How Companies Can Make Marketing Personal

When you open your favorite music streaming app and see a playlist that feels made for you, or you scroll through an online store that seems to ‘know’ what you like, you’re seeing the power of personalization in action.

For companies, personalization promises higher engagement, stronger loyalty, and more efficient marketing. But as technology evolves, privacy concerns have also grown, and marketers are called to strike a balance between how to personalize effectively and responsibly.

In the paper, “Personalization and Targeting: How to Experiment, Learn & Optimize,” published in the International Journal of Research in Marketing, Carl Mela, the T. Austin Finch Foundation Professor at Duke University’s Fuqua School of Business, and co-authors* argue that tailoring firm decisions to individual consumers or groups based on observed characteristic has become the "heartbeat of modern marketing."

Crucially, the researchers also show how experimentation and causal inference are the principles for designing and testing marketing strategies that learn from customer preferences.

“Personalization isn’t just about data,” Professor Mela explained. “It’s about learning systematically from experience — testing, adjusting and improving the consumer experience with each interaction.”

From guesswork to experimentation 

With the dramatic growth of individual-level data—from browsing histories and app activity to geolocation and social media interactions—companies can now tailor offers and messages to each consumer.

Previous personalization methods mostly relied on algorithms and machine learning which, based on data collected from past customer behavior, try to predict who might buy what.

However, Mela and co-authors explain that recent technological advancements are making it feasible to go beyond simple prediction.

"There is a difference between forecasting who will buy, and who will buy in response to marketing, as the marketers’ goals pertain more to inducing purchase,” Mela explained. “Thus, a central question centers on learning which specific marketing action tipped the customer into buying.”

The researchers frame personalization as a causal inference problem handled through a ‘test and learn’ approach, which moves marketing beyond a mere prediction of what might happen, to identifying the precise drivers of customer actions—what caused the result. 

”Imagine a food delivery service testing different discount levels,” Mela said. “The service might find that loyal customers order again regardless of discounts, while occasional customers need the incentive. The company will save money by targeting those whose behavior actually changes because of the offer, not those already likely to order.”

Technological advances behind the new era of personalization

An array of technological innovations has transformed how companies collect and act on customer data. From mobile apps and social media to wearable devices and smart TVs, firms now have unprecedented access to the individual-level signals that reveal when, where, and how customers engage.

“Businesses can now capture not just what consumers buy, but the context around those choices—whether they’re shopping on a phone during their commute or browsing late at night at home,” Mela said.

Modern communication channels also enable marketers to connect with customers one-on-one. Mobile notifications, personalized emails, and targeted ads replace the one-size-fits-all approach of traditional mass media. 

But what the researchers consider perhaps the most transformative development is the rise of ‘continuous experimentation.’ Companies like Amazon, Booking.com, and Google have the data and the resources to run thousands of randomized tests daily, quietly fine-tuning everything from webpage layouts to pricing strategies to discount levels. 

Integrating causal inference with machine learning, these experiments allow firms to detect individual responses to marketing actions, Mela said. But it all relies on the abundance of individual-level data.

“If a firm has two consumers with a similar profile and finds that a particular marketing tactic worked for only one of them, how can it confidently conclude that the tactic is successful half of the time?” Mela said. “But with thousands of trials, a firm’s algorithm will be able to confidently predict what tactic is likely to pay off.”

The challenges of learning from data

Despite all the potential, the researchers describe a set of challenges in the way of truly effective personalization. Among them, “you may not have any information about new customers,” Mela said, referencing what academics call the cold-start problem. Another issue is noisy signals from human behavior, where confounding factors beyond the marketing tactic are the drivers of consumer responses.

To overcome these challenges, Mela said companies can use techniques such as reinforcement learning to determine what actions actually cause customers to respond.

What may be more nuanced and trickier to address are the concerns around privacy, Mela said.

The power of personalization undoubtedly has the potential of crossing ethical lines, he explained—for example, when algorithms inadvertently discriminate by race, gender, or income. 

Technology itself could help boost the safety of these practices, the authors note, with tools that promote “fair personalization” and privacy—such as bias-eliminating algorithms, and methods that preserve privacy by adding small amounts of noise to data to protect individuals’ identities.

Although worries of consumer harm are understandable, Mela thinks these risks must be balanced with the “unintended consequences” some regulations may generate, since consumers may end up losing something valuable.

“Consumers view personal data transactions like monetary transactions, expecting more in return for their data than what they pay,” Mela said. “For example, I enjoy being able to upload a picture of my past Halloween costume and receive recommendations for the current year based on that picture. But I wouldn’t like to have a tracking device in my car, which could lead to a subpoena in case of an accident.”

AI and the future of personalization

The potential of modern personalization is further empowered by the developments in generative AI (GenAI) and large language models, the authors note. These tools can synthesize information across text, images, and customer behavior to craft personalized messages at scale, enabling marketing teams to create ads that matter to consumers, Mela said.

“As a first step, I can use natural language processing or machine learning to understand what messages matter to different people,” he said. “But then I need to create the ad, and I can use GenAI to create the text, image and video those people care about.”

Mela also mentioned that as businesses collect more data about customer touch points with the brand, they can also use AI to analyze the customer interactions with the company and study what marketing tactics have affected their decisions.

A framework for the modern marketer

By reframing personalization as an experiment-based learning process, Mela and his colleagues provide marketers with a practical guide to navigate a world of abundant data.

While large companies have both resources and time to build personalized tools and techniques to process data into ‘test and learn’ experiments, smaller companies can still be part of the game by buying these services from agencies or consultants, Mela said.

In the end, the message is that personalization can become “the heartbeat of modern marketing,” a way of channeling the proliferation of data points into the understanding of the ROI of each tactic at the individual level.

“Every customer interaction is a mini-experiment,” Mela said. “If firms learn how to use this feedback—and do it responsibly—they’ll make better offers and build better relationships.”

---

* The paper is also co-authored by: Aurélie Lemmens, Jason M.T. Roos, and Sebastian Gabel (Erasmus University, Rotterdam), Eva Ascarza and Ayelet Israeli (Harvard University), Hernán A. Bruno (University of Cologne), Brett R. Gordon (Northwestern University), 
Elea McDonnell Feit (Drexel University), and Oded Netzer (Columbia University)