Earnings call analysis that tests every call against your thesis and ties every claim to its transcript line
Every call read the day it reports, every claim tied to its transcript line.
See one case, screen by screen ↓Earnings Call Analysis is AI earnings call analysis for buy-side analysts: it turns each earnings call into a cited thesis-impact note, in any language. Seven agents listen to every covered call, tag guidance live, compare it with last quarter and the analyst's model, and look for the case against before any thesis pillar moves. The analyst decides each move, a second analyst confirms any Broken pillar, and only signed notes reach portfolio managers, with every claim tied to its transcript line.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Earnings call analysis on the buy side is a race against the calendar. In the peak weeks of earnings season an analyst covering 30 to 60 names gets several overlapping calls a day, some held in Japanese, Korean or German with no English transcript for hours. Before each call there is last quarter's guidance, consensus and the model to pull up; after it, the transcript to re-read, the model and the variance table to update, and a note to write before the portfolio managers' morning meeting.
Most earnings call analysis tools stop at an AI earnings call summary: what was said, not what it means for the investment thesis. And an analyst reading alone tends to read the call to confirm it. Notes reach portfolio managers with claims nobody can trace to a line, review of a rating or target change is often informal, and a one-on-one at a broker conference can carry material nonpublic information into a note. With research budgets flat while assets grow, every extra name has to come from reading faster, not from spending more.
Estimated split for one covered company's quarterly call, by hand and with the solution. The post-call read-and-write cycle alone runs about one to two hours a name.
Seven agents listen, translate, compare and challenge every call; the analyst decides, a second analyst confirms, the analyst signs.
Quillfield Capital is a fictional global asset manager. On Thursday, October 29, 2026, Hoshida Robotics — a Tokyo Prime name held in five Quillfield funds for $423M — holds its Q2 FY3/27 results call in Japanese, 58 minutes and 214 lines. Here is what happens next, screen by screen, in the working solution: the analyst asks the agents in plain words, sees the line behind every answer, and decides.
Dana covers 38 names. Her Today screen shows four calls — Japanese, Korean, German and English — three things waiting for her, each with one next step, and 20 of 38 names signed this season. The agents have already read the Hoshida call and list what they did, each line cited; under the list she can ask them about her coverage, or tell them what to do, in plain words.
Needs you: “Hoshida Robotics: the agents propose ‘China demand’ → Weakened. Backlog fell 12% to ¥412.6B, 9.3% under your model.”
Intake joined the call on the firm's recording line, and the Listener tagged 14 guidance and key statements in 214 lines as it ran, in Japanese, with a tone ribbon that marks where the call turned cautious. The Translator put English beside every line and checked 23 figures in both languages. On the right are Dana's four thesis pillars, each with a needle tied by a thin line to the transcript lines that moved it.
“China demand recovers in H2 FY3/27” · Confirmed → Weakened · 3 lines moved it · 2 counterweight · 86% confidence.
On the call screen, Dana asks the agents in plain words. The Listener answers with eight lines, each with its timestamp — orders down 27%, backlog down 12%, the EV question, customers cautious, price competition, the Q4 recovery, two projects in final specification, and North America as context — and filters the transcript to them.
The agents answer with the line behind every point, and act on the screen when asked.
Dana clicks the backlog timestamp in the answer and the CFO's line opens: the Japanese beside the English, with Play from 00:31:12. When the vendor transcript arrived at 4:41 p.m., the Translator reconciled it with the firm's recording and logged the 2 lines it corrected.
“Translation check: every figure is the same in both languages (12%, ¥412.6 billion).”
The Comparer checked 7 statements against Q1 guidance and Dana's model and found 4 differences over the threshold. Operating margin guidance was cut from 14.5% to 13.2%, the 15% target moved from FY3/28 to FY3/29, China orders fell 27% against −10% in the model, and backlog of ¥412.6B is 9.3% under the model's ¥455.0B. It proposes six cell changes, v14 to v15, each with the line behind it.
The Comparer proposes and Dana approves: nothing is written to the model until she does, and v14 is kept.
