How to convert a PDF to Excel

PDF to Excel is one of the most common data conversion needs — and one of the most frustrating when the tools don't work. The core problem is that PDFs were designed for printing, not data extraction: columns are defined by visual position rather than structure, and there's no concept of 'this is a table' baked into the format.

The three types of PDF tables (and how hard each is)

• Bordered tables with visible grid lines — easiest. Camelot's lattice mode handles these reliably.
• Whitespace-aligned tables (columns separated by spaces, no borders) — medium difficulty. pdfplumber and Tabula usually work.
• Scanned or image-based tables — hardest. No text layer, so standard tools return nothing. Need OCR or AI extraction.

Method 1: Tabula desktop (no code, point-and-click)

Download Tabula from tabula.technology. Open your PDF, draw a box around the table, and click Export. This is the fastest option for a one-off conversion and requires no programming. The GUI shows you the detected columns in real time so you can adjust the selection before exporting.

Limitation: manual, one table at a time, requires Java. Not suitable for processing many PDFs.

Method 2: pdfplumber + openpyxl (Python)

For automating the conversion across many PDFs:

• pip install pdfplumber openpyxl
• import pdfplumber; from openpyxl import Workbook
• wb = Workbook(); ws = wb.active
• with pdfplumber.open('report.pdf') as pdf:
• for page in pdf.pages:
• table = page.extract_table()
• if table:
• for row in table: ws.append(row)
• wb.save('output.xlsx')

This writes raw text into cells. Numbers come out as strings — you'll need to convert them if you want Excel formulas to work. Use ws.cell(row=r, column=c).value = float(val.replace(',','').replace('$','')) to force numeric types on amount columns.

Method 3: pandas (cleaner for data analysis)

tabula-py wraps Tabula as a Python library and integrates directly with pandas, which makes type conversion and Excel writing cleaner:

• pip install tabula-py pandas openpyxl
• import tabula, pandas as pd
• dfs = tabula.read_pdf('report.pdf', pages='all', multiple_tables=True)
• # Combine all tables into one sheet, or write each to a separate sheet:
• with pd.ExcelWriter('output.xlsx') as writer:
• for i, df in enumerate(dfs):
• df.to_excel(writer, sheet_name=f'Table_{i+1}', index=False)

pandas infers types automatically — columns that look numeric become float64, which means Excel formulas work without cleanup. For invoices and financial documents where dollar signs and commas are in the cells, you'll still need to clean before pandas can infer the numeric type.

Method 4: AI extraction (for invoices, statements, forms, and scans)

For documents where the data isn't a simple table — invoices with vendor headers and line items, bank statements with account info plus transaction rows, forms with field-value pairs — or for any scanned PDF, AI extraction produces a clean Excel file with proper structure: header strip on one sheet, data rows on another, numeric values already typed correctly.

Why 'PDF to Excel' tools often produce garbage

Most online 'PDF to Excel' converters take a shortcut: they rasterize the page, run OCR, and dump the raw text into cells. The result looks like the original PDF but isn't actually structured — all data in column A, numbers as text strings you can't sum, merged cells that break filters. The useful conversion isn't 'put the text where it appears', it's 'understand what the data is and put it in columns'.

Getting proper numeric values in Excel

The most common problem after extraction: amount fields that look like numbers but Excel treats as text. Signs: left-aligned in the cell, sum returns 0, VALUE error on arithmetic. Fix: Data → Text to Columns → Delimited → Finish, or use VALUE() formula, or fix the currency symbol stripping in the extraction step.

Multi-sheet Excel for complex documents

Invoices and statements aren't one table — they're a header block plus line items plus totals. openpyxl lets you write each section to its own sheet:

from openpyxl import Workbook wb = Workbook() header_ws = wb.active header_ws.title = "Header" header_ws.append(["Vendor", "Invoice #", "Date"]) header_ws.append(["Acme Corp", "INV-1042", "2025-03-15"]) items_ws = wb.create_sheet("Line Items") items_ws.append(["Description", "Qty", "Unit Price", "Amount"]) for row in extracted_rows: items_ws.append(row) wb.save("invoice.xlsx")

pandas ExcelWriter with sheet_name per DataFrame is cleaner when you already have DataFrames. Name sheets predictably — downstream macros often hard-code sheet names.

Camelot for bordered financial tables

When pdfplumber splits columns wrong on a grid-lined P&L, try Camelot's lattice mode — it traces visible lines:

import camelot tables = camelot.read_pdf("report.pdf", pages="5", flavor="lattice") print(tables[0].df) print(f"Accuracy: {tables[0].accuracy:.0f}%")

If accuracy drops below 80%, the PDF probably uses whitespace alignment instead of borders — switch to flavor='stream'. Camelot requires Ghostscript and OpenCV; factor that into Docker images.

Bank statements are the worst case

A statement is a header block, a running transaction ledger that continues across page breaks, and a closing balance that has to tie out. Column detection alone gets you a spreadsheet that looks right and reconciles to nothing: repeated page headers land in the middle of the ledger, and multi-line descriptions split one transaction into two rows.

Validating numbers after export

• Sum the amount column and compare to the PDF total — mismatch means a dropped row or merged cell.
• Check for parentheses negatives: (1,234.56) must become -1234.56, not a text string.
• Watch thousand separators: European PDFs use 1.234,56 — strip dots before parsing if your locale is US.
• Merged header rows produce NaN column names — forward-fill the header row before to_excel().