No Silver Bullet: Finding the right recipe for effective brand advertising

The largest academic study into brand building effectiveness. Aimed at identifying how successful advertisers are at building brands with their current media mix. We use a dataset of approximately 1400 advertising campaigns to discover which combinations of media channels work for achieving which types of outcomes, we move beyond channel dyads to model dense combinatorial interactions, as well as modelling multiple outcomes. Our guiding analogy is the 'recipe,' which media channels use as 'ingredients.'

Jason BellFelipe Thomaz and Andrew Stephen

Partner: Kantar

Marketing image

Social media advertising effectiveness

This is a global study of over 200 digital advertising campaigns run by over 100 brands on Facebook and Instagram, looking to see how ads in these media channels impact brand-level metrics. Findings show that social media ads can have a positive impact on brand, particularly in generating awareness, but only for brands that act in a more social and 'human' manner in social media. 

Oxford: Andrew Stephen, Felipe Thomaz
External: Yakov Bart (Northeastern University)
Partner(s): Kantar Millward Brown, Facebook

Marketing image

Social media wellbeing

This project is a longitudinal study of people in the US and UK that attempts to understand the relationship between time spent using social media and their wellbeing (psychologically, financially, and physiologically). 

Oxford: Andrew Stephen, Cammy Crolic, Gillian Brooks
External: Peter Zubcsek (Tel Aviv University)

Marketing image

Blame the bot, anthropomorphism and anger in customer-chatbot Interactions

Chatbots have become common in digital customer service contexts across many industries. While many companies choose to humanise their customer service chatbots (e.g., providing them with names and avatars), little is known about how anthropomorphism influences customer responses to chatbots in customer service settings. Across five studies, including the analysis of a large real-world dataset from an international telecommunications company and four experiments, the authors find that when customers enter into a chatbot-led service interaction in an angry emotional state, chatbot anthropomorphism has a negative effect on customer satisfaction, overall firm evaluation, and subsequent purchase intentions. However, this is not the case for customers in non-angry emotional states. The authors uncover the underlying mechanism driving this negative effect (expectancy violation) and delineate implications for managers. Notably, the findings suggest it is important to both carefully design chatbots and consider the context in which they are used, particularly in customer service interactions that involve problem resolution or customer complaints.

Oxford: Andrew Stephen, Cammy Crolic, Felipe Thomaz, Rhonda Hadi

Marketing image

Honeymoon effects with advertising effectiveness when using new media channels

This is a global study of over 8000 digital advertising campaigns run on desktop and mobile devices over approximately 7 years, looking to see how effectiveness varies by media format and over time. We find that for newer formats (mobile in the earlier part of the dataset) effectiveness increases and delivers supernormal performance (i.e., the honeymoon period). As formats mature, however, effectiveness declines over time. 

Oxford: Andrew Stephen, Felipe Thomaz
External: Yakov Bart (Northeastern University)
Partner(s): Kantar Millward Brown

Marketing image

Media recipes: channel combinations for multiple outcomes

We use a dataset of approximately 1400 advertising campaigns to discover which combinations of media channels work for achieving which types of outcomes. We move beyond channel dyads to model dense combinatorial interactions, as well as modeling multiple outcomes. Our guiding analogy is the 'recipe,' which media channels as 'ingredients.'

Oxford: Andrew Stephen, Felipe Thomaz, Jason Bell
Partner(s): Kantar

Marketing image

Use of hypergraphs in early trend detection

We apply a hypergraph solution to trend detection, using social listening data to create a mapping of a multidimensional conversational space constructed of individuals, topics, locations, and time. The resulting network allows us to identify the importance of individual members of the conversation space, and importantly, which topics occupy a structural position indicating a significant future importance in the space.

Oxford: Greg Clarke, Felipe Thomaz, Andrew Stephen
Partner(s): Kantar

Marketing image

Value of creativity distinctiveness 

Identify and quantify the value of Creativity (Big C - meaning large conceptual ideation) in advertising effectiveness. Large program of study with a significant potential for a large number of sub-projects to develop.

