Article summary
Normalize New Zealand +64 variable-length phones and national zero notation, then test WS Gender/Age missingness and fairness without treating inferred profiles as identity.
The key to a New Zealand WhatsApp Gender and Age check is not dividing +64 phones into demographic buckets. It is correctly handling variable-length phones and identifying where profile-field missingness is concentrated. Results may support a limited aggregate study; they do not establish individual identity, ethnicity, credit or eligibility for service.
+64 is not one fixed length
The New Zealand dialling plan published through the ITU allows different national-number lengths for different service ranges. Its 2020 official update lists separate mobile and geographic ranges. Normalization needs a current range table rather than one universal length expression.
Preserve national zero and international notation
| Input | Treatment | Caution |
|---|---|---|
| +64… | Validate national significant part | Do not add country code again |
| 0064… | Recognize international access prefix | Keep raw value |
| National notation beginning 0 | Convert by range when NZ evidence exists | Do not strip every zero blindly |
| Unclear length/country | format_exception | Never guess |
Trans-Tasman files can mix codes
Australian and New Zealand customers, migrants and travel phones can coexist in one CRM. Country code describes a numbering plan, not current residence, citizenship or language. Preserve country_evidence separately from market_location.
The actual WS Gender and Age schema
AIPUSH WS Gender and Age may export phone, age, gender, avatar and mapped WhatsApp phone. Age, gender and avatar may be unknown or depend on limited public signals. A mapped phone is a relationship observation, not identity confirmation.
Accept with a stratified missingness matrix
| Dimension | Compare | Response to issue |
|---|---|---|
| Range/format | Coverage and errors | Review normalization rule |
| Source | Unknowns for form, store and event | Limit generalization |
| New/old cohort | Observation time and mapping collision | Mark data lifetime |
| Customer stage | Profile visibility difference | Do not call it value difference |
Never infer ethnicity from profile observations
New Zealand includes diverse communities, but age, gender, avatar, name and +64 cannot reliably establish Māori, Pacific or any cultural identity. Do not create such derived labels. Cultural adaptation should use community partnership and voluntary aggregate research.
Keep TXT separate from CRM context
The upload contains one phone per line and no name, ethnicity, region, order or permission. An internal crosswalk joins contact_id, and Excel lands in staging. Human review handles mapped-phone collisions; households and shared business records are never merged automatically.
Use profile data only for low-risk aggregate experiments
An approved content study uses broad age bands, minimum cells and neutral creative, reporting unknowns and uncertainty. Profile observations cannot decide price, finance, employment, insurance, treatment of minors or access to support.
Permission outranks a profile
A message still requires source, channel, purpose and no opt-out even when profile or activity is returned. English, Māori or another language preference comes from the person, never avatar and age.
Deletion and recheck
Raw image URLs and individual inference receive short retention; aggregate findings may remain for the stated purpose. Stop segmentation after expiry and recheck only when the research and a fresh approval still exist. Do not build a permanent demographic warehouse.
What a qualified result looks like
It records +64 rule version, stratum inputs and exceptions, field coverage, unknowns, mapping collisions, allowed use, roles and deletion date, explicitly rejecting ethnicity and personhood inference. Transparent limits matter more than a high “profile completeness” number.
