Skip to Content

Data-Driven Insights - New interpretable AI model aims to improve mall performance

Malls are important physical platforms for stores and generate trillions of dollars in transactions and asset value, but competition from online businesses has led to declines in customer traffic. Many malls are closing, affecting rent revenue.

However, those that incorporate new technologies and AI tools are growing and opening branches in new cities, says Yu Jeffrey Hu, who holds the Accenture Professorship of Information Technology at Purdue’s Mitchell E. Daniels, Jr. School of Business.

In “Evaluating the Effectiveness of Marketing Campaigns for Malls Using a Novel Interpretable Machine Learning Model,” Hu shows how a novel interpretable AI model is being designed to improve mall performance using AI-enabled sensor technology. The study, which was published in the journal Information Systems Research, was coauthored by Tong Wang from Yale University’s School of Management, Cheng He from the Wisconsin School of Business, University of Wisconsin-Madison, and Fujie Jin from Indiana University’s Kelley School of Business.

Yu Jeffrey Hu“The key metric to gauge the performance for malls is customer traffic, which is strongly correlated with mall revenue and is the main index determining the rental prices for tenant stores to lease spaces in malls,” Hu says. “Malls invest in various ways to increase traffic, including regular maintenance, renovation, and most importantly, running a number of marketing campaigns throughout the year that have proven to be effective in attracting customers to visit malls. These include promotional events around the holiday time, themed events with special decorations, or even entertainment events such as concerts.”

However, data limitations have constrained studies on marketing performance for malls, with studies mostly focusing on a single store or a specific type of campaign. According to Hu, there remains a research gap in examining the performance of different types of campaigns for brick-and-mortar malls. “It is of high practical relevance and research value to explore how new granular data and novel prediction models can be applied to improve campaigns,” he says.

The study uses a unique AI model to understand the relationship between campaign budget and customer traffic in malls using AI technology. Data was collected from 25 malls spanning two years using AI-chip-embedded sensors. Campaigns were then categorized by approach and timing based on previous literature and conversations with industry leaders. Weather, temperature and seasonality information was also combined to compile a rich list of features for analysis. Named a generalized additive neural network model (GANNM), this AI tool is designed to accurately measure marketing campaign effectiveness and present interpretable insights on how customer behavior changes depending on the campaign budget.

“The malls in our data typically schedule all campaigns for the coming year and decide upon the timing, approach and budget by the end of the year,” Hu says. “These campaigns vary in duration, ranging from one or two days to several weeks, and the campaign schedule also varies across malls, so we can use the variations across malls to evaluate the impact of different types of mall campaigns.”

A new model

The new model, GANNM, achieves better interpretability and higher prediction accuracy in characterizing how the impact of budget spending on customer traffic differs across campaign types. In addition, the optimal budget allocation derived from GANNM improves customer traffic lift by 11%, a significant improvement compared with the 3% lift from the baseline interpretable model. The findings provide helpful guidance for managerial practices, indicating that malls should focus more on campaigns that enhance customers’ shopping experience, especially during the off-peak period. Likewise, campaigns that enhance customers’ shopping experience with omnichannel retailing could be more effective in increasing customer traffic than campaigns with sales-only incentives.

“For example, the French beauty brand Sephora provides customers with hyper-personalized consultations based on the data from their skin analyses and shows AI-generated makeup trends and tutorials on the latest beauty looks with product recommendations. Sephora stores also offer beauty services and exclusive classes,” Hu says. “To expand offline, the London fashion store Missguided created a flagship store inspired by a TV studio, with huge screens that stream customer-generated social media content. Our results provide empirical evidence showing that this recent trend of employing novel approaches for enhancing customer experience in physical stores can effectively encourage customers to visit malls.”

While most of the campaign budget of a mall is typically allocated to peak period campaigns, the study suggests that the marginal benefit is smaller than during off-peak period campaigns. “This implies that mall managers should increase marketing spending to areas that are likely overlooked and avoid over-crowding the budget to campaigns during times with high levels of competition that are likely already over-marketed,” Hu says.

Adjusting marketing campaigns

With the continuing rise of e-commerce, the findings also have implications for how malls should adjust their campaigns to counter competition from online businesses, suggesting that online promotions could create opportunities for offline businesses. According to Hu, sufficient investment in campaigns to raise customer awareness during major online promotion periods could significantly increase customer traffic for malls.

“The findings from this study have important managerial implications for offline retailers undergoing digital transformation,” Hu says. “We show that by using new AI-chip-embedded sensors, malls can gain business insights from interpretable AI models for a more accurate evaluation of the effectiveness of marketing campaigns. In the future, interpretable AI models that have both high prediction accuracy and high interpretability will become more important in many business contexts. Such models will help managers go beyond what is predicted and understand why such a prediction is made.”