Before any pillar moves, the Skeptic must find the best case against the move in the same call. It searched 214 lines and found two: the CEO expects China orders to recover in Q4 as large EV projects restart, with two projects in final specification. So the agents stop “China demand” at Weakened, not Broken. Dana accepts Weakened; she could choose another state or reject with a reason code.
Skeptic agent: “Your rule says it breaks if China orders stay below −20% for two quarters or backlog falls under ¥400B.”
The CFO cut margin guidance and moved the 15% target a year later. Dana marked “Cobot mix lifts operating margin to 15% by FY3/28” Broken with reason code PB-01. Tomás Aguirre, senior analyst in London, confirmed it with CF-01 and reviewed the target price change from ¥8,400 to ¥7,600.
“Maker-checker complete. Marked by Dana Whitlock, confirmed by Tomás Aguirre. The same person can never do both.”
The Note writer updates NOTE-2026-10-4471 to version 3: China recovery pushed out, 15% margin a year later, Overweight kept, conviction High to Medium, target ¥7,600. All 14 claims cite a call line, the release or the model, and Dana's own view is labeled as hers. Before you sign shows 6 of 6 checks passed, including the compliance pre-check.
A sentence added without a source is flagged, and signing waits until it has one or is marked as the analyst's view.
Dana attests that she read the cited line behind every claim, that every source is public with no material nonpublic information, and that her views are labeled. The Compliance agent confirms Hoshida is not on the restricted list and every source is public, then sends the note to Helena Vos, Rafael Duarte, the 11 Asia Industrials readers and the research management system as RMS-118842.
“Your signature releases version 3 to the readers you chose. It is recorded with the time, your attestation and every source behind the note.”
Helena runs the Quillfield Japan Equity Fund, where Hoshida is 2.1%. Before the open she sees the Hoshida note marked New, with each pillar's state, signed by Dana Whitlock at 6:42 p.m. Singapore time and reviewed by Tomás Aguirre. Portfolio managers only ever see signed notes; drafts stay locked.
“Signed by Dana Whitlock · Today 6:42 p.m. SGT · 6:42 a.m. ET · reviewed by Tomás Aguirre.”
The same earnings call, seen by five of the people who carry it — what earnings season looked like, and what it looks like now.
Seven agents listen, translate, compare and challenge every call; the analyst decides, a second analyst confirms, the analyst signs.
Matches every call on the earnings calendar to the coverage list, joins the firm's recording line and attaches the vendor transcript when it lands.
Tags every guidance and key statement as the call runs, in the language spoken — metric, value, direction, period, speaker and timestamp — and scores the tone of each line: the earnings call sentiment analysis behind the tone ribbon.
Puts English beside every original line, checks that every figure is the same in both languages, and reconciles the recording with the vendor transcript, logging each correction.
Holds each statement against last quarter's guidance and the analyst's own model, and proposes model changes cell by cell for approval.
Before any pillar moves to Weakened or Broken, searches the same call for the strongest counter-evidence, weighs it and states the pillar's break rule.
Drafts the thesis-impact note from the decided pillars, the guidance changes and the model changes, cites every factual claim and labels the analyst's views.
Checks every note against the restricted and watch lists and research blackouts, checks every source is public, and distributes the note after signature.
Decide, confirm, sign, clear. The agents propose; a named person decides.
Every main screen has the agents close at hand. Ask about a call, a pillar, a note or a setting in plain words — the named agent answers with the line, release or model cell behind it, and acts only with your approval.
Why is the Hoshida China pillar weakening?
The CFO said order backlog fell 12% quarter on quarter to ¥412.6B, 9.3% under your model, and China orders fell 27%. The Skeptic found the CEO's Q4 recovery comment, so the agents propose Weakened, not Broken. Opening the call for your decision.
Show me every China comment
Eight lines touch China: orders −27%, backlog −12%, the EV question, customers cautious, price competition, the Q4 recovery, two projects in final specification and North America as context. I have filtered the transcript to them.
How does the backlog compare with my model?
Backlog ended Q2 at ¥412.6B, down 12% from ¥468.9B and 9.3% under your model's ¥455.0B. I can propose the model update on the Guidance tab; I never write to your model without your approval.
