How do you migrate CRM data without losing or duplicating records?

What belongs in a CRM data migration checklist?

1. Scope. Decide which objects and which years you migrate. Not everything has to move.
2. Backup. Take a full backup of the old system before you touch anything.
3. Export. Export every object you keep, with IDs, so links between records survive.
4. Field mapping. Map each old field to a new field, or decide to drop it.
5. Clean and dedupe. Fix formats and merge duplicates on a clear key such as email or domain.
6. Test import. Run a dry run, then import into a test workspace.
7. Review. Let the pilot team check real records, not just counts.
8. Final import and freeze. Import, then set the old system to read-only.

Which file format should you use for a CRM migration?

CSV is the common denominator, and RFC 4180 describes the format most tools expect, including how quotes and line breaks inside fields work ^[1]. A lot of broken imports come from exactly that, a note field with a line break or a semicolon export from a German Excel, and not from the CRM.

So check the encoding and the delimiter before anything else, and open a sample in a text editor, not in Excel.

How do you avoid duplicates during the import?

Pick one dedupe key per object, email for contacts and domain for organizations usually, and stick to it. Then import with that key, so a second run updates records instead of creating them twice. The GDPR's accuracy principle asks for personal data that is accurate and kept up to date ^[2], and a CRM with three versions of the same person fails that quietly.

How can an AI agent help with a CRM migration?

Agents are good at the boring middle part: finding near-duplicates that a key misses, normalising company names, checking which contacts have no organization. Over the MCP server Claude can run a dry run, show what would change and only import after you approve. I wouldn't let an agent decide the field mapping though, that's a business decision.

What is your rollback plan if the import goes wrong?

You need one for the old system and one for the new. For the old one it's the backup from step two. On a PostgreSQL database, pg_dump writes a consistent backup even while the database is in use ^[3]. For the new system, check whether imports can be undone as one unit.

How should a CRM import be checked before it goes live?

With a dry run first and real records second. At JadenX we built the Sigma CRM import so that it imports and exports CSV and Excel for every object, runs a dry run that previews the counts without writing, asks for approval before the real import, and can roll an import back. Duplicates that remain can be merged into the contact you keep. The same tools work in the interface and for Claude over MCP.

Frequently asked questions