Financial statement translation that proves every figure before an analyst reads it
Local filings in English the day they are published, every figure proven to the original.
See one case, screen by screen ↓Financial statement translation turns a company's financial statements and filings — annual and half-year reports, earnings summaries, exchange notices — from the language they were published in into another, with every line and figure faithful to the original. Botminds Financial Statement Translation is financial translation software built for the issuers a fund holds: eight agents pick up each local-language filing within minutes of publication, draft the English line by line with the firm's accounting glossary, prove every figure digit for digit and check each other's work, and a bilingual reviewer approves before analysts read it.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Financial statement translation is the quiet bottleneck in global research. A global fund holds hundreds of issuers that report first, or only, in their own language: German half-year reports, Japanese earnings summaries (kessan tanshin), Korean half-year reports, Brazilian quarterly forms, Indonesian interim statements. Tokyo has required Prime Market companies to publish key disclosures in English at the same time as Japanese only since April 2025; Korea's English disclosure duty reaches companies with KRW 2 trillion or more in assets in 2026; most markets have none. So financial report translation for the largest positions goes to an agency and waits days, and the long tail is pasted into a generic translator or skimmed.
The risk is in the numbers nobody checks. A generic translator can leave 1.284,6 in German notation, miss a scale header (Tsd. €, 百万円, 万元) and render Wertberichtigung, an allowance, as a write-up. One wrong scale header is a thousandfold error in a model. The same term is rendered three ways over three years, so comparisons drift, and when a figure turns out to be wrong, research compliance cannot say who translated it or which analysts relied on it.
Estimated split for a priority half-year report, by agency and by hand against the solution; the agency's 6–10 business days are published vendor figures.
Eight agents draft the English line by line, prove every figure and check each other; a bilingual reviewer decides only what they could not settle.
Brennwald Chemie AG, a German specialty chemicals group held in 14 Quillfield funds ($612 million, 1.8% of the European Value Fund), publishes its 48-page Halbjahresfinanzbericht at 06:00 London time. Hannah Osei's results call is at 14:00. Here is what happens next, screen by screen, in the working solution.
The exchange filing feed carries the report at 06:00:00; at 06:01:48 the Intake agent matches DE000BRWC264 to 14 funds and $612 million, sets priority P1 and starts a 4-business-hour clock: due 10:00. The Page reader reads 48 pages (one scanned, lowest line confidence 0.93): 1,184 lines, 38 tables, 412 figures. The Translator drafts line by line with German glossary v14 and the FY2025 annual report approved in March as memory; the Terminology agent applies 11 glossary decisions, the Number Guard proves 411 of 412 figures, and the Back-translation agent translates the English back with a different model and finds 2 lines drifting. At 06:17 they hand Kenji Morita 3 things to decide.
“3 things to decide, each with a proposal, the evidence and the agents' confidence.”
Kenji reviews German and Japanese from London. His Today screen shows 4 items waiting, 41 of 64 filings already released in English and a median of 2 h 47 m from publication to English. Every finding since midnight cites its page, and he can ask the agents in plain words what could miss its deadline.
“The Number Guard stopped a 1,000× scale error in Brennwald's related-party note: the table is in thousands.”
The statement opens interlinear, with side-by-side and English-only views a click away. Every figure is locked. He hovers pension provisions: 1.284,6 becomes 1,284.6, and the Number Guard shows its proof — digits 1-2-8-4-6 match, the decimal comma becomes a point, Mio. € becomes € million, the sign is positive. Click any figure and page 14 opens with the passage highlighted.
The rail says it plainly: “3 things need you before release. The agents settled the rest: 412 figures checked, 11 terms applied.”
In note 14 the draft says “write-ups” on trade receivables. Translated back it reads Zuschreibungen, the opposite of Wertberichtigungen: drift 0.71 against a 0.25 threshold. The agents propose “Changes in valuation allowances on trade receivables” at 97% confidence. Kenji asks why, and the agent answers with IFRS 9, the FY2025 report he approved and the movement table on page 31.
“Under IFRS 9 the allowance for expected credit losses reduces the carrying amount of receivables; a write-up (Zuschreibung) increases an asset's value.”
