Confidence by evidence
Five levels, derived from the kind and clarity of the source and from whether the documents agree. Your team checks the uncertain values first.
A thousand pages in. Every value out, with its source.
Klartext Extract reads annual reports, contracts and other PDFs cover to cover and extracts exactly the fields you need. Your team reviews instead of retyping.
- EU-local
- Built for 1,000+ page documents
- Pricing on request
For credit and rating teams
Financial Spreading in minutes, not days.
Financial Spreading turns financial statements into one standard template, the basis for credit decisions, ratings and peer comparisons. Klartext Extract maps every line of the balance sheet, income statement and notes to your template, across years and with its source.
In the statementYour template20242023
- Revenue from sales of goods and servicesp. 87Revenue184.6171.2
- Cost of materials and purchased servicesp. 87Cost of materials92.387.9
- Wages, salaries and social securityp. 87Personnel expenses48.144.6
- Depreciation of property, plant and equipmentp. 87Depreciation9.89.1
- Operating resultp. 87EBIT21.718.4
- ComputedEBIT margin11.8 %10.7 %
EUR million
- Your own template and line items
- Several years side by side
- Ratios calculated in code, not by the AI
Scale
Built for thick documents. And for lots of them.
An annual report runs to 300 pages, a framework agreement with annexes to 150, a prospectus to 500, and you have hundreds of them. Klartext Extract reads every document in full instead of in snippets, even past 1,000 pages, and works through whole folders on its own.
- 1,000+
- pages per document, read in full
- 10
- PDFs per extraction, merged into one record
- 1
- click from any value to its source in the PDF
- 15 min
- instead of 4 days per company at a rating agency
Upload a folder. Get one record per case.
- Every subfolder becomes its own extraction
- Duplicate files are skipped automatically
- Email when the whole batch is done
Traceability
No number without evidence. One click shows where it came from.
Every value is linked to its page and passage. Klartext Extract shows what kind of source it was, how confident the value is and why it was chosen, even when two documents disagree.
Pick a value
Annual_Report_2024.pdfp. 87
Kestrel Logistik GmbH87
Consolidated income statement
for the 2024 financial year · EUR million
| 2024 | 2023 | |
|---|---|---|
| Revenue | 184.6 | 171.2 |
| Cost of materials | -92.3 | -87.9 |
| Personnel expenses | -48.1 | -44.6 |
| Depreciation | -9.8 | -9.1 |
| Operating result (EBIT) | 21.7 | 18.4 |
| Net income | 14.9 | 12.6 |
- Evidence
- Audited table
- Confidence
- Very high
- Sources
- Annual report, p. 87Annual report, p. 3
AI explanation
Taken from the audited consolidated income statement, 2024 column. An audited table is the highest-ranking source, hence very high confidence.
How it works
From a folder of PDFs to a reviewed table.
- 01
Define your fields
Together we define the fields you need, with description, unit and format: from a few dozen to several hundred fields, with formulas for derived values.
- 02
Upload your documents
Up to ten PDFs per case, or a whole folder with one subfolder per case. Scanned pages get text recognition automatically.
- 03
Let it extract
The AI reads every document in full, finds each value with its source and merges the documents. Conflicts are settled by source rank: audited table before table before narrative text.
- 04
Review and export
Your team reviews sorted by confidence, approves, corrects and comments. Every edit goes into the change log, and the result flows into your systems as Excel or JSON.
Features
Built for teams who have to sign off.
Computed, not guessed
Sums, ratios and cross-checks are calculated by code from extracted values, traceable down to the formula.
Comments and change log
Discuss values right at the field. Every edit shows the original AI value and who changed it.
Review at keyboard speed
Move with ↑ ↓, approve with A, reject with R. No mouse needed.
Reproducible runs
Every run stores the schema and prompt version it was produced with.
Use cases
Wherever data hides in long PDFs.
Bonds & prospectuses
Key terms from every prospectus
Volume, coupon, maturity, call rights and issuer data from securities prospectuses of 500 pages and more.
| Field | Value | Source |
|---|---|---|
| Issue volume | EUR 500 m | p. 12Very high |
| Coupon | 4.25 % | p. 12Very high |
| Maturity date | 15.06.2031 | p. 13High |
| Issuer call right | Yes, from 2028 | p. 47Medium |
Comparison
Not a chatbot. A pipeline you can audit.
Manual analysis is thorough but slow. A general-purpose AI chat is fast but impossible to audit. Klartext Extract gives you both.
| Manual analysis | General AI chat | Klartext Extract | |
|---|---|---|---|
| Minutes per case instead of days | No | Yes | Yes |
| Whole documents of 1,000+ pages | Yes | Partly | Yes |
| Hundreds of documents in one go | No | No | Yes |
| The same fields for every document | Yes | No | Yes |
| Every value with its page and passage | Partly | Partly | Yes |
| Confidence per value | No | No | Yes |
| Conflicts between documents flagged and explained | Partly | No | Yes |
| Review workflow with change log | Partly | No | Yes |
| Data stays in the EU | Yes | Partly | Yes |
YesPartlyNo
In production
A European rating agency
Their analysts used to do Financial Spreading by hand, working through reports with hundreds of pages for every risk assessment. Today Klartext Extract delivers the values with their sources, and the team reviews and signs off.
Read the case study- 4 days → 15 min
- per analysis, review included
- >97 %
- accuracy at scale
- 1,000+
- pages per analysis, across several documents
Figures from the published case study on klartext-ai.com.
Security & deployment
Your data stays where it belongs.
Built for companies that work with confidential documents: hard boundaries between customers, and results you can trace.
EU-local
Hosting, storage and processing in the EU.
Your own instance
Enterprise customers get their own server, database, file storage and domain.
Strictly separated
Every query is scoped to your organization. An enterprise build contains only your own code and schemas.
People decide
The AI proposes and cites, your team reviews and approves. Every edit is logged.
Invite-only access · HTTPS everywhere · backups in a second region
Which documents can Klartext Extract process?
PDFs such as annual and sustainability reports, financial statements, contracts, prospectuses, policies or certificates, including scanned ones. A single document can have more than 1,000 pages, and one extraction can merge up to ten PDFs. Processing is tested on German and English documents.
How is this different from ChatGPT and similar tools?
A chat answers questions. Klartext Extract fills a fixed schema: the same fields and units for every document. Every value has a source and a confidence level, computed values are calculated by code rather than the AI, and your team reviews everything in one workspace with a change log.
How accurate is the extraction?
It depends on your documents and your fields, so we measure it on your own documents before you commit. The confidence level on every value shows your team where a review pays off: high-confidence values are wrong far less often than low-confidence ones.
Can we define our own fields?
Yes. We set up your schema with you, with descriptions, units, formats and formulas for derived values. Schemas in production range from a dozen to several hundred fields.
Where is our data processed?
In the EU. Enterprise customers get their own instance with a separate database, file storage and domain.
How does the data get into our systems?
As an Excel or JSON export, structured by your schema, ready for your own database or tools.
What does it cost?
Pricing is on request. It depends on your document volume and on the deployment, a shared platform or your own instance. Write to us and we will send you an offer.
Derived from § 14; confirmed with legal.
Edited by M. Huber