Nicolas Padilla

Publications

• Probabilistic Machine Learning: New Frontiers for Modeling Consumers and their Choices
  Dew, Ryan, Nicolas Padilla, Lan E. Luo, Shin Oblander, Asim Ansari, Khaled Boughanmi, Michael Braun, Fred Feinberg, Jia Liu, Thomas Otter, Longxiu Tian, Yixin Wang, and Mingzhang Yin (2024)
  Accepted at International Journal of Research in Marketing
  [Show Abstract] [Download Paper] [Code Companion]
  

  Making sense of massive, individual-level data is challenging: marketing researchers and analysts need flexible models that can accommodate rich patterns of heterogeneity and dynamics, work with and link diverse data types, and scale to modern data sizes. Practitioners also need tools that can quantify uncertainty in models and predictions of consumer behavior to inform optimal decision-making.

  In this paper, we demonstrate the promise of probabilistic machine learning (PML), which refers to the pairing of probabilistic modeling and machine learning methods, in pushing the frontier of combining flexibility, scalability, interpretability, and uncertainty quantification for building better models of consumers and their choices. Specifically, we overview both PML models and inference methods, and highlight their utility for addressing four common classes of marketing problems: (1) uncovering heterogeneity, (2) flexibly modeling nonlinearities and dynamics, (3) handling high-dimensional and unstructured data, and (4) addressing missingness, often via data fusion. We also discuss promising directions in enriching marketing models, reflecting recent developments in representation learning, causal inference, experimentation and decision-making, and theory-based behavioral modeling. .

• The Customer Journey as a Source of Information
  Padilla, Nicolas, Eva Ascarza and Oded Netzer (2024)
  Quantitative Marketing and Economics
  [Show Abstract] [Paper] [Download Paper]
  

  We introduce a probabilistic machine learning model that fuses customer click-stream data and purchase data within and across journeys. This approach addresses the critical business need for leveraging first-party data (1PD), particularly in environments with infrequent purchases, which are characterized by minimal or no prior purchase history. Combining data across journeys poses a challenge because customers’ needs might vary across different purchase occasions.

  Our model accounts for this “context heterogeneity” using a Bayesian non-parametric Pitman-Yor process. By drawing from within-journey, past journeys, and cross-customer behaviors, our model addresses the “cold start problem,” enabling firms to predict customer preferences even without prior interactions. Notably, the model continuously updates the inferred preferences as customers interact with the firm.

  We apply this model to data from an online travel platform, revealing significant benefits from consolidating 1PD from both current and previous customer journeys. This integration enhances managers’ understanding of customer needs, allowing for more effective personalization of marketing tactics, such as retargeting efforts and product recommendations, to better align with customers’ dynamic preferences.

• Gender Differences in Preferences for Meaning at Work
  Vanessa Burbano, Nicolas Padilla and Stephan Meier (2024)
  AEJ: Economic Policy
  [Show Abstract] [Paper] [Download Paper]
  

  Scholars have examined whether preferences for job characteristics help explain why men and women sort into different occupations, but have overlooked preferences for meaning at work. We first document gender differences in preferences for meaning in a large-scale survey covering individuals in 47 countries. We then conduct a choice-based conjoint analysis of a cohort of MBA students at a leading business school to study gender differences in preferences for meaning compared to other job attributes.

  We show that gender differences in preferences for meaning at work are widespread and partly explain gender differences in behavioral outcomes, including industry of work.

• Overcoming the Cold Start Problem of CRM using a Probabilistic Machine Learning Approach
  Padilla, Nicolas and Eva Ascarza (2021)
  Journal of Marketing Research
  [Show Abstract] [Paper] [Download Paper]
  

  The success of customer relationship management programs ultimately depends on the firm's ability to identify and leverage differences across customers—a difficult task when firms attempt to manage new customers, for whom only the first purchase has been observed. The lack of repeated observations for these customers poses a structural challenge for firms to infer unobserved differences across them. This is what the authors call the “cold start” problem of customer relationship management, whereby companies have difficulties leveraging existing data when they attempt to make inferences about customers at the beginning of their relationship.

