Article summary
After India +91 mobile cleanup, do not treat avatar hit rate as customer completeness; audit nulls, bias and expiry across language, channel and urban/rural provenance.
The main quality question in an India WhatsApp Profile Picture check is not whether the hit rate looks high. It is which phones are more likely to expose an avatar and which cohorts become nulls. When a sample spans states, languages, urban and rural channels and acquisition years, one aggregate percentage hides structural missingness and tempts a content team to mistake visibility for a customer trait.
Constrain the input to a mobile-phone problem
Official Indian Department of Telecommunications material uses country code +91 and documents mobile dialling and full mobile number portability. A WhatsApp cohort should validate current mobile ranges and ten-digit mobile phones rather than mixing fixed lines, short codes, service numbers or M2M identifiers into one task.
Leave transformation evidence in the preparation table
| Column | Meaning | Purpose |
|---|---|---|
| phone_raw | Source-system text | Recovery and audit |
| phone_e164 | Rule-passing +91 candidate | TXT input |
| source_state | State/channel/form cohort | Stratified analysis |
| rule_version | Range-table version | Handle updates |
A prefix no longer proves current operator or region
Mobile number portability makes prefix-based claims about the current carrier unreliable. A phone also does not prove the holder’s current state, first language or urban/rural residence. If carrier or location is genuinely needed, obtain it from a separately authorized source rather than completing it from prefix or avatar.
Read the four Profile Picture outputs separately
AIPUSH WS Profile Picture may include phone, business-account observation, avatar and mapped WhatsApp phone. Business status is a platform observation, not company certification. Avatar is visibility at a time, the mapped phone can collide, and phone is a task key. None independently verifies a human identity.
Build a representation matrix by provenance
| Stratum | Compare | Response to difference |
|---|---|---|
| Stated language preference | Avatar unknown rate | Never infer language from image |
| Urban/non-urban source | Visibility and collision | Limit generalization |
| Online/offline channel | Phone exceptions and nulls | Repair collection process |
| Acquisition year | Expiry and mapping change | Apply data lifetime |
Do not infer caste, religion or human value from an image
Name, clothing, religious symbol, background and facial appearance cannot reliably establish caste, religion, ethnicity, health or economic capacity. Even if a task returns a skin-tone-related field, restrict it to an approved aggregate quality study; never build individual scores or exclusion rules from it.
Minimize TXT and quarantine Excel
The upload is TXT with one phone per line and no name, state, language, order or permission. An internal crosswalk retains contact_id and source stratum. Exported Excel first enters staging, mapping collisions never auto-merge, and image references receive narrow access roles and a short retention period.
Marketing experiments start with user choice
Use the person’s chosen language and content preferences, while permission and opt-out outrank any avatar observation. If an approved content test is run, use sufficiently large aggregate cohorts, neutral creative and explicit stopping rules, reporting unknowns. Never target inferred sensitive identities to increase conversion.
When a recheck is justified
Recheck only the necessary small cohort near actual use while purpose and permission remain valid. Avatars and mappings change quickly, so expired values stop being used or are deleted. Do not repeatedly collect every contact merely to maintain an appearance of profile completeness.
Acceptance centers on a bias statement
The final report lists inputs, exceptions, duplicates, avatar observed/not observed/unknown, business-account coverage, mapping collisions and checked time by provenance stratum. It also states which cohorts are underrepresented and which inferences are prohibited—more useful evidence than one hit-rate number.
