Personalising ads and offers in real time has become a central challenge in digital marketing, as customer preferences, contexts, and available options may change continuously. For many organisations, that ambition still clashes with technical and analytical constraints. New research by Hong Deng seeks to tackle these challenges.
On Friday, 27 February, Hong Deng will defend her PhD dissertation, which focuses on developing advanced methods for real-time personalisation in digital marketing. Her work speaks directly to practitioners working on recommendation systems, customer relationship management, and data-driven personalisation, but also to academics who research the intersection of marketing analytics, econometrics, and data science.
From static models to real-time decisions
When a customer visits a website, the system must decide which offer to show: perhaps a discount, a product suggestion, or no offer at all. That decision depends not only on customer characteristics, but also on context: time of day, device, location, or recent interactions. Deng’s methods are designed to process this information on the fly and select the most effective action at that moment.
‘Most personalisation models are still difficult to implement in real time,’ Deng explains. ‘They can be computationally complex.’ That makes them less suitable for environments where customer interactions happen continuously and conditions change rapidly. Deng’s dissertation develops algorithms that learn and adapt while they are being used.
Personalisation in dynamic environments
A core contribution of the dissertation is its focus on dynamic environments. In practice, the set of available offers is rarely fixed. Products go out of stock, campaigns change and new options are introduced. ‘Think of featured content on a news platform’, she explains. The available articles change constantly, and new content has no historical performance data. ‘How does a platform choose which content to feature for a specific user?’ she illustrates. Deng’s methods allow systems to learn the effectiveness of new options while continuing to personalise recommendations.
Beyond this, the dissertation addresses two additional challenges. First, it tackles the use of high-dimensional customer data. Companies often have rich databases, but including too many variables in real-time models can reduce performance. Deng proposes methods that automatically learn which features are truly relevant.
Second, her work accounts for time-varying effectiveness. Seasonality, competitor actions, or shifts in consumer preferences can all change what works. The proposed algorithms detect such changes and adapt targeting strategies accordingly.
What this means for practitioners
The motivation behind the research is strongly practice-driven. ‘These challenges arise from real problems companies face,’ Deng explains. The methods are designed to be implemented in languages such as Python or Julia and integrated into existing systems.
The implementation code is available upon request, and Deng is keen to collaborate with organisations interested in applying the algorithms in real-world settings. ‘We would really welcome practitioners who want to try out these personalisation methods in their own campaigns and make their targeting more efficient,’ she says. ‘The algorithms are flexible and can be tailored to different applications.’
- More information
Find more information on Hong Deng's PhD defense on Friday, 27 February.
For more information, please contact Ronald de Groot, Media and Public Relations Officer at Erasmus School of Economics, rdegroot@ese.eur.nl, or +31 6 53 641 846.
