Clean up a CSV file before importing it into a CRM

01What the tool fixes, row by row

A botched import rarely shows on screen. It shows three weeks later, when a rep searches for “Dupont” among first names, or when the same client exists three times over. Here is what the tool repairs before the file ever reaches a CRM.

02Why “DUPONT Jean” so often comes out backwards

A naive split treats the first word as the first name. Yet directory exports and back-office software often put the last name first, so “DUPONT Jean” turns into a first name of DUPONT and a last name of Jean, with no warning anywhere. The tool reads the order from the value itself:

A particle left dangling at the end of a first name goes back to the last name, and a single word (“Madonna”) becomes a first name rather than an orphaned last name. The preview shows, row by row, which order was chosen and why. If your file follows a different convention, one checkbox swaps first and last name on every row.

03A “Name” column does not mean the same thing in every file

Next to a “First name” column, a “Name” column holds the last name. On its own, especially in French exports, it often holds the full name. A tool that decides from the header alone is bound to get one of the two cases wrong. This one looks at the values: when at least 60% of the cells have two words or more, the column is treated as a full name and the tool offers to split it.

Other columns are recognized the same way, header first and values second: email, phone, company, LinkedIn profile, web address, date, number. Email is tested before date, and date before phone, because a date like 2025-12-18 has the shape of a phone number. The detected columns table also shows how full each column is.

04Phone numbers: one format to recognize a person

A CRM compares phone numbers to recognize a person. Yet 06 12 34 56 78, 06.12.34.56.78 and +33 6 12 34 56 78 are the same line without being the same string. The tool brings them all to +33612345678, the international format defined by ITU recommendation E.164. French numbers have 10 digits, as Arcep, the French telecoms regulator, puts it: the leading 0 drops and the country code takes its place.

French overseas mobiles take their own country code: +590 for a number starting 0690, +594 for 0694, +596 for 0696, +262 for 0692 or 0639, according to Arcep’s national numbering plan. A number already written with + or 00 keeps its country code. A number with no country code is read in the country you pick, France by default. Anything still unreadable (“06 11 22 33 44 ext 12”) is left as it was and highlighted in the preview.

05Duplicates: email or phone, never the name

Two rows are duplicates when they share an email address or a phone number. Emails are compared ignoring case and spaces, but dots and +tag aliases are kept: j.dupont@ and jdupont@ are two separate inboxes with most providers. Phones are compared once converted to international format, which matches 06 12 34 56 78 with +33 6 12 34 56 78.

The name plays no part: two people called Jean Martin are often two different people. You choose between adding a column that points at the first occurrence (“row 2”) or removing duplicate rows and keeping the first one. Merging two records that complement each other is CRM work, not file work.

06Comma or semicolon: separator and encoding

RFC 4180, which has described the CSV format since October 2005, separates fields with commas and wraps in double quotes any field that contains a comma, a quote or a line break. Excel plays by another rule: Microsoft explains that it switches to semicolons when the comma is the decimal separator, which is the case on a machine set to French.

The tool detects the separator (comma, semicolon, tab or pipe) and the encoding: a file that is not valid UTF-8 is reread as Windows-1252 instead of showing diamonds where the accents should be. On export, pick the comma for a CRM import, or the semicolon to reopen the file in Excel on a machine set to French. The file comes out in UTF-8, with the byte order mark that lets Excel display accents correctly.

07Your file stays on your computer

Reading, fixing and downloading all run in JavaScript, inside the open tab. Not a single row of your file is sent, to Kasar or anywhere else: no request contains your data. You can check it in the Network tab of your browser’s developer tools: the site’s audience measurement shows up there, your file never does. A 50,000 row file is processed in under a second on a recent computer.

08Next: importing into Kasar

Kasar makes these fixes during the import itself. You drop a CSV or an Excel file, up to 150 MB per file and 10 files at once. Kasar detects the encoding and the separator, suggests where each column goes, or lets Leo prepare the mapping with you. It splits names with the same rules, with a swap option. It converts phone numbers to international format based on the country you pick.

Before you launch, the preview shows duplicates and errors. Contacts are recognized by email, phone or LinkedIn profile, and you choose to create, skip or update existing records. A report then details what was created, updated or set aside. A CSV export from HubSpot, Pipedrive or Salesforce imports the same way, and a Notion database connects directly.