Discover and classify
Identify PII and particularly sensitive data. Record the relevant metadata and accountable owners.
Data & AI Engineering
Protect personal data without blocking analytics
Your teams need meaningful analysis, but identity, health and contract data cannot be made broadly accessible. We build data flows with clear classification, access controls and targeted masking.
What technical PII masking must achieve
PII masking hides or changes personal values for defined access contexts. It is not the same as encryption, pseudonymisation or irreversible anonymisation.
What matters is where the control takes effect: on ingestion, in storage, at query time or when data is exported.
Four control points
Restrict access
Grant roles, attributes, rows and columns according to a legitimate business purpose.
Mask or pseudonymise
Select the right mechanism for analytics, testing or development and protect any required re-identification process.
Operate with evidence
Make changes to permissions, data flows, exports, retention and deletion traceable.
Technical options and their limits
ClickHouse Cloud and other analytics platforms offer native masking policies. Microsoft Presidio can assist with PII detection; OpenMetadata and DataHub can support cataloguing and lineage. The right choice depends on the data flow, access model and operational environment.
Masking is not an access ban. Small cohorts, join keys and residual metadata can still reveal individuals. We assess minimisation, every access path and the controls applied to queries and outputs together. See Data & AI Engineering.
Data masking in practice
Our ClickHouse tutorial for a pension fund demonstrates how a database role receives masked values at query time. This is one control inside a wider governance model that also covers ownership, quality and auditability. Explore Data Governance.
Frequently asked questions
When should PII be masked rather than deleted?
When the data may lawfully be retained but certain roles do not need to see the original values. Masking never removes an applicable duty to delete data.
What is the difference between masking and tokenisation?
Masking changes the visible representation. Tokenisation substitutes a token whose mapping back to the original can be controlled separately.
Which systems can be integrated?
Databases, warehouses, data pipelines, APIs and analytical applications. The specific security architecture determines the implementation.
Can an AI model use masked data?
That depends on the purpose and what information remains. Training, inference, logging and external model providers all require their own data-flow and risk assessment.
Next step
Review the data architecture and access model.
We will review your data sources, responsibilities, access paths and the scope of a practical implementation pilot.
Aleksandar Basara
Managing Director
aleks@valanor.co
+49 (0) 711 16 44 6 83
Lindenspürstrasse 22, 70176 Stuttgart
