A/B Test — Registration Experience Activation

Flow B won. Automatic INE scanning reduced friction, decreased CURP entry errors, and increased the activation completion rate. This result determined the final registration flow design. Manual entry remains available as a fallback option but represents a minimal share of total accesses.

Registration data showed high dropout and error rates in the activation flow, with no evidence of which approach would be most effective for new users.

Minimum sample of 1,000 unique users per variant for statistical significance. Key metrics: registration completion rate, average completion time, data entry error rate, and INE scan adoption rate in Flow B. Duration: 4-8 weeks. Instrumenting the test required creating new analytics events as part of the migration from Google Analytics 3 to GA4, internalizing the event creation process. I participated in building the tracking dashboard in Looker via BigQuery; the Data team later migrated the infrastructure to DBT.

A/B test between Flow A (manual form: phone + CURP with optional INE scan) and Flow B (INE scanning as the first step to auto-fill CURP). 50/50 distribution managed via Firebase Remote Config, limited to first installs during the test period.

1. Justified friction reduces errors: making INE scanning the first step (Flow B) seemed more complex, but it eliminated CURP typos and increased completion rates. Perceived complexity does not equal real friction.

2. Firebase Remote Config as a design tool: distributing the A/B at 50% from remote configuration (no code changes per variant) was the decision that made the experiment viable. Without this pattern, the test would have required two development sprints.

3. Creating analytics events is part of the design sprint: GA4 instrumentation was key to measuring error rates by field type. Without that data, it would have been impossible to confirm which variant won and why.

1. Propagate the INE scanning pattern to other document collection flows — proof of address and CLABE are the immediate candidates where autocomplete could reduce errors.

2. Measure the impact of the GA3-to-GA4 migration on event consistency: the dashboard was built in Looker on BigQuery and later migrated to DBT; validate that registration flow events remain consistent post-migration.