Research
Research interests
Substantive: Digital Marketing, Quantitative Marketing, Unstructured Data
Methodological: Causal Inference, Econometrics, Natural Language Processing, Interpretable Machine Learning and Deep Learning
Working papers
Can Lower Expert Opinions Lead to Better Consumer Ratings?:The Case of Michelin Stars (Job market paper).
With Yiting Deng, Puneet Manchanda, and Bert De Reyck.
Reject & Resubmit at Management Science
Abstract: Expert opinion exerts tremendous influence on the purchase journey, but its effect on overall consumer experience is ambiguous as it can give rise to both "expectation" and "reputation" effects. This paper explores the effect of expert opinions on consumer experience via the lens of consumer reviews in the restaurant industry, where the expert opinions are conveyed by Michelin stars. We construct a unique data set based on the Michelin Guide for Great Britain & Ireland from 2010 to 2020. The data include consumer reviews on TripAdvisor for all restaurants that were awarded Michelin stars during these 11 years, and a large pool of potential control restaurants. We apply two synthetic-control-based methods to estimate the effect of Michelin star changes on the sentiment and content of consumer reviews. We find that decreases in Michelin stars improve consumer review ratings, suggesting that the expectation effect of expert opinions is stronger than the reputation effect. The analysis of review content further shows that service and "value for money" appear to be the key drivers of the customer experience. When a restaurant loses or receives fewer Michelin stars, consumers become less demanding on service aspects and also focus less on value considerations. We discuss the implications of our findings for restaurant managers, the Michelin Guide, and other businesses that provide experience goods.
With Viviana Culmone, Bert De Reyck, Onesun Steve Yoo.
Under review at Production Operations Management
Partly funded by UKRI Innovate-UK
Abstract: Conducting a spend analysis of a procurement practice is a challenging task for manufacturers. It requires making sense of large-scale spend data in the form of unstructured texts and identifying savings opportunities. This process relies on procurement expert’s know-how and is often performed manually, a laborious task often leading to missed savings opportunities. We propose a three-component classification model to automate spend analysis and replicate the expert's know-how. It utilizes "small data" of detailed hierarchical taxonomy guides and "big data" of supplier information collected from various sources to train classifiers. It improves classification accuracy by cleverly combining top-down classifier that classifies a supplier’s general categories with bottom-up classifier that classifies its specific product-level categories. Finally, our decision support tool performs a Kraljic analysis to identify the product categories with the highest savings potential, and helps recommend specific suppliers to seek savings. Using the spend data from Cranswick plc, a major food producer in the UK, we test the accuracy of our methodology and show superior performance compared to benchmark models. Simulation of implementation estimates that automation of spend analysis contributes to £16-22m in annual savings for Cranswik plc.
Working in progress
Convolution Neural Networks for Scheduling Streaming Ads.
With Yiting Deng and Puneet Manchanda.
Data analysis in progress
Collaborated with a streaming ads company