Article summary
Use counterfactual questions to test whether Botswana +267 WhatsApp demographic segmentation creates unjustified differences and preserve small-sample unknowns.
Before using Gender and Age fields on a Botswana +267 list, ask whether swapping those fields would change the service without a defensible reason. This counterfactual check reveals when segmentation is becoming differential treatment.
Four counterfactual questions
| Question | Risk signal |
|---|---|
| Would price change? | Differential pricing |
| Would eligibility change? | Exclusionary decision |
| Would service quality change? | Unfair routing |
| Would unknown receive the worst outcome? | Missingness penalty |
Narrower possible uses
Controlled aggregation, content-coverage review or sampling-bias investigation may be considered. Identity, credit, employment, insurance and critical eligibility must not use these observations.
+267 is not the complete market
Separate numbering plan from customer market. A foreign phone can belong to a Botswana service customer, while +267 does not prove present location.
Inspect provenance distribution first
Compare unknowns across first-party forms, partners and legacy CRM. If one source dominates, do not generalise the result to a country population.
Use coarse output for a small sample
Combine age bands, preserve unknown gender and publish only an overall count below the minimum cell. Precision is not a synonym for quality.
Isolate TXT from decision fields
Upload one phone per line while eligibility and permission stay internal. Returned observation cannot write directly into a decision system.
Give humans a veto
Any low-risk segmentation using profile observations needs visible reasons, correction and an override path.
Record the option not taken
If the counterfactual test fails, document why the team used a non-profile queue. That is governance evidence, not project failure.
The Botswana value
The page does not teach finer human labelling; it helps decide whether a segmentation should happen at all.
