Case Study 12:
Analytics Challenges

Challenge of finding better simulating methods

Consultants of DigitCompass LLC worked on Joinpoint Regression Program utilizing the Joinpoint regression model to analyze the trends of cancer rates. One important of measurement of trends is the confidence intervals of trends in a time period. But the confident intervals by current methods at that time were too wide, sometime meaningless. Our consultants proposed an innovative simulation method, ‘Empirical Simulation Method’, which generates the best confident intervals compared with the current methods at that time. The work was summarized in the paper ‘Improved Confidence Interval for Average Annual Percent Change in Trend Analysis’ published in Statistics in Medicine 2017.
Challenge of insufficient data

Consultants of DigitCompass LLC had a challenge of building call-impact models with insufficient data when helping customers allocate resources for solicitation calls. The customers have clients and want to sell a product to the clients. But resources are limited and not every client can get a call. Resources must be allocated to those clients who have largest difference between the chance of responding to the solicitation call and the chance of buying the product without the solicitation call. The call-impact models actually evaluate the impact of a marketing operation (call) on the clients’ behavior. The challenge is that the available data have records of clients who either did not receive the call or received the call. Splitting data into two segments receiving calls or not receiving calls leads to poor model performances. Our consultants proposed a latent-factor method that efficiently leverages the available data and enhances the performance of the resources allocation by 11%. The work is summarized in the paper ‘A method of evaluating impact of business actions and its implementation’.

Challenge of computation-intensive problems

While working on the research of cancer census tract, consultants of DigitCompass LLC had a challenge of computation-intensive problems. It took months to get simulation results for the study. Instead of sticking to the traditional technique, our consultants proposed a new method of high performance computing, the general parallel computing model (GPCM), and implemented it in R platform called ‘parallelcomp’. With the software, ‘parallelcomp’, the simulation results of cancer census tract only took hours. Efficiency was improved hundreds of times. Our consultants utilized innovative approach and successfully solved the challenge.