Authors

Description

Intro [5 mins]. Describe the principles where CI acts, what the challenges of the industry are in measurements and decision making, and the need to use this statistical theory applied in real problems of different natures. Mercado Libre (MeLi) - pushes [8 mins]. Method: Classic experimental A/B testing. Describe the random partition of two known a priori groups into test and control, and the treatment of the test group and its subsequent measurement. Measurement: CUPED-CUPAC Describe the limitations in standard measurement (variance, bias) and why it is necessary to strengthen the measurement. Also comparison of both methods, and a real example of how they improve the precision of the measurement. Glovo - clustered [7 mins]. Method: Clustered. Describe the special use case of these experiments, when they appear naturally or by default, and the need to treat them appropriately to avoid erroneous conclusions otherwise. Measurement: Clustered Errors. Describe how clustered errors generalize the assumption of independence, and allow the calculation of standard errors in accordance with which the appropriate significance of the estimator is then inferred. Doordash - Bernoulli (7 mins). Method: Bernoulli. Briefly describe the Bernoulli use case, and its most common use cases. Measurement: DiD. Explain the problem of bias due to lack of control in the variability of observations and the context when the treatment is applied, therefore the use of comparisons with previous windows to improve the precision of the measurement. Closing pitch (3 mins). Summary of the impact that decision making based on the application of CI methods has on the tech industry, especially within exposed companies, and how this effect can be extrapolated to other industries. Current and potential challenges to continue scaling this knowledge.