You have this file. Everybody has this file. It's an export from QuickBooks or Mailchimp or a list somebody kept on the side, and it's a mess: Robert Kessler, Bob Kessler, and Robert Kesler are all the same person, Dan Whitfield shows up again as Dana, phone numbers are in four different formats, and half the rows are missing something.
The old answer is to hand it to somebody and lose an afternoon. In this episode, Jason attaches the file, writes one plain-language prompt, and walks away.
The prompt does the whole job: clean it, merge duplicates keeping the most complete version of each person, normalize phone numbers, create or match each company and link it, skip rows with no name and no email. The last line is the one most people forget, and it's the most important: tell me how many you imported, how many you merged, and how many you skipped. Same thing you'd ask of an intern.
What happens next is the interesting part. Claude comes back with questions before it does anything destructive: the CSV has fields your database doesn't, what should it do with them? Two rows share an email, merge or keep both? Which phone format? Then it works, and it catches something nobody asked it to catch: the one contact that had been added by hand in the previous episode, which it recognizes and merges instead of duplicating.
Final result: 44 contacts, 43 companies, four duplicate sets merged, two rows skipped, company name variants normalized (Alamo Steel LLC and Alamo Steel are now one company), and every contact linked to the right company page.
Homework: run your own file, spot check a few of the records you know were questionable, and then never open that spreadsheet again.
Builds on the CRM from the previous episode. Works the same if your CRM is HubSpot or Salesforce.