Enlighten Data Story: iFood Performance

01. Study Objective

The primary objective of this study was to evaluate the statistical efficacy of a newly proposed marketing campaign (Campaign B) against the existing baseline campaign (Campaign A). In data-driven decision making, relying solely on absolute conversion rates can lead to Type I errors (False Positives). Therefore, a rigorous A/B testing methodology was implemented to determine if the observed differences were statistically significant or merely due to random variance.

02. Methodology & Data Pipeline

The analytical pipeline was constructed using a dual-stack approach to ensure maximum accuracy and data integrity:

• Data Preprocessing (Python): Raw datasets were ingested, cleaned, and transformed using pandas and numpy. The observed sample size (n=333 valid conversions) was measured against the theoretical sample target (n=339, calculated at a 95% Confidence Level and a 5% Margin of Error), achieving a highly representative 98.2% adherence.
• Modeling & Testing (Power BI & DAX): The curated dataset was imported into Power BI. Custom DAX measures were engineered to dynamically calculate the Conversion Rates and execute the Z-Test for proportions, yielding the final P-Value.

03. Results & Strategic Recommendation

The statistical analysis yielded a P-Value of 0.14. Because this value is strictly greater than the established alpha threshold (0.05), we failed to reject the null hypothesis. In business terms: there is no statistical evidence to assert that Campaign B outperforms Campaign A.

Furthermore, multivariate analysis revealed that demographic factors—specifically Income Level and Educational Attainment (e.g., PhD cohorts)—were substantially stronger predictors of conversion than the ad creative itself.