Volumes that need a repeatable process
Thousands of folders and mixed formats: NymData processes a folder and its subfolders in one go, then only what is new on each run.
Manufacturing and industry
Give your AI vendor your data, not your customers
Service reports, customer records, spreadsheets and correspondence:
NymData pseudonymizes years of history on your servers before the vendor building your AI gets access.
The data stays usable for training models.
Real names never leave.
Before: customers and contacts identifiable, in red
Maintenance Department · Service Report
Service call on the packaging line
Customer: John Carter, Site Manager
Contact: (555) 013-2847 · [email protected]
Abnormal vibration on conveyor no. 3. Replaced the motor-side bearing, adjusted belt tension. Technician: David Miller.
Customer and technician names in every report
After: names pseudonymized, technical content kept
Maintenance Department · Service Report
Service call on the packaging line
Customer: Ryan Mitchell, Site Manager
Contact: [PHONE] · [EMAIL]
Abnormal vibration on conveyor no. 3. Replaced the motor-side bearing, adjusted belt tension. Technician: Kevin Hayes.
Same customer, same pseudonym across the whole history
- Years of history processed
- Same pseudonyms everywhere
- Natural text for AI
- Scheduled processing
How do you anonymize data for AI training?
You need to anonymize the identities of customers, employees and partners while keeping text and data that models can use.
With NymData, installed on your servers, you select your folders, files or databases.
Names are replaced with realistic pseudonyms, always the same for the same person.
Every run produces an audit report.
The vendor receives a pseudonymized copy, never the source files.
The case
Twenty years of history, an outside vendor, no identities to hand over
To build an in-house AI, a manufacturer brings in a specialized subcontractor.
The vendor needs real business data, not customer names.
And the history covers more than twenty years of files on the company's servers.
Consistency over the years
The same person gets the same pseudonym in every Word, Excel, CSV and PowerPoint file and across every year, with no mapping table to store.
Natural text for models
Realistic pseudonymization replaces "John Carter" with another plausible name: the text stays natural, which suits model training.
Documents covered
What manufacturers prepare for AI
Twenty years of formats, processed in the same project.
- Reports and minutes in Word or PDF: service calls, audits, meetings. Word
- ERP and CRM exports in Excel or CSV: customers, orders, complaints. Excel and CSV
- Databases in PostgreSQL or MySQL, copied to a working database for the vendor. Databases
- Quotes, invoices and letters in PDF, often the largest share of the archives. PDF
Use cases
When should you anonymize this data?
In-house AI with a vendor
Give the subcontractor a pseudonymized corpus to train and evaluate models.
Integrators and software vendors
Share realistic data for a migration or configuration project.
Test databases
Populate business application test databases with data that contains no real customers.
Putting archives to work
Use your history for analytics without exposing people.
Framework
What the GDPR says about this kind of sharing
Pseudonymized data is still personal data: the vendor acts as a processor, and the contract between you must govern the processing.
Pseudonymization greatly reduces the risk if the vendor suffers a breach, because it never receives the identities.
For data that must fall outside the scope of the GDPR, you need effective anonymization: no re-identification possible by reasonable means.
The choice depends on the use case and is made with your privacy officer.
Which method should you choose? · Pseudonymization vs anonymization: what is the difference?
Scope
What NymData handles for you
- SupportedMixed formats
- PDF, Word, Excel, PowerPoint, CSV, text files and PostgreSQL or MySQL databases in a single project.
- SupportedScheduled processing
- Daily, weekly or monthly: only new or modified files are processed.
- Good to knowLarge volumes
- The GPU option installs on your own hardware for large archives.
- SupportedTraceability
- Every run produces a timestamped, sealed audit report to add to your AI project file.
FAQ
Frequently asked questions
Does the vendor see the original data?
No. NymData runs on your servers and creates a pseudonymized copy in a separate folder: that copy is what you hand over.
Source files are never modified and stay on your premises.
Does the same person keep the same pseudonym across the whole history?
Yes. The pseudonym is computed from the original value and a project-specific key: the same person gets the same pseudonym in every Word, Excel, CSV and PowerPoint file and across every year, with no mapping table.
Can pseudonymized data be used to train an AI?
Yes. Realistic pseudonymization replaces names with other plausible names: the text stays natural and the links between documents are preserved.
Can new data be processed automatically?
Yes. A project can be scheduled daily, weekly or monthly: only new or modified files are processed on each run.
Do I need to send my files to NymData?
No. The software is installed on your servers: no file and no data is sent to NymData or to any cloud service.
See NymData prepare a corpus for AI, live
A demo on sample documents, remote or on site.