Article summary
Separate Kosovo’s +383 numbering evidence, customer market and WhatsApp profile observations without turning age or gender fields into identity claims.
Kosovo uses the +383 calling code, but a collection of +383 phones is not a register of “Kosovo identities.” The prefix first identifies a numbering plan. Residence, language and self-identification require separate evidence such as an order, an address or a user-provided preference.
Draw three evidence lines that must not merge
| Business field | Acceptable evidence | What cannot replace it |
|---|---|---|
| number_plan | +383 and phone syntax | Avatar or name |
| customer_market | Order, address, self-report | Calling code alone |
| profile_observation | Dated checking task | Market or identity label |
| contact_permission | Consent, relationship, opt-out | Registration or profile result |
Keep +383 exceptions in a phone queue
Preserve phone_raw, source_file and source_row before normalizing international notation. Do not prepend +383 to a phone that already has it. A row whose length or service range cannot be supported enters format_exception. It must not become “no age” or “unknown gender,” because that corrupts the denominator of the profile report.
Age and gender outputs are not self-declarations
AIPUSH WhatsApp Gender and Age checking can return phone, age, gender, avatar and a mapped WhatsApp phone. These are limited observations at check time, so retain checked_at and task version. Do not convert them into verified identity or infer ethnicity, religion, politics or citizenship.
Small markets create small cells quickly
Crossing source, location, age band and gender may leave only a few records in a cell. Set a minimum-cell rule before publishing: combine bands, suppress the split or retain only an overall aggregate. Person-level Excel should not be available to roles without a defined operational need.
Do not fill missing language with a guess
Albanian, Serbian or another language preference comes from user choice or a communication record. A name, avatar and country code cannot populate the language field. Unknown is an honest state, not a defect to be filled with the answer that seems most likely.
Place an isolation layer between TXT and CRM
The upload is TXT with one phone per line. Name, order, address, language and permission stay in an internal crosswalk. Returned Excel enters staging and reconnects through batch_row plus the submitted phone. Mapping disagreements receive human review instead of silently changing the CRM master number.
Attach a denominator to every reported figure
Show submitted, format-accepted, returned, unknown and aggregation-eligible counts separately. Comparing unknown rates by provenance is more useful than announcing one national percentage: it can reveal a stale partner file without presenting a data-quality defect as a trait of people in Kosovo.
Permission remains independent of profile data
Valid syntax, an observed WhatsApp profile or an age band does not create marketing permission. Confirm provenance, purpose, frequency and opt-out before messaging, and follow the WhatsApp Business Messaging Policy.
The acceptance question for this Kosovo guide
The handoff should identify what came only from +383, what came from business records, what came from a dated observation and what remains unknown. Maintaining those boundaries lets checking support controlled segmentation without manufacturing an unsupported identity dossier.
