Financial Documents

Pull the numbers out of research PDFs

Sell-side reports, earnings decks, industry bulletins and statistical releases become clean rows in your schema, delivered by API or spreadsheet every time a new one comes out.

Q1Q2Q3Q4
Company Name
Total Revenue
Revenue Chart
Net Income
Fiscal Quarter

Extracted Data

Company Name

Acme Corporation

Total Revenue

$394.3B

Revenue Chart

Quarterly Revenue Data:

Q1$88.2B
Q2$97.3B
Q3$89.5B
Q4$119.3B

Net Income

$96.99B

Fiscal Quarter

Q4 2023

Extraction Complete5/5 fields

Send us a report and the columns you need

Attach one example document and describe the series. We will send back a sample table, usually overnight.

Our clients

Trusted by leading hedge funds

Powering data-driven decisions for the world's most sophisticated investment firms

$100B+

in Assets Under Management

100K+

Data Points Extracted & Validated

The data is in the PDF. Getting it out is someone's morning.

Broker research arrives as PDFs. Industry bodies publish monthly bulletins and change the layout every year. Analysts retype numbers into Excel, and the time series a model needs stops wherever someone ran out of patience.

  • •Manual transcription: an analyst retypes hashrate, volumes or prices from every new report.
  • •Layouts that drift: a table moves, a column is renamed, a footnote changes the unit, and the parser breaks.
  • •Backfill: the model needs years of history, and the archive is a folder of PDFs and scans.
  • •Many languages: statistics agencies and port authorities publish in their own language and format.

How it works

1

Point at the source

Upload the PDFs, send links, or name the publisher's page where new reports appear.

Scanned or digital.

2

Describe the series

Write the columns the way your model wants them: "miner, month, hashrate in EH/s" or "country, month, export volume in tonnes".

Plain language, your units.

3

Get the history, then the updates

sieve extracts the back catalogue into one series, then watches for new releases and adds them as they are published.

Backfill plus daily monitoring.

Reads tables by meaning

The agent finds the table by what it contains, not where it sits, so a new layout does not break the series.

History and new releases in one pipeline

Years of back issues and next month's report land in the same table, with the same columns.

One schema across sources

Totals, splits by direction, different units and languages are mapped into the columns you asked for.

Checked numbers

Automated checks run on every extraction. On enterprise plans, reviewers with financial data experience verify each data point.

What investment teams send us

  • Bitcoin miner metrics such as hashrate and coins mined from sell-side analyst reports
  • Monthly production volumes from industry bodies such as the National Cotton Council, backfilled for years
  • Country export statistics for a commodities desk, extracted the day they are published
  • Port throughput (TEU) records in several languages, normalized into one global series
  • Tables from earnings presentations, investor decks and annual reports

Upload a report, get a table

Send the PDF with a plain-language instruction. The result comes back as CSV or JSON. Setup and full examples

with open("broker-report.pdf", "rb") as f:
    start = requests.post(
        f"{BASE_URL}/api/scrapes",
        headers={"Authorization": f"Bearer {API_KEY}"},
        data={"instruction": "Extract monthly hashrate (EH/s) and bitcoin mined for every miner in this report"},
        files={"file": f},
    )
session_id = start.json()["session_id"]
# Poll GET /api/scrapes/{session_id} until status == "done", then download the CSV.

Our investors

Backed by the best

backed by world-class investors who believe in our vision

Y Combinator logo

Questions people ask

Yes. Scanned pages, low-quality images, stamps over text and tables that span pages go through the same pipeline as digital PDFs.

Nothing on your side. The agent reads the table by meaning, so a moved column or a renamed header does not need a new template, and the series keeps the same columns.

Yes. Send the archive or the publisher's index page and we extract the full history into one series, then keep it current as new reports come out.

No. Results can arrive as a spreadsheet by email. Engineers can call the same extraction from the API and get CSV or JSON.

2025 Sieve Data Inc. All Rights Reserved.