CML Research Focus Areas

We develop next-generation research and ideas on topics such as algorithmic pricing, computational advertising, and online experimentation.

Algorithmic Pricing

Algorithmic pricing leverages data, algorithms, and economic theory to develop strategies for effective pricing.

Effective pricing requires bringing together careful measurement that isolates the causal effects of prices, well-articulated models of consumer demand response; consideration of heterogeneity in responses across units, and well-posed frameworks of competitive firm interaction.

Research in the lab combines experimentation and quasi-experimental methods to address the causality; economic and psychological theories of choice to understand consumer response; statistical and machine learning methods to uncover differences across units; and game theoretic models to assess competition.

Combining these, we analyze, measure, and develop pricing strategies in diverse areas such as multiproduct pricing in e-commerce, advertising pricing in digital ad-auctions and dynamic pricing and promotions in retail.

Featured Research

This handbook chapter reviews the empirical marketing literature analyzing diffusion and pricing over the product life cycle. It discusses how recent work has linked outcomes to micro-foundations and accommodated a role for forward-looking consumers and firms. It emphasizes a more nuanced perspective of the product life cycle that has emerged in the literature, as an endogenous outcome arising from the interaction of preferences, expectations, costs, and competition in the market, rather than as an exogenously specified process against which marketing strategies should be optimized.

Computational Advertising

Computational advertising pertains to the design, allocation, measurement, payment, and management of modern digital advertising using data-driven and algorithmic methods.

Research in the lab addresses several aspects of computational advertising including the economic and psychological mechanisms by which advertising works; how advertising targeting can be implemented at scale; effective mechanisms such as auctions to allocate ads and to facilitate payments; how advertising-driven business models work in technology markets; and the use of sophisticated experiments to measure the causal effects of digital advertising and to develop effective attribution strategies.

We research all aspects of digital advertising including display, search, social, and television markets. Apart from online advertising, the lab is also interested in studying offline direct-to-consumer advertising such as in-store advertising and business-to-business advertising such as that mediated via salespeople. Also of interest is the interaction between modern advertising markets and privacy, and the design and functioning of the digital advertising ecosystem composed of advertisers, publishers, and intermediaries such as demand and supply-side platforms and ad-exchanges.

person on computer
Featured Research

Advertisers that engage in real-time bidding to display their ads commonly have two goals: learning an optimal bidding policy and estimating the effect of exposing users to their ads. By framing the problem as a multi-armed bandit, we develop a Thompson sampling algorithm that learns the optimal bidding policy and estimates the expected effect of displaying the ad while minimizing losses from sub-optimal bidding. Simulations show that the proposed method accomplishes the advertiser’s goals at a much lower cost than conventional experimentation policies aimed at performing causal inference.

Online Experimentation

Online experimentation research at the lab explores strategies for leveraging experimentation at scale to resolve uncertainty in complex domains and to make better decisions, particularly on digital platforms.

One area of interest is to combine experimentation with economic theory to address business problems of interest. Another area of research is how experimentation can be implemented in online settings where data are arriving sequentially and exploration and exploitation of information for decision-making occurs simultaneously, such as in online advertising and pricing settings. For this, we leverage algorithms such as multi-arm bandits and other reinforcement learning methods and study how they can be engineered and implemented in technology-driven organizations. We are also interested in the broader question of how organizations can become better at measurement of their innovations, strategies, and processes and how these translate to better business outcomes.

Featured Research

Dramatic improvements in the ease of experimentation in the digital era has increased the propensity to experiment on platforms. As a consequence, it is common for many firms to be experimenting simultaneously, and for consumers to be simultaneously in many experiments. How should experimentation and policy be developed in such environments? To do this, this paper studies in detail the interpretation, analysis, and estimation of treatment effects when many experiments are running in parallel and applies it to the problem measuring the effect of digital advertising campaigns.