How to convert CSV to Excel without breaking the data

A CSV is text with commas in it. It has no column types: nothing in the file says that 00412 is a customer number and not the number 412, or that 2026-03-04 is a date. Every program that turns a CSV into a spreadsheet guesses, and for IDs, postcodes, phone numbers and long reference numbers the guess is wrong. Converting without breaking the data means making those decisions yourself, column by column.

import pandas as pd

df = pd.read_csv("customers.csv", dtype=str, keep_default_na=False)
df["joined"] = pd.to_datetime(df["joined"], format="%Y-%m-%d").dt.date
df["balance"] = pd.to_numeric(df["balance"])

df.to_excel("customers.xlsx", index=False)

What the guessing does

The test file is three customers, written the way a CRM export writes them (UTF-8, comma-separated):

customer_id,name,postcode,phone,order_ref,joined,balance
00412,Zoë Müller,02134,0612345678,7310000012345678901,2026-03-04,1234.50
00977,Łukasz Nowak,08001,0031201234567,7310000012345678902,2026-11-12,87.05
01500,Søren Ågård,2100,+4512345678,7310000012345678903,2026-01-31,-15.00

Each converter's .xlsx was opened again to read the type and value of every cell. The first data row:

The order numbers are the worst case. Read as numbers, all three rows become the same value, 7310000012345679000, because a spreadsheet number holds about 15 significant digits and these have 19. LibreOffice's 7.31000001234568e+18 is that value. Microsoft's own page on keeping leading zeros and large numbers gives the same limit for Excel: 15 digits, and anything longer must be text.

The browser converter's date is the date serial 46085 with the number format yyyy-mm-dd. It is worked out from the digits alone, so the time zone doesn't come into it: we got exactly 46085 with the computer set to New York.

There is one difference between pandas and the browser converter that the first row doesn't show. pandas was told that postcode is a text column. The browser converter decides cell by cell, so Søren's postcode 2100, which has no leading zero, became the number 2100, in a column where the other two are text. Sorting and lookups treat the two kinds differently. When a column has to be text all the way down, say so in pandas or LibreOffice.

Convert with pandas

The script above is the whole job for one file. Reading with dtype=str keeps every value exactly as it was written; keep_default_na=False stops pandas turning empty cells and words like NA into missing values. Then convert the columns that are numbers or dates on purpose.

Dates need .dt.date. Without it pandas writes a date and time (format YYYY-MM-DD HH:MM:SS), and with openpyxl as the writer the date_format option of ExcelWriter was ignored in our test (pandas 2.2.3).

If you would rather name the text columns and let pandas read the rest, pass a dictionary. This version also handles a European export, with semicolons between fields and a comma as the decimal mark:

import pandas as pd

df = pd.read_csv("customers-semicolon.csv", sep=";", decimal=",",
                 keep_default_na=False,
                 dtype={"customer_id": str, "postcode": str,
                        "phone": str, "order_ref": str})
df["joined"] = pd.to_datetime(df["joined"], format="%Y-%m-%d").dt.date
df.to_excel("customers-eu.xlsx", index=False)

1234,50 comes out as the number 1234.5. Three more things the file may need:

• Day-first dates (04/03/2026 for 4 March): format="%d/%m/%Y". An explicit format is the only safe way. Guessed, that date is 3 April.
• A Windows export in the old Western code page: encoding="cp1252". Without it, pandas stops with UnicodeDecodeError rather than writing broken accents, which is the better failure.
• A byte order mark at the start of the file: nothing to do. pandas 2.2.3 skipped it and read the first header as customer_id.

A folder of exports

from pathlib import Path
import pandas as pd

TEXT = {"customer_id": str, "postcode": str, "phone": str, "order_ref": str}

for csv in Path("exports").glob("*.csv"):
    df = pd.read_csv(csv, dtype=TEXT, keep_default_na=False)
    df["joined"] = pd.to_datetime(df["joined"], format="%Y-%m-%d").dt.date
    df.to_excel(csv.with_suffix(".xlsx"), index=False)
    print(csv.with_suffix(".xlsx"), len(df), "rows")

exports/february.xlsx 3 rows
exports/january.xlsx 3 rows

Convert with LibreOffice

LibreOffice converts from the command line without opening a window, and its CSV filter takes the column types as a string of numbers:

soffice --headless \
  --infilter="Text - txt - csv (StarCalc):44,34,76,1,1/2/2/2/3/2/4/2/5/2/6/5/7/1,1033" \
  --convert-to xlsx customers.csv

convert /w/customers.csv as a Calc document -> /w/customers.xlsx using filter : Calc Office Open XML

The string, token by token:

So columns 1 to 5 are text, column 6 is a year-month-day date and column 7 is read as a number. In the result, the IDs, postcode, phone and order numbers are text cells with Excel's text format (@), so they stay text if someone edits them later. LibreOffice's CSV filter options page lists the rest.

For the semicolon file with decimal commas, change two tokens: 59,34,76,1,1/2/2/2/3/2/4/2/5/2/6/5/7/1,1031. For day-first dates, give that column format 4 instead of 5. For a folder, pass a wildcard and an output directory:

soffice --headless \
  --infilter="Text - txt - csv (StarCalc):44,34,76,1,1/2/2/2/3/2/4/2/5/2/6/5/7/1,1033" \
  --convert-to xlsx --outdir out exports/*.csv

Without --infilter, LibreOffice guesses: it dropped the leading zeros, turned the order numbers into 7.31000001234568e+18, and put the semicolon file into one column, split again at each decimal comma. It did read the ISO dates as dates.

In the browser

The CSV to XLSX Converter types each cell by rules that never change a value: text that only looks like a number stays text, plain numbers become numbers, and YYYY-MM-DD becomes a date. It is the quick route for one file when you can't install anything.

The workbook made from our test file holds what the preview shows: 00412, 02134, 0612345678 and the order numbers as text cells, Zoë Müller with its accents, the date as 46085 formatted yyyy-mm-dd, and 1234.5 as a number.

Two settings cover what a file can't say for itself:

• Decimal separator. A semicolon file is read with decimal commas when its numbers look like 1234,50 and none look like 1234.50, so 1234,50 becomes 1234.5. A comma-separated file always uses a point. Set it by hand for a file that mixes the two.
• Dates like 04/03/2026. Nothing in the file says whether that is 4 March or 3 April, so by default such dates stay text, exactly as written. Choose EU (DD/MM/YYYY) or US (MM/DD/YYYY) and they become real dates, shown as yyyy-mm-dd.

Which to use

• One file, no install: the CSV to XLSX Converter, after a look at the preview. It keeps IDs, phone numbers and long references as text and reads ISO dates as dates without being told.
• A column that must be one type all the way down (postcodes where some have no leading zero, a code that is sometimes all digits): pandas with dtype=str, or LibreOffice with a filter string. They type the column, not the cell.
• Many files with the same columns: the pandas loop or soffice with a wildcard. Check one result before you send the batch.
• A one-off file in Excel itself: Data › From Text/CSV, with types set per column. See why your CSV opened wrong.

Before you send the .xlsx on, open it and click a cell in each ID column: the leading zeros should be there, and the full number should show in the formula bar.