New paper @ QME! Mobile Geo-Tracking data and its privacy implications for predicting retail visits

Not all papers are created equal. And their fate — where they end up in the review process, how long they take, how they evolve — is not always linear or predictable.

This is another post in my “journeys of papers” series (the first was on my job market paper’s eight-year journey). My hope is to make my experiences with academic research and the publication process more transparent, especially for junior faculty and PhD students. You are not alone.

Here’s the paper:
Link: Privacy and prediction: how useful are geo-tracking data for predicting consumer visits?

And since videos can sometimes speak louder than words, here’s a short overview of the findings as I teach them in my MBA courses:
Link: Video summary

My Toggl time tracker tells me that just this paper was over 1700 hours in the making (for the first versions between 2020 and 2023). But there is more to a paper than the hours and the outcomes. This blog post is the story behind the paper — a story of pivots. 

Early ideas and a chance encounter

Spring 2018. I was a PhD student at Texas A&M, teaching an undergraduate course in market research. One rainy afternoon, during office hours, a woman walked in with boxes of pizza. Their career event on campus had been cancelled, so she stopped by instead. She turned out to be the CEO and founder of a mobile app company focused on safe driving — rewarding people for driving without looking at their phones.

We kept in touch. That summer, I began a sort of “pseudo-internship” with her team, helping shape their data strategy. At that time, they weren’t even collecting location data, though users had opted in. I nudged them to think of data not just as a by-product, but as part of the product — something that could power dashboards, partnerships, and insights for clients. Eventually, they launched a premium membership built on location intelligence, like comparing foot traffic at competing restaurants.

In parallel, I had grown fascinated with IO courses and demand estimation frameworks in my own PhD coursework. Once I had an NDA signed, had sufficient data collected, and was able to check the quality of the data, I approached and started working with my co-author Fernando Luco at Texas A&M econ. Like me, he was interested in new methods and bringing Machine Learning to demand frameworks.

Very soon, we realized that the app data alone won't suffice for a research project of the scale we were planning. So we brought in Safegraph, Yelp, Data Axle, ACS, and at one point -- even menu prices data -- to the mix. 

Very soon, we also realized that a project predicting users' visits to retail stores alone won't suffice. I am a mobile-first researcher. It wasn't until our discussions much later--when the prediction problem had taken shape--that we decided to bring in the privacy perspective. Looking back, this was the natural step.

Pivot 1: From data to research question

Initially, the project was about predicting where individual users would go — their individual visits to restaurants and retailers — over a two-month period. Privacy was framed as tracking frequency: more frequent location pings meant less privacy.

But as we gathered feedback, we realized two things:

1. Restaurants care more about aggregate demand, especially about fluctuations in demand, than individual visitors.
2. Privacy concerns aren’t just about frequency — many consumers care about being tracked at all or what data are tracked about them.

So, we pivoted. We shifted to predicting weekly demand fluctuations at the restaurant level, and expanded our privacy lens beyond tracking frequency to include restrictions on what and where data are collected. 

Pivot 2: From raw coordinates to meaningful features

Working with geo-location data was another learning curve. At first, I thought raw latitude-longitude points, fed into advanced models like LSTMs, CNNs, or Transformers, would improve predictive power. It didn’t. What mattered were the features we engineered — clusters, behavioral summaries, and patterns from movement data. Complexity alone didn’t improve predictions. Thoughtful feature engineering did. That lesson has stayed with me: conceptual clarity beats fancy models esp. when the signal is messy.

Pivot 3: Managerial decision making

We received criticism that the usefulness of geo-tracking data won't translate to managerial decisions. We were asked if there was a way we could quantify how managers will tweak their strategies and decisions in response to the use of geo-tracking data. This is one critique we perhaps couldn't fully resolve due to data limitations on not observing managerial decisions. However, we collected revenue, cost, and profit information based on industry reports and the paper now shows about 5% of restaurant revenues can potentially be "saved" with using these geo-tracking data. 

The first version of the paper went to the Journal of Marketing Research (JMR). It survived through three tough rounds, even an editor change, but ultimately didn’t make it. My understanding is this was mainly due to the last point of not observing managerial decisions, something we couldn't have resolved. We simply had no data on how managers used this information other than the interviews we conducted and the conversations with the app platform. 

That third-round rejection stung. The timing was awful, especially since I had just presented the paper at European schools, receiving overwhelmingly positive feedback (privacy being even more salient post-GDPR) and got the rejection on my flight home. But the truth is: some limitations cannot be resolved, and not every paper belongs everywhere.

At Quantitative Marketing and Economics (QME), the editors encouraged us to write the paper we wanted to write. The process was efficient and direct. I am deeply grateful to the editors for seeing value in the contribution as it stood i.e., predictive power of geo-tracking data, and the trade-offs introduced by privacy restrictions.

Looking back, this paper carries many “firsts” for me. My first project co-authored with someone outside of marketing and also a junior faculty at the time. My first collaboration with a company relationship I had built from scratch. My first foray into mobility data and not just where people shop, but how they move. I learned SO MUCH about modeling consumer location footprints and connecting them with online data.

The paper also carries important lessons: Relationships matter. That serendipitous meeting in 2018 shaped years of work (even as the firm went through major ownership changes in 2020). Insights matter more than raw data. Pivots are part of the process. Whether in methods, framing, or journals, flexibility keeps research alive and relevant.