every note before release: the restricted and watch lists from the compliance system, public sources only, expert-call content cleared and disclosures present. A restricted name is blocked before drafting, and only Research Compliance can clear it. Every check is on the audit trail.
Yes. Releases are read in the original language, such as a Japanese kessan tanshin, a German quarterly statement or a Brazilian release, and shown side by side with the English. Each English figure is checked against the original and cited to its page.
No. In this kind of automated equity research the agents catch, read, tie, test and draft; the analyst decides. Estimate changes, thesis claims and the signature belong to the analyst, the Director of Research or a regional head releases every note, and rating changes need the Director's co-sign. The design target is 20–30% more names per analyst on the same research budget.
The Agentic Solution Engine builds and deploys it from your requirements and documents: your coverage universe, your analysts' model templates, your note template and your rules. It goes live once every quality gate has passed. We will walk you through it on last season's releases for your own covered names first.
Every call read the day it reports, every claim tied to its transcript line.
1-min demoFinancial Data ExtractionEvery covered name's model refreshed the day it files, every number tied to its page.
1-min demoFinancial Statement TranslationLocal filings in English the day they are published, every figure proven to the original.
1-min demoWe'll run Equity Research Automation on last season's releases for your own covered names and your own analysts' models.