Overview of my research:

I study topics related to marketing analytics, digital marketing, and customer relationship management (CRM). The problems I try to address can be divided into two major categories: (1) how to draw causal inferences on marketing decisions, and (2) how to inform better marketing decisions.

Job market paper:

Leveraging Large-Scale Granular Single-Source Data for TV Advertising

with Rex Du and Shijie Lu

R&R from Marketing Science

Show abstract

Automatic content recognition (ACR) technology, coupled with the wide adoption of smart TVs and set-top-boxes, enables the collection of highly granular TV viewership data at scale. The current research proposes a novel instrumental variable (IV) approach to estimate the causal effect of TV advertising by leveraging large-scale, granular single-source data that merges household-level linear TV viewership data with first-party purchase data. Our approach decomposes each ad exposure by each household into an endogenous component that is a confluence of the focal brand’s ad buy decision and the household’s TV viewing decision, and an exogenous component that results from the TV network’s quasi-random within-show ad insertion timing decision. We show how the exogenous component can be extracted empirically and serve as a valid and strong IV. We apply the proposed approach to investigate the impact of TV advertising on household food delivery ordering from a major online platform in the U.S. We examine the evolution of household ad responsiveness as a function of their purchase history and measure the same-day and 30-day cumulative ad impact. For the focal brand, the average same-day ad elasticity is 0.032, and household ad responsiveness varies as a function of purchase frequency and recency. Model-based simulations show that same-day ad response, carryover, and state-dependence account for, respectively, 43%, 36%, and 21% of 30-day cumulative ad impact; targeting based on purchase history enhances profitability, while targeting based on same-day impact can lead to suboptimal efficacy in terms of long-term impact.

Publication:

Leveraging Online Search Data as a Source of Market Intelligence

Foundations and Trends in Marketing

with Rex Du

Show abstract

Every year billions of users around the world submit trillions of queries through online search engines such as Google, Bing, Baidu, and Yandex. Over the years, aggregated and anonymized search volume data on keywords contained in all these queries have formed an epic database of human intentions that continues to expand every day. Thanks to platforms such as Google Trends, Google Ads Keyword Planner, Microsoft Advertising Keyword Planner, Baidu Index, and Yandex Wordstat, advertisers can readily assess search engine users’ collective interests over time and across geographic areas to optimize their search engine marketing efforts. In this monograph, we illustrate how online search volume data, indexed or otherwise, can be leveraged as a powerful source of market intelligence for purposes beyond search engine marketing. We do so by offering a brief tutorial on Google Trends and Google Ads Keyword Planner, two popular (and free) platforms for gathering online search trend and volume data, respectively. We review prior studies that have examined the use of aggregate online search data as 1) predictors for nowcasting and forecasting, 2) dependent variables in market response modeling, and 3) proxies for hard-to-measure constructs. In each of these three areas, we provide specific examples of applications to illustrate the power and versatility of online search data. We conclude by offering several ideas for future research where we see the full potential of online search data is still to be uncovered.

Working paper:

Measuring Targeting Effectiveness in TV Advertising: Evidence from 313 Brands

with Samsun Knight and Yakov Bart

R&R from Journal of Marketing Research

Show abstract

We estimate the heterogeneous effects of TV advertising on revenues of physical retail stores and restaurants in the United States for 313 brands using a novel panel of store-level revenue data and a two-way fixed effects design. We find a mean revenue elasticity to TV advertising of 0.094 and a median elasticity of 0.044, along with a significant estimated S-curvature in the marginal effect of advertising. We document significant heterogeneity in estimated effective- ness across store-level characteristics, and in particular find that advertising is more effective for stores in denser areas and for stores in areas with higher numbers of competitor locations. We then use these heterogeneity estimates to construct optimal allocations of ad expenditure across DMAs and project that these counterfactual reallocations would increase returns on advertising by a median of 4.3% and a mean of 22.1%. This study advances recent research demonstrating that TV advertising is less effective than is generally assumed by highlighting the role of suboptimal geographic targeting and by quantifying how much realized effectiveness can understate advertising’s potential effect. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4638083

Assessing the Potential of Addressable TV Advertising

MSI working paper

with Rex Du

Show abstract

Addressable linear TV allows advertisers to target individual households watching the same linear programming with different ad insertions. This study assesses the potential lift in ad efficacy of addressable linear TV over traditional linear TV. We calibrate an ad response model by matching, for a panel of 731,393 households over 15 months, second-by-second linear TV viewing data with conversion data from an online vendor of personal financial information and services. Our model estimates (1) the same-day and one-month carryover effects of the focal brand’s TV ads, and (2) two dimensions of heterogeneity—the likelihood of a household being in the focal market, and conditional on being in the market, its responsiveness to the focal brand’s TV ads. Using the calibrated model, we predict the average incremental conversions per one thousand ad impressions that could result from alternative targeting strategies. Simulation results suggest that there can be a substantial lift in efficacy when traditional TV advertisers switch some ad buys to addressable TV and target households with the highest incremental conversion potentials. This lift in ad efficacy gets amplified when a larger portion of linear TV ad inventory becomes addressable or addressable TV ads are deployed for a lower level of reach.