Canero

Privacy Model

Privacy by design, not by policy

Canero's privacy model combines identity isolation, k-anonymity-style threshold screening, and multi-layer checks to ensure individual feedback can never be traced.

Foundation

Identity isolation by architecture

The most important privacy property in Canero isn't a policy - it's an architectural constraint. An employee's identity and their conversation content are stored in completely separate systems, and the application serving the product has no permission to access identity information.

Identity stored in an isolated layer

Content stored with pseudonymous identifiers only

Application has no access to identity mappings

Connection between identity and content is architecturally prevented

Participant Abstraction

Identity Layer

Pseudonym:abc123xyz
Real identity:••••••••
↓ no connection possible ↓

Content Layer

Pseudonym:abc123xyz
Feedback:[themes only]

Threshold Screening

Multiple checks, every report

Privacy screening runs at report generation time - not as a post-processing step. If any of the independent checks fail, the report is withheld entirely. There is no partial report, no redacted version - just no report.

This approach is designed to prevent re-identification attacks including elimination, homogeneity, and auxiliary information attacks.

Privacy Screening ThresholdsEnforced per report

Organization-wide reports

10+conversations

Minimum before any org-level insights are generated

Segment reports

10+conversations

Higher bar because segment data is more identifying

Theme reports

7+conversations

Topic groupings require a diversity buffer

Segment headcount

5+members

Segments smaller than this are treated as identity-equivalent

Sentiment diversity

2+distinct buckets

Blocks homogeneous themes that could reveal individual views

Report available

All thresholds met

Report withheld

Insufficient participation

5 distinct failure modes evaluated on every report generation

Privacy screening checks

Each check addresses a distinct re-identification risk. All must pass before a report is generated.

1

Insufficient conversations

A report requires a minimum number of completed conversations for the period. If participation falls short, no report is generated.

2

Insufficient segment size

Even if a segment has enough conversations, the segment must represent a minimum headcount to prevent elimination attacks.

3

Insufficient topic coverage

Theme tags require enough conversations mentioning the theme before that theme appears in a report.

4

Insufficient segment headcount

Segments smaller than a minimum size are treated as identity-equivalent - a unique segment of one is the same as being identified directly.

5

Insufficient sentiment diversity

If a theme is mentioned only by people with the same sentiment, the report could reveal who holds a specific view. Diversity of sentiment is required.

Aggregation

k-Anonymity and beyond

Canero's screening model implements principles inspired by k-anonymity and l-diversity. The threshold checks ensure that any single person's response is indistinguishable from at least k-1 other responses. Sentiment diversity checks address the l-diversity requirement for attribute protection.

These are architectural constraints that exceed what most commercial employee feedback tools implement.

k-Anonymity principle

Each report requires enough participants that no individual can be distinguished from the group. The org threshold of 10 means your response is indistinguishable from at least 9 others.

l-Diversity for attributes

Sentiment diversity screening ensures that within a theme, there are at least 2 distinct sentiment buckets. This prevents identifying who holds a specific viewpoint even in a large group.

See the full technical picture

Learn more about our security architecture, data isolation approach, and identity separation guarantees.