Trust
Privacy is the product.
Labs want the structure of real work — not the identities of the people in it. Our pipeline keeps the first and removes the second.
The pipeline
Seven steps from raw archive to licensed dataset.
- 01
Scope
Systems, date ranges and exclusions are fixed in the contract. Excluded data is never pulled.
- 02
Ingest
Read-only export into our isolated environment. Access is logged and limited to the anonymization team.
- 03
Resolve identities
People, companies, addresses, emails, phones and account numbers are detected and mapped to one identity graph per company.
- 04
Tokenize
Every identifier becomes a consistent token, so the same person is PERSON_014 in every tool.
- 05
Strip
Secrets, credentials, card numbers and government IDs are removed outright. Faces, logos and signatures are blacked out in images and PDFs.
- 06
Review
Automated leak checks, a human audit sample, and your own review of output from your data before delivery.
- 07
Delete
Raw archives are deleted within 30 days of sign-off. You receive a deletion certificate.
See it
The same thread, before and after.
Same person → same token, across every tool.
# eng-billing · Slack
PERSON_014
@PERSON_027 the COMPANY_221 invoice batch failed again — 1,284 rows. Customer contact is PERSON_390 (EMAIL_390).
PERSON_027
Rolling back PR #2213. Opening BILL-482 and I'll call her at PHONE_390 before EOD.
Re: March invoices · Email
From: PERSON_390 <EMAIL_390> To: PERSON_014 Hi PERSON_014 — our AP team flagged duplicate charges on the March invoices. Card ending 4417. Can you confirm a credit before our close on the 5th? — PERSON_390, COMPANY_221 ADDRESS_118
Employee data standard
The people in the data didn't sign the deal. We protect them anyway.
- ✓HR, payroll, benefits, medical and legal-privileged folders are excluded by default.
- ✓Direct messages are excluded unless the company specifically includes them and has the right to do so.
- ✓Every remaining personal identifier is tokenized; no buyer ever receives a real name or contact detail.
- ✓Licences prohibit re-identification, profiling, or contacting anyone represented in the data.
- ✓Data is licensed for model training and evaluation only — never resold for marketing or people search.