How to extract data from an annual report

Annual reports and 10-K filings pack the numbers you need — revenue, net income, segment breakdowns, EPS — into dense PDFs that weren't designed for copying. Here are four methods to get the data out, ordered from simplest to most powerful.

Method 1: Copy-paste (works for one table)

Open the PDF in Adobe Acrobat or your browser. Click and drag to select a table. Paste into Excel. This works about half the time on text-based PDFs — the rest of the time, the columns collapse or the numbers land in the wrong cells.

Use this when you need one table from one filing and you can spend five minutes cleaning it up. It breaks immediately on multi-page tables, scanned reports, and anything with merged cells.

Method 2: Tabula (free, open-source)

Tabula is a free desktop app that draws a selection box around a table in a PDF and exports the contents as CSV. It works well on clean, text-based PDFs where the columns are visually obvious.

• Download Tabula (tabula.technology), open your annual report PDF.
• Draw a selection box around the income statement or balance sheet.
• Click Export and choose CSV or Excel.
• Repeat for each table you need.

Limitations: Tabula can't read scanned PDFs (it needs a text layer), it fails on tables that span multiple pages, and you have to select each table manually — so extracting five years of filings is tedious.

Method 3: Python with pdfplumber

For batch extraction — pulling the income statement from twelve quarterly filings, or extracting the same metric across a dozen competitors — Python gives you a repeatable pipeline.

pdfplumber is the most reliable library for table extraction from text-based PDFs:

• pip install pdfplumber pandas
• import pdfplumber; pdf = pdfplumber.open('10k.pdf')
• tables = pdf.pages[12].extract_tables() # page number varies by filing
• import pandas as pd; df = pd.DataFrame(tables[0][1:], columns=tables[0][0])
• df.to_csv('income_statement.csv', index=False)

The hard part is that page numbers and table positions change between filers and between years. You'll need to inspect each PDF to find the right page, and the column names will vary ("Net revenues" vs "Total revenue" vs "Revenue, net"). For consistent schema extraction across multiple companies, you need something that understands the semantics, not just the layout.

Method 4: AI-powered extraction (handles scans, inconsistent layouts, any schema)

AI extraction reads the document semantically instead of by layout coordinates. You describe what you want — "income statement with revenue, operating income, and net income by year" — and the model finds and normalizes the data regardless of how the filing is formatted.

This handles the cases that break Tabula and pdfplumber: scanned annual reports, multi-page tables, IFRS vs GAAP formatting differences, non-US currencies, and consistent field extraction across many companies with different layouts.

What you can extract from an annual report

• Income statement — revenue, gross profit, operating income, net income, EPS by year
• Balance sheet — total assets, current assets, liabilities, equity, cash, long-term debt
• Cash flow — operating, investing, and financing activities; free cash flow; capex
• Segment breakdown — revenue and profit by business unit or geography
• Key metrics — growth rates, margins, guidance, KPIs
• Risk factors — material risks as structured rows with title and summary
• Executive compensation — named officers, base salary, stock awards, total comp
• ESG data — emissions (scope 1/2/3), diversity %, board governance

Scanned vs digital PDFs

If your annual report was scanned — common for older filings and international companies — copy-paste and Tabula won't work at all. pdfplumber will return empty tables. AI extraction (method 4) handles scanned PDFs because it applies OCR before extraction.

For most modern 10-K and 20-F filings from public companies, the PDF is text-based, so all four methods apply. If you're pulling older filings (pre-2010) or private-company annual reports, assume scanned and plan accordingly.

Extracting the same fields across multiple companies

Competitive analysis — pulling the same income statement metrics from five companies' annual reports — is where the difference between methods is most visible. Copy-paste and Tabula require manual effort per filing. Python scripts require per-filer tuning. AI extraction uses the same prompt on all filings and returns a consistent schema regardless of how each company labels their line items.

The output is a JSON array where each row is a year and each key is a field you asked for — ready to paste into Excel or import into a model.