Draft a one-line update for the morning meeting
“Hoshida Robotics: China recovery pushed out (backlog −12% q/q to ¥412.6B) and the 15% margin target moved a year; Overweight kept, conviction lowered, target ¥7,600.” It is a draft: it goes nowhere until you sign the note.
13 screens from the working solution, on its sample data. Pick one to see it large.
How strict the Skeptic is, when a line counts as unclear, the clocks and how long records are kept are settings — versioned, logged and approved by the Director of Research.
| Setting | Default | Choose from |
|---|---|---|
| Skeptic strictness | Strict | Standard · Strict · Very strict |
| Confidence below which a line is “unclear” | 80% | A slider from 60% to 95%; lines under it wait for a person |
| Listen live on the recording line | On | On · Off |
| Read the firm's recording first when the vendor is late | On, after 30 minutes | On · Off |
| Note signed after the call | Within 24 hours, then escalates to the Director | 2 · 4 · 8 · 12 · 24 · 48 hours |
| Second analyst confirms a Broken pillar | Within 4 hours | 2 · 4 · 8 · 12 · 24 · 48 hours |
| Keep notes, citations and the audit trail | 7 years | 5 · 7 · 10 years |
| Where notes and recordings are stored | By office: US East, UK South, Southeast Asia | By office (recommended) · All in US East |
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 of the 44 largest asset managers, H1 2025; Correlation One, 2025, on buy-side analysts' coverage and call loads; Tokyo Stock Exchange English disclosure survey, 2025). Quillfield Capital, the people, issuers, brokers and identifiers named on this page are fictional — characters and sample data in the working solution.
Earnings Call Analysis is AI earnings call analysis for buy-side analysts: it turns each earnings call into a cited thesis-impact note, in any language. Seven agents listen to every covered call, tag guidance live, compare it with last quarter and the analyst's model, and look for the case against before any thesis pillar moves. The analyst decides each move, a second analyst confirms any Broken pillar, and only signed notes reach portfolio managers, with every claim tied to its transcript line.
Most AI equity research tools stop at a summary of what was said. Earnings Call Analysis tests the call against the analyst's own investment thesis, pillar by pillar, each with its confirm and break rules and the model cells behind it, and shows which lines moved which pillar. A Skeptic agent argues the other side before anything moves, and the result is a note the analyst signs, not a panel to read.
Yes. The Listener reads the call on the firm's recording line in the language spoken, and the Translator puts English beside every line and checks that every figure is the same in both. When the vendor transcript arrives, it is reconciled with the recording and each correction is logged, so earnings transcript analysis starts while the call runs, not when the English transcript lands hours later. In the working solution, 143 of the 350 calls read this season were in a language other than English.
Yes. The Comparer holds every guidance statement against the prior quarters and the analyst's model, cell by cell, and every value links to its line, release or cell. It proposes model changes — six cells from v14 to v15 on the Hoshida call — and nothing is written until the analyst approves; the earlier version is kept.
Before release, the Compliance agent checks the restricted and watch lists, research blackouts and whether every source is public. A note citing a private meeting or expert call — such as a CFO's comment at a broker conference, a possible selective disclosure under Regulation FD — is held for Research Compliance, who releases it, removes the paragraph or blocks it, with a reason code. The analyst also attests at signing that no material nonpublic information was used.
Yes. The agents propose and never decide: the analyst accepts, changes or rejects each pillar move with a reason code, a Broken pillar needs a second analyst who is never its author, and rating and target changes need a senior analyst. Only the covering analyst signs, and the model is written only after approval.
Every read, proposal, decision, signature and send, by people and agents, with who, what, when, why and the evidence, plus agent versions and prompt hashes. It is hash-chained, kept seven years by default — Rule 204-2 asks for at least five — and exports for internal audit, an SEC or FCA request, or client due diligence.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your coverage list, thesis pillars, note template and a sample of past notes — and it goes live once every quality gate has passed. First-time setup is a guided conversation with the agents in six steps, each approved before the next, and nothing goes live until the last one. We will walk you through it on your own calls first.
Every result read, modeled and noted before the open, every figure tied to its page.
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 Earnings Call Analysis on your own past calls, thesis pillars and notes.