For some time now, we had been eagerly awaiting a research talk from Eva Ascarza, the Jakurski Family Associate Professor of Business Administration from the Harvard Business School, who finally visited us last week.

Eva presented her recent research paper written where with her co-author, they attempt to provide a framework which helps to address and overcome the issue of ineffective long-term targeting when only insufficient information based on short-term signals is available. The paper shows that due to accumulation of unexplained variation, the usual approach of combining experiments and historical customer data to predict targeting can become inadequate, if not harmful. Instead, the authors propose a novel method which they dub separate imputation. This solution replaces the signals with simplified noise-reduced proxy of the outcome variable in CATE (Conditional Average Treatment Effect) models and helps to diminish uncertainty of making targeting predictions. It also outperforms alternative methods that use and as a benefit, does not require vast amounts of the customer data.

It goes without saying that we are very excited and grateful for enriching talks by researchers of Eva’s caliber and are looking forward to more visits in the near future. Thank you, Eva!