Global Retailer's Regional Rollout Under Legal Constraint
A retailer with sites in the US, Germany, and Brazil wants to launch a homepage hero test in Q3. Legal has confirmed that Germany requires opt-in consent before any A/B test cookie is set, while the US and Brazil permit implied consent. Marketing wants a single global activity so results are comparable across regions. Traffic is 3.2M monthly sessions: 55% US, 30% Germany, 15% Brazil. The team has one Target implementation, no data layer yet, and a 6-week deadline. Analytics shows the German homepage has the highest revenue per session. The CRO lead proposes launching globally at 100% allocation on day one to "get significance fast."
1. Evaluate the CRO lead's proposal and recommend an activity architecture that respects the consent constraint while preserving comparability.
Read the worked answer
The 100% global launch is wrong for two reasons. First, it violates the German consent requirement - visitors who have not opted in must not be bucketed into the test, so a global 100% allocation would either breach policy or silently exclude Germany and break the "comparable" goal. Second, 100% allocation removes a control experience and eliminates the baseline needed to measure lift.
Recommended architecture:
- Keep one activity definition (same offers, same success metric) so the hypothesis and measurement are comparable.
- Use audience targeting to include only consented German visitors plus US/Brazil visitors; exclude non-consented DE via the consent flag in the data layer.
- Set allocation to a controlled split (e.g., 50/50 or 34/33/33), not 100%.
- Report Germany separately because its consented population is a biased subset (opt-in skews toward engaged users); do not pool it blindly with US/BR for the primary readout.
- Build the data layer before launch - the consent flag and revenue-per-session metric both depend on it.
2. The team wants to know whether one global activity can produce a valid decision for all three regions. What would you tell them, and what metric guardrails would you set?
Read the worked answer
A single global activity can produce a directional read but not a clean per-region decision, because:
- The German sample is restricted to consented users, so it is not representative of the German population.
- Revenue per session differs materially by region, so a pooled average will be dominated by the US (55% of traffic).
Guardrails:
- Pre-register the primary metric (e.g., revenue per session) and a secondary (conversion rate).
- Set a minimum sample per region before reading regional segments; a pooled win does not license a regional rollout.
- Treat Germany as a separate decision with its own significance threshold, and flag that its result applies only to consented traffic.
- Use a sequential or fixed-horizon test plan agreed in advance to avoid peeking-driven false positives.
If the business needs per-region decisions, run region-scoped activities sharing one offer library rather than forcing one pooled readout.