  The authors propose a solution to the cold start problem by developing a probabilistic machine learning modeling framework that leverages the information collected at the moment of acquisition. The main aspect of the model is that it flexibly captures latent dimensions that govern the behaviors observed at acquisition as well as future propensities to buy and to respond to marketing actions using deep exponential families. The model can be integrated with a variety of demand specifications and is flexible enough to capture a wide range of heterogeneity structures. The authors validate their approach in a retail context and empirically demonstrate the model's ability to identify high-value customers as well as those most sensitive to marketing actions right after their first purchase.

Working Papers

• Your MMM Is Broken: Identification of Nonlinear and Dynamic Effects in Marketing Mix Models
  Ryan Dew, Nicolas Padilla, Anya Shchetkina (2024) (Authors contributed equally)
  [Show Abstract] [Download Paper]
  

  Recent years have seen a resurgence in interest in marketing mix models (MMMs), which are aggregate-level models of marketing effectiveness. Often these models incorporate nonlinear effects, and either implicitly or explicitly assume that marketing effectiveness varies over time. In this paper, we show that nonlinear and time-varying effects are often not separately identifiable: while certain data patterns may be suggestive of nonlinear effects, such patterns may also emerge under simpler models with time-varying effects. Moreover, problematically, these two types of effects may suggest fundamentally different optimal marketing allocations.

  We examine this identification issue through theory and simulations, describing the conditions under which conflation between the two types of models is likely to occur. We show that conflating the two types of effects is especially likely in the presence of autocorrelated marketing variables, which are common in practice, especially given the widespread use of stock variables to capture long-run effects of advertising. We illustrate these ideas through numerous empirical applications to real-world marketing mix data, showing the prevalence of the conflation issue in practice. Finally, we show how marketers can avoid this conflation, by designing experiments that strategically manipulate spending in ways that pin down model form.

• Heterogeneity in HMMs: Allowing for heterogeneity in the number of states
  Padilla, Nicolas, Ricardo Montoya and Oded Netzer (2019)
  [Show Abstract] [Download Paper]
  

  Hidden Markov Models (HMMs) have been widely used in marketing to study dynamics in customer behavior. HMMs have been successfully applied to model how customers transition among diverse latent states such as attention levels, web search behavior, customer's relationships, and purchase intent. While most HMMs in marketing allow for heterogeneity in the model's parameters, these models assume that the number of latent states is common across customers. In this work, we analyze the potential bias of making such an assumption, assess to what extent heterogeneity in the model's parameters can mitigate the impact of such bias, and provide a mixture of HMMs model that relaxes this assumption.

  Using a comprehensive Monte Carlo simulation exercise and secondary data from an online role playing game, we demonstrate that ignoring heterogeneity in the number of states could lead to model identification problems and to erroneous interpretations of customer dynamics. In particular, we show that: (1) even when only a small proportion of customers have a larger number of latent states (as compared to the majority of customers), the best fitting model would be an expensive HMM in terms of number of states; (2) even when heterogeneity is accounted for in the HMM parameters, the inference from analyzing the population estimates, a common practice in the literature, can be biased; (3) even the individual-level estimates of customers with the correct number of states can be biased.

  We propose a mixture of HMMs with different number of states to account for heterogeneity in the number of states which captures well the behavior at both individual and population level.

Research in Progress

• Unified Marketing Measurement: Incorporating Experiments in Media Mix Models
  with Ryan Dew
  [Show Abstract]
  

  Digital marketers are struggling to measure campaign effectiveness due to the loss of customer-level tracking, rendering multi-touch attribution models obsolete. Moreover, constantly running experiments may be a costly alternative if effectiveness changes over time. As a consequence, firms have turned to using classic measurement tools like media mix models, which have always been built on potentially endogenous aggregate measures of campaign spend and performance.

