Step 3: Delete High-Risk Fields

Some personal data has no analytics value but high privacy risk. Delete it entirely rather than trying to anonymize it.

The Goal

• Delete customer_name - No aggregate analytics value
• Delete customer_address - Too specific, no value anonymized
• Delete customer_dob - But convert to age bucket first

Why Delete vs. Hash?

Treatment	Use When	Result
Delete	No analytics value, high risk	Field removed entirely
Hash	Need to count unique values	Irreversible pseudonym
Generalize	Need distribution analysis	Reduced precision

Names and addresses have no meaningful aggregate analytics value—you can't sum them or count unique patterns. Delete them.

Implementation

step-3-delete.yaml

pipeline:
  processors:
    # Steps 1-2 from previous...

    # Step 3: Delete high-risk fields
    - mapping: |
        root = this

        # Full name - no analytics value, high risk
        root = root.without("customer_name")

        # Full address - no analytics value
        root = root.without("customer_address")

        # Date of birth - convert to age bucket before deleting
        root.customer_age_bucket = match {
          this.customer_dob == null => "unknown",
          now().ts_unix() - this.customer_dob.ts_parse("2006-01-02").ts_unix() < 25 * 365 * 24 * 3600 => "18-24",
          now().ts_unix() - this.customer_dob.ts_parse("2006-01-02").ts_unix() < 35 * 365 * 24 * 3600 => "25-34",
          now().ts_unix() - this.customer_dob.ts_parse("2006-01-02").ts_unix() < 45 * 365 * 24 * 3600 => "35-44",
          now().ts_unix() - this.customer_dob.ts_parse("2006-01-02").ts_unix() < 55 * 365 * 24 * 3600 => "45-54",
          now().ts_unix() - this.customer_dob.ts_parse("2006-01-02").ts_unix() < 65 * 365 * 24 * 3600 => "55-64",
          _ => "65+"
        }
        root = root.without("customer_dob")

Understanding the Code

Expression	What It Does
root.without("field")	Remove field from record
now().ts_unix()	Current time as Unix timestamp
.ts_parse("2006-01-02")	Parse date string
match {...}	Pattern matching for age calculation

Age Calculation Logic

The age bucket calculation:

1. Parse DOB as timestamp
2. Subtract from current time (gives age in seconds)
3. Compare against thresholds (years × 365 × 24 × 3600)
4. Return appropriate bucket string

Expected Output

Input:

{
  "customer_name": "Hans Schmidt",
  "customer_address": "Hauptstraße 42, 10115 Berlin, Germany",
  "customer_dob": "1985-03-15",
  ...
}

Output:

{
  "customer_age_bucket": "35-44",
  ...
}

Note: customer_name, customer_address, and customer_dob are completely removed.

Adaptation checks

Audit What Was Deleted

Track the selected deletions for debugging and review:

root._gdpr_compliance.fields_deleted = ["customer_name", "customer_address", "customer_dob"]
root._gdpr_compliance.deletion_timestamp = now()

Configurable Age Buckets

Make buckets configurable for different markets:

# Some markets use different age groupings
let buckets = env("AGE_BUCKETS").or("18-24,25-34,35-44,45-54,55-64,65+")

Handle Missing DOB

Gracefully handle records without DOB:

root.customer_age_bucket = if this.customer_dob == null {
  "unknown"
} else if this.customer_dob == "" {
  "unknown"
} else {
  # age calculation...
}

EU-Specific Age Categories

For GDPR, consider child protection (under 16):

root.customer_age_bucket = match {
  # GDPR considers under-16 as children requiring parental consent
  age_years < 16 => "child",
  age_years < 18 => "16-17",
  age_years < 25 => "18-24",
  # ... rest of buckets
}

---

Next Step

📖 Step 4: Hash Identifiers →