Guides

Anonymize customer emails and support tickets

Support tickets are full of the things an AI is good at: summarizing a long thread, drafting a reply, sorting complaints by topic. They are also full of customers: names in the greeting and the signature, email addresses in the quoted thread, phone numbers, addresses and payment details.

Where tickets go

In each case the content of the ticket matters and the customer's identity doesn't. For a reply you want to send back, use placeholders and restore the answer (see removing personal data before ChatGPT). For everything that leaves for good, fake data reads better.

Why fake data for tickets

A ticket full of [PERSON_1] and ***** is hard to read and looks odd in training material. The Anonymizer's Support ticket preset replaces values with realistic stand-ins instead: a name becomes another name, an email another email at an example domain, an IBAN a made-up one that still passes its check digit. The same value gets the same stand-in each time, so a thread with ten quoted replies still reads as one conversation.

Fake data is one-way: the originals are not kept. That is the point when the text is shared for good.

Step by step

  1. Copy the ticket or the email thread, including quoted replies and signatures.
  2. Open the anonymizer with the Support ticket preset and paste the text.
  3. Review the highlighted findings. Signatures are where most misses hide: a job title, a department, a direct line written in an unusual format.
  4. Add your own identifiers to the custom terms list: customer numbers, order numbers, contract IDs, product or project names that point to one customer.
  5. Copy the result, or download it if you opened the ticket as a file.

What detection can't know