Skip to main content
PulsarKit

Convert a bank statement PDF to Excel

Pull tables out of a PDF into a real XLSX workbook, one sheet per table, each value in its own cell. Runs on our server; files are deleted automatically within about two hours.

Drop your files here

.pdf · up to 300 MB · up to 30 jobs an hour · uploaded to our server and deleted after processing; the result is wiped within about two hours

Table detection
More settings

How to Extract to Excel

  1. 1

    Upload your PDF

    A report, invoice, statement or any document with tables, up to 300 MB and 200 pages.

  2. 2

    Pick the detection mode

    Auto works for most documents; choose Lattice or Stream yourself if a table has an unusual layout.

  3. 3

    Download the workbook

    One sheet per table, each value in its own cell. Cells arrive as text, numbers included, so SUM skips them until you convert them. Excel usually marks such cells: select them, open the alert next to them and choose Convert to Number.

What Auto did with two statements

The page opens with “Auto” under “Table detection”, the PDF to Excel default. On October 21, 2026, we tried it on a statement we made up: three pages, 75 rows. Its columns are Date, Description, Debit, Credit and Balance. On a ruled copy, with lines around every cell, Auto followed the grid: three tables of a header and 25 rows. A description printed on two lines stayed in one cell. On a borderless copy, Auto read how the text lines up instead. Every date and amount still landed in its own column, but each table began with the heading lines printed above it. Forcing “Stream (borderless tables)” on the ruled copy gave that result too, so leave “Auto” on.

One sheet for the whole statement

A statement is one long list, so on this page “Merge tables with the same columns” is already on under “More settings”. The ruled copy then came back as one sheet: the header once and all 75 rows below it. On the borderless copy, four heading lines sat on top, and the second line of each long description became a row of its own, 15 in all, with an empty Date cell. Delete those rows and 76 remain, as on the ruled copy, with long descriptions cut to their first line.

The amounts arrive as text

Every cell comes through as text, dates and amounts included: all 147 amounts on our statement were text cells. Excel's SUM ignores text values, Microsoft's help says, and in LibreOffice Calc a SUM over our Debit column gave 0. Microsoft's help page on numbers stored as text gives the fix, as of October 21, 2026: select the amount cells, open the alert beside them and choose “Convert to Number”. A formula in a spare column also works: =IF(C2="",0,NUMBERVALUE(C2,".",",")). It reads the statement's own decimal point and thousands commas, whatever your regional settings, and counts an empty cell as 0. Converted that way in LibreOffice, our Debit and Credit columns came to 11,692.64 and 7,166.41, the sample's own totals.

Your statement goes to our server

Table detection runs on our server, so the PDF is uploaded there. It is deleted as soon as the job finishes, and the workbook is wiped from the server within about two hours. Our runs took three to six seconds from “Extract to Excel” to “Download”. A statement that asks for a password is refused. Remove the password first in Unlock PDF; the password is never sent anywhere. A scan needs OCR PDF first. Our scanned copy then gave the same 75 rows, every amount intact. To continue in another tool, PulsarKit may store a local copy of your input or result in this browser. It can remain after you close the tab or restart the browser. Clear PulsarKit’s site data to delete it.

Processed on our server, deleted automatically

Table detection runs on our server with the camelot library, so the file is uploaded for processing. The uploaded file is deleted as soon as the job finishes. The result stays downloadable for one hour and is wiped from the server within about two hours, whether or not you download it. There is no account and nothing to clean up by hand.

Frequently asked questions

What's the difference between Lattice and Stream?

Lattice detects tables drawn with visible grid lines and is the more precise option when those lines exist. Stream infers columns from text alignment and spacing, for tables that have no lines at all. Auto tries Lattice first and falls back to Stream if needed.

What does “Minimum accuracy” do?

It sets how cleanly a table's rows, columns and cell boundaries must be detected for the table to reach the workbook: from 50 to 95 percent, 80 by default. Tables below it are left out, so garbled data does not end up hiding in a sheet. If none is left, the page says so and suggests the other detection mode.

Why does a table without visible lines detect worse?

Without grid lines, detection has to guess column boundaries from spacing alone, which is less certain than reading a drawn line. Try Stream mode, since it is built for this case.

What if my PDF is a scan?

A scan has no text layer to read tables from. Run OCR PDF on it first, then extract tables from the result.

Do formulas carry over?

No. A PDF never contains formulas, only the values a spreadsheet produced when it was exported.

Why does SUM give 0 on the amounts?

Because every cell arrives as text, numbers included. On October 21, 2026, all 36 cells of our test table came out as text, “1,250.00” and “-12.99” among them, and Excel's SUM ignores text values. To turn them into numbers, select the cells and choose Convert to Number from the alert Excel usually shows next to them, or use the VALUE function in a new column. If your Excel uses a comma as the decimal mark, those two may misread “1,250.00”; NUMBERVALUE with “.” and “,” as the separators converts text to a number regardless of regional settings. All three are from Microsoft's Excel help, checked on October 21, 2026.

How do I get CSV instead of an Excel workbook?

Use PDF to CSV. It runs the same table detection and gives you individual CSV files instead.

How long is my file kept on the server?

The uploaded PDF is deleted as soon as the job finishes. The workbook stays downloadable for one hour and is wiped from the server within about two hours, whether or not you download it.

Related tools