  We propose a \textit{unified marketing measurement} (UMM) framework that allows us to measure time-varying marketing effectiveness. Methodologically, we use a modern Bayesian nonparametrics framework that fuses the (available) experiments with aggregate media-mix data and leverages the exogenous variation in experiments to de-bias a media mix model. Using Gaussian Processes, our model regularizes ad effectiveness over time smoothly, allowing the experiments to separate marketing effectiveness from the correlation between sales and ad spending close to the experiment.

  Our modeling framework also provides uncertainty quantification on ad effectiveness, which can be leveraged to determine if further experiments are needed. Using a series of simulations, we show the conditions for properly inferring ad effectiveness over time. We further show that endogeneity bias in observational data induces higher posterior uncertainty on the effectiveness and structural correlation estimates, which does not decrease with more observational data. This means we can use posterior uncertainty quantification to diagnose when additional experiments are needed.

• The Impact of Generative AI on Online Publishers
  with Anja Lambrecht, Tai Lam, and Brett Hollenbeck
  
• Understanding the Non-linear Customer journey: Identifying the "Moment of Truth"
  coauthored with Oded Netzer and Vicki Morwitz
  [Show Abstract]
  

  For decades, the traditional purchase funnel model was proposed as means to capture how consumers chose products, mainly on the consumer packaged goods category. The typical purchase process suggested that consumes advance through the funnel from awareness of the products to developing interest and consideration, to choice and purchase. The model was also extended by including a loop from the purchase to incorporate experience from consumption which builds loyalty leading to repeat purchases. For customers in typical offline retail settings, the decision process was straight forward: they placed themselves in front of the shelf, examined the brands, attributes and prices, selected a few products to consider and made a purchase decision.

  Nowadays, with the rise of online retailers and marketplaces, the assortment available for the consumer has increased dramatically, offering consumers a large variety of products targeted to fulfill heterogeneous needs, and high variation in pricing from the same products in different channels. Additionally, the ease of search and product comparison online has led to a reduction in search cost. This change has led consumers to engage more actively in the process of searching for a product. At the same time, customers are using new technologies that allow them to compare prices, sort products, filter results by features, and return to previously seen products to make the decision. Thus, the purchase process has moved from a mostly linear funnel to a complex (back and forth) journey in which consumers go iteratively between the steps of awareness, search, and consideration, gaining awareness for new alternatives and product features, adding and removing products from the consideration set, and revisiting previously considered products throughout the process, making decisions along the way, making the purchase funnel complex and highly non-linear.

  Although complex, unlike the possibly simpler and linear offline journey, the online journey is mostly observed to the researcher and the firm, making it possible to better map the journey and aim to understand the deliberate process consumers go through in choosing products or services. Customers compare products based on (mostly) observable features, making observable decisions along the journey such as searching, clicking on products, and analyzing products information.

  The purpose of this research is to investigate the online customer journey. First, we aim to map the non-linearity of the online customer journey. Although the journey is complex, different customers follow similar patterns of journeys, leading to predictive patterns of the new purchase journey. Second, we show that firms can use the new customer journey to predict how likely a journey is to finish with a purchase. For example, is a customer who searches for a flight and returns from the flights option page to make another search query more or less likely to eventually purchase a flight than a customer who went linearly search to flight options? On the one hand, the customer signals that they did not find the fight they looked for in the flights option page, but on the other hand, the customer indicates that they are engaged in the process. Finally, we investigate which steps or paths of the customer journey are most indicative that the journey is likely to end in purchase. In doing so we aim to identify the moment of truth, which is step in the journey that increases the likelihood of purchase the most.

  As an empirical setting to investigate the non-linear customer journey we use search and purchase data from one of the largest online travel agencies, analyzing data from flight search of over 106,245 customers, making 650,317 purchase ,journeys. We identify that customers indeed tend to go through a highly non-linear purchase funnel. On average a customer went through 5.6 steps in their customer journey, revisiting the query for a flight, changing the parameters of the search and clicking on multiple options in a highly non-linear fashion.

• Avoiding Fifty Shades of Grey: Uncertainty in Customized Assortment Optimization
  coauthored with Oded Netzer
  

• [Download Full Version]