A/B Testing at Vungle Case Study

Added on - 21 Apr 2020

  • 12


  • 848


  • 129


  • 0


Showing pages 1 to 4 of 12 pages
A/B Testing at Vungle CaseStudy
The company named Vungle is a firm that carried out the production ofvideo advertisements. In its initial stage, the company during 2011 mademoney by the funds borrowed as investments from Jaffer's girlfriend, who ishis future wife and from his business professor. The two funds borrowedwere considered as the investment, which ranged to $15 thousand fromeach person. Later in 2012, along with this investment, the founders utilizedtheir own creative video production technology for grasping the attention ofthe San Francisco based start up incubator named, AngelPad. This programprovided opportunity for the Company with $120,000 in its seed funding.The first challenge for Jaffer revolves around the installation rates bythe mobile device users, the publishers, the advertisers and finally theplatform which matches with the user's choice. The other set of challengeincludes meeting the demands of the advertisers who are interested topurchase and supply the advertisements. Then, attracting and increasing theeagerness and interest in the users is quite challenging. Because, eagernessand interest is what that allows the users to increase the installation rates.Further, the evaluation of the two algorithms namely, Vungle's presentAlgorithm and the data science approach, is also challenging for Jaffer thatdetermines which algorithm serves best.According to the Mobile in-app advertising funnel, the process includesrequests, impressions, completes, clicks and installs. By evaluating the twoalgorithms i.e, the existing algorithm and the new data science the valuesdetermine that, in the new data science, Impressions, Completes and Clicks2
must to be improved. This improvisation will help the new data sciencealgorithm to be even better.H0-Frequency distribution method determines effective resultsby comparing A and B algorithms.It provides only the frequency information for comparing the twoalgorithms. Thus, it is not an effective approach.H1- Normal distribution method provides better comparisonresults from the A/B test.Whereas, the normal distribution approach provides mean, varianceand standard deviation for comparing the two algorithms. Thus, it canprovide effective result.The assumptions which underline the analysis are illustrated below.5,500,0006,000,0006,500,0007,000,0007,500,0008,000,0008,500,0009,000,0009,500,00005101520253035HistogramUnnamed Series 1ImpressionsPercent3
Figure: A-Condition for Impression("Frequency Distribution of DataUsing Megastat", 2017)420,000440,000460,000480,000500,000520,000540,000560,000580,000600,0000510152025303540HistogramUnnamed Series 1ImpressionsPercentFigure: B-Condition for Impression5,000,0005,400,0005,800,0006,200,0006,600,0007,000,0007,400,0007,800,0008,200,0000510152025HistogramUnnamed Series 1CompletesPercentFigure: A-Condition for Completes4
You’re reading a preview

To View Complete Document

Become a Desklib Library Member.
Subscribe to our plans

Download This Document