He accepts the fix, and the Number Guard re-checks: figures unchanged. Next, Bilanzgewinn on page 9 has no entry in glossary v14. The Terminology agent offers three candidates with their history: “net retained profits” (25 of 37 past uses, 82% confidence), “distributable profit” and “balance sheet profit” (literal, avoid). He picks the first and proposes it to the German glossary for Sabine Krüger to approve.
The related-party table on page 38 is in Tsd. €, but the draft header says € million: sales to joint ventures would read €14.9 billion instead of €14.9 million, almost four times group revenue. He accepts “€ thousand”. With all three settled he approves: 412 of 412 figures proven, no line drifting over 0.25, reason code R1, an attestation, signed with single sign-on.
“I have reviewed the exceptions and the glossary decisions; the English is faithful to the German original.”
The Release agent publishes the English to the research management system with its page anchors, files the original, English and evidence in the document store for 7 years, and tells Hannah Osei, five other analysts and 3 portfolio managers. 412 figures proven, 3 exceptions settled, 3 h 52 m from publication to English.
Hannah reads the approved English first, with the original one click away. She asks why pension provisions fell and gets the answer with its pages: a discount rate of 4.0% against 3.6%. She exports the balance sheet to her model with a source link per cell, four hours before her call.
“Only approved English appears here.”
Kenji's proposal waits for Sabine Krüger; he cannot approve his own change. The impact is shown: Bilanzgewinn appears in 37 released translations from 9 issuers since 2024, and 12 used “balance sheet profit”. When she approves, the agents re-check those translations and list them for re-issue, which needs two people, and every analyst who read an affected version is told.
Hoshimura Denki published its earnings summary in Japanese and English at the same time, as Tokyo has required of Prime companies since April 2025. The Gap checker compares 286 figures and 398 sentences and finds 3 differences: operating profit ¥8,421 million in the English against 8,412百万円 in the Japanese, an omitted exchange-rate assumption, and “decreased slightly” for a 12.4% fall. The recommended choice: analysts read Quillfield's own English, every figure proven, with a note on the differences.
“The agents never contact an issuer.” A question to investor relations is sent only after the Head of Research Operations approves it.
The same morning of filings, seen by five of the people who carry it — what results season looked like, and what it looks like now.
Eight agents draft the English line by line, prove every figure and check each other; a bilingual reviewer decides only what they could not settle.
Watches 14 exchange filing feeds in 13 markets and the issuer mailbox, matches each filing to holdings in portfolio accounting, classifies its type, period, language and standard, sets priority from position size and watchlists, and starts the clock at publication.
Reads every page, including tables and scanned pages, with a confidence for each line, and anchors every line to its page and position.
Drafts the English line by line, keeping the table structure, using the firm's glossary and the issuer's approved past translations as memory.
Applies the firm's accounting glossary per language and standard, lists the term decisions for each document, and asks the reviewer when a term has no entry, with candidates and their past use.
Proves every English figure equals the original digit for digit through separators, scales (Tsd., Mio., Mrd., 百万円, 億円, 万元, 억원, R$ mil), signs (−, △, brackets) and percentages, and re-adds totals in both languages.
Translates the English back with a different model, compares the meaning line by line and flags drift above the threshold, 0.25 by default.
Compares the issuer's own English with the original for figures, omissions and tone, so analysts know when the official English cannot be trusted.
After approval, publishes the English to the research management system and document store, tells the analysts and portfolio managers who cover or hold the issuer, and answers their questions from the approved English with page citations.
Approves with a reason code. The agents propose; a named person decides.
Reviewers ask the agents about a report; analysts ask the approved English. Every answer names the agent and cites the original page.
What could miss its deadline today?
Two filings are inside 30 minutes: Brennwald Chemie AG (yours, 3 to decide, due 10:00) and PT Sinar Kencana Agro (Dewi Halim, 3 scanned lines, due 10:00). I escalated Sinar Kencana to the Indonesian lead at 75% of its clock.
Why is p.38 in thousands?
Only the related-party table is in thousands; its header reads “in Tsd. €”. Brennwald has shown this note in thousands since its FY2024 report.
Why did pension provisions fall?
Pension provisions fell to €1,284.6 million from €1,341.2 million, mainly because the discount rate rose to 4.0% from 3.6%; the €52.8 million actuarial gain went to other comprehensive income.
