Article summary
Design an Eastern Europe Viber/WhatsApp pilot with preregistered hypotheses, shared sampling, randomized channel assignment and sequential stop rules—without double-messaging people.
An Eastern Europe Viber-versus-WhatsApp pilot should not message the same people on both channels. Its core is a hypothesis, sample, metric and stop rule written before sending. Number checking creates platform observations, random assignment reduces selection bias, and permission plus user preference controls who can enter the experiment.
Preregister a falsifiable hypothesis
For example: “Among recent voluntary inquiries in Romania who did not choose a channel, one human follow-up on Viber versus WhatsApp changes qualified reply rate by at least the predefined threshold.” Select the primary outcome and minimum meaningful difference in advance. Do not switch to clicks or a favorable country after seeing results.
Define inclusion and exclusion
| Rule | Example | Reason |
|---|---|---|
| Include | Traceable source, same purpose, current permission | Lawful comparable cohort |
| Exclude | Opt-out, current complaint, unknown language | Protect customer |
| Set aside | Mapping collision, bad format, platform unknown | Do not force into a group |
| Honor choice | Explicit Viber or WhatsApp preference | Not randomized |
Observe registration on the same phone set
Run Viber Registration and WhatsApp Registration against the same frozen normalized sample. Both corresponding AIPUSH tasks return phone and registration result. Preserve unknown, error and no-registration-observed separately instead of folding the first two into a negative.
Randomize channel, never permission
Random assignment is possible only when registration is observed on both platforms, no preference exists and both channels fit the original permission. Each contact enters one send arm. Consider a no-send or business-as-usual group if the organization needs an incremental estimate.
Keep treatment consistent
| Control | Method | Bias if ignored |
|---|---|---|
| Content | Same value proposition and local language | Copy effect looks like channel effect |
| Sender | Teams with equivalent skill | Human ability contaminates outcome |
| Time | Balance weekday and hour | Timing bias |
| Follow-up | Same steps and interval | Frequency changes complaints |
Use qualified reply as the primary outcome
A qualified reply advances service, clarifies need or explicitly declines. Delivery, opening and platform activity are not business success. Secondary outcomes include resolution, opt-out, complaint, human time and cost per useful conversation. Negative outcomes appear beside positive ones.
Apply sequential stop rules
At predefined sample points, review safety signals. Stop an arm immediately when complaints or opt-outs exceed a threshold. End for futility when the minimum meaningful difference is no longer plausible. Do not inspect daily and declare victory at a random lead.
Keep platform-specific fields out of the common endpoint
Viber Activity may include mid and offline time; WhatsApp Activity may include activity time, active days and mapped phone. Those can support platform-specific diagnostics, but the pilot’s main result should use customer-conversation outcomes measured fairly on both sides.
Handle crossover and spillover
If a customer assigned to Viber voluntarily moves to WhatsApp, record crossover and continue serving them. Analyze the primary result by original assignment and describe actual channel separately. Customer experience outranks statistical purity.
The result may not be one winner
Evidence may support routing by country, source or user choice, or show no incremental value for either platform. Preserve the sample flow, exclusions and uncertainty. A small win cannot be published as “Eastern Europe prefers one platform.”
A reproducible pilot package
Archive the preregistration, TXT hash, both task schemas, random seed, message versions, stop events and analysis code. A new market may reuse the experimental discipline, but it needs a new hypothesis and local content. Methods transfer; conclusions do not.
