Article summary
Report usable, unknown and provenance gaps together so a Tanzania +255 WhatsApp demographic sample is not mistaken for all customers.
When only part of a Tanzania +255 list receives gender and age observations, the first question is not the percentages among visible rows. It is why everyone else never entered the visible sample. Ignoring unknowns creates coverage bias.
Draw the coverage funnel first
| Stage | Count | Explanation required |
|---|---|---|
| Raw submitted | N0 | Provenance and permission |
| Format-processable | N1 | Country evidence |
| Task returned | N2 | Exceptions and unknowns |
| Profile usable | N3 | Scope of use |
Compare dropout by acquisition source
Calculate N3/N0 separately for website forms, stores, partners and legacy CRM. A low source ratio calls for collection or format repair before demographic analysis.
Do not fill city or region from the phone
+255 supports a numbering plan but does not prove a specific residence. Never reverse-infer urbanity, region or community from an avatar, name or visible result.
Demographics are not identity verification
Age and gender fields are limited observations that need observed_at and confidence. They cannot authenticate a person or overwrite information the customer provided.
Split unknown into causes
Count format_exception, not_observed, visibility_limited and task_error separately. This tells the team whether the remedy belongs to data operations or source collection.
Do not present only attractive percentages
Publish denominator, missing rate and provenance composition together. Mark a small or clearly biased sample as unsuitable for population inference.
Separate TXT and analysis layers
AppShai accepts a TXT file with one phone per line. Provenance and permission remain internal; returned Excel lands in analytical staging before any purpose-specific aggregate.
Prefer signals closer to the business question
For content choice or service planning, explicit interest, language and order stage are usually more interpretable than inferred age and gender.
The Tanzania decision line
Profile output becomes analytically useful only when the team can explain who was observed, who was not and why. Otherwise it is merely a visible fragment.