How much covenant headroom is there?
Net debt / adjusted EBITDA is 2.85x against a 3.50x limit (nota 17, p.41). On our arithmetic EBITDA could fall about 19% before the limit. Net debt includes R$600,000 thousand of lease liabilities.
13 screens from the working solution, on its sample data. Pick one to see it large.
The checks, clocks and coverage are settings. Threshold changes need a second approver, and every change is on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Number Guard | 100% of figures proven | Fixed · no override |
| Back-translation drift threshold | 0.25 | Lines above it go to the reviewer |
| Page reading confidence floor | 0.85 | Lines below it go to a person |
| Bulk approval | Up to 5 pages | Only with no exceptions, 100% of figures proven and no reviewer edits |
| Turnaround from publication | P1 results: 4 business hours | P1 ad hoc notices 1 business hour · P2 results 1 business day · annual reports 3 business days · P3 5 business days |
| Priority P1 | Held at 0.5% or more of any fund | Or on a watchlist in the order management system |
| Languages | 12 of 14 on | German, Japanese, Chinese (Simplified and Traditional), Korean, Portuguese, Spanish, French, Italian, Dutch, Swedish, Indonesian · Turkish and Polish off |
| Retention and regions | 7 years | EU issuers in the EU region · Asia-Pacific in Singapore · Americas in the US · legal hold |
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 (translation vendors' published turnaround of 6–10 business days per language pair for an annual report, directional; JPX on same-day English for Tokyo Prime Market companies from April 1, 2025; FSC Korea and KRX on English disclosure for companies with KRW 10 trillion or more in assets from 2024 and KRW 2 trillion or more in 2026; arXiv 2107.08357 on machine translation of separators and scale words). Quillfield Capital and the people, issuers, funds and identifiers named on this page are fictional — characters and sample data in the working solution.
Financial statement translation turns a company's financial statements and filings — annual and half-year reports, earnings summaries, exchange notices — from the language they were published in into another, with every line and figure faithful to the original. Botminds Financial Statement Translation is financial translation software built for the issuers a fund holds: eight agents pick up each local-language filing within minutes of publication, draft the English line by line with the firm's accounting glossary, prove every figure digit for digit and check each other's work, and a bilingual reviewer approves before analysts read it.
The Number Guard proves every English figure equals the original digit for digit: decimal and thousands separators (1.284,6 becomes 1,284.6), scale headers such as Tsd. €, 百万円, 万元 and 억원, and signs in brackets or △. It re-adds totals in both languages. 100% must be proven before release, with no override; in the demo it stopped a table in thousands from going out as € million.
No. It translates the language, not the currency. Figures stay in the issuer's own currency and scale — € million, R$ thousand, SEK million — and are proven equal to the original, so an analyst's model receives the numbers exactly as published.
Back translation translates the English back into the original language and compares the meaning, which catches errors that read fluently. Here a second agent does it on a different model from the Translator and flags any line drifting above 0.25. In the demo it caught “write-ups” for Wertberichtigungen, which is an allowance — the opposite meaning.
Translation vendors publish 6–10 business days per language pair for an annual report. Here the clock starts at publication, with targets of 4 business hours for priority results and 3 business days for annual reports. In the demo a 48-page German half-year report reached analysts in 3 h 52 m, and the day's median was 2 h 47 m.
Yes. When an issuer publishes English, as Tokyo has required of Prime Market companies since April 2025, the Gap checker compares it with the original for figures, omissions and tone. Analysts are told when the official English differs, and any question to the issuer is sent only after the Head of Research Operations approves it.
Yes. The agents never approve or release a translation, change a figure, approve a glossary change or contact an issuer. A bilingual reviewer approves each release with a reason code and an attestation, glossary changes and corrections need two people, and every step is on an audit trail kept 7 years.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your glossaries, approved past translations, languages, sources, holdings and turnaround targets — and it goes live once every quality gate has passed. First-time setup takes six steps and ends with a trial on 20 of your past filings.
Every call read the day it reports, every claim tied to its transcript line.
1-min demoEquity Research AutomationEvery 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 demoWe'll run Financial Statement Translation on a trial of 20 of your own past local-language filings.