Oxford: Andrew Stephen, Felipe Thomaz, Jason Bell, Natalia Efremova, Greg Clarke
Partner(s): Google

Marketing image

Consumer trust in mobile reviews

This project studies how consumers trust user-generated product/service reviews when they are written on mobile vs non-mobile devices. Using data from millions of TripAdvisor hotel reviews and a series of experiments, we find that consumers appear to trust reviews from mobile devices more. This is because they believe that more effort goes into writing them on mobile devices. 

Oxford: Andrew Stephen
External: Lauren Grewal (Dartmouth College)

Marketing image

How posting about products in social media affects offline purchase Intentions

This project examines the phenomenon of posting about desired products on social media platforms such as Facebook and Pinterest. Contrary to marketers’ hopes, we find that this can lead to consumers posting about products wanting those products less in real life. 

Oxford: Andrew Stephen
External: Lauren Grewal (Dartmouth College), Nicole Coleman (University of Pittsburgh)

Digital ROI measurement

This is a large research project aimed at developing new and better methods for measuring the ROI of digital advertising. 

Oxford: Andrew Stephen, Felipe Thomaz, Natalia Efremova
Partner(s): L’Oréal, Facebook

Marketing image

A comparison of machine learning approaches for efficient advertisement budget allocation

Initial investigation demonstrated that the traditional approach to marketing mix modelling (eMMM) lacks flexibility, and highly sensitive to errors, data issues, and assumptions. The goal of this project is to develop a more accurate system for MMM, that flexibly allows for both traditional and newer marketing investments.

Oxford: Andrew Stephen, Felipe Thomaz, Jason Bell, Natalia Efremova
Partner(s): L’Oréal

Food image

A feast for the eyes, how augmented reality Influences food desirability

Augmented reality (AR) has generated enormous industry investment and buzz, and the food and beverage industry has been quick to embrace the technology in hopes of augmenting the customer experience. However, limited research had empirically explored how this nascent technology might actually influence consumer judgements and behaviours. Our research demonstrates, for the first time, that because AR visually superimposes objects into a consumer’s real-time environment, it increases the ease with which consumers mentally simulate consuming a pictured food, which can in turn increase their desire for the food, purchase likelihood, and consumption enjoyment. Through two field studies and a laboratory experiment, we demonstrate the effect of AR presentation on these outcomes, support the mediating role of mental simulation, and show that the presence of nutritional information attenuates the effect. 

Oxford: William Fritz, Andrew Stephen, Rhonda Hadi

Understanding customer attention on mobile devices

This paper aims to describe the specifics of consumer attention to the advertisement on mobile platforms. The main purpose of the study is to understand  perceptual processing of the advertisement on mobile devices (also check if users exhibit any emotions during the interactions with each type of content: static, dynamic, scroll-trough, highly-engaging etc.). We will measure the amount of attention users pay to each type of ads and compare it between the different platforms. The difference with previous research is that we don't use any additional equipment, which could affect user's behaviour (eye-tracker and glasses), just the web camera on mobile device.

Oxford: Natalia Efremova, Felipe Thomaz, Andrew Stephen

Marketing image

Demystifying 'black box': applications of machine learning tools to marketing problems

In this work, we will take intuition from the field of computer science and machine learning and apply the most useful tools the field offers at the moment to common marketing problems, such as understanding social media content, sentiment of the discussion thread or budget allocation problem. We will create  a few datasets:  image and video data, text data, time series (sales) data and customer behaviour data.  We then apply state of the art models to the existing data or create code in Python that addresses the particular problem and publish it on GitHub to make it easily accessible to any researcher in marketing so everyone will be able to use it.

The purpose of this study is to facilitate the adoption of the most successful machine learning models by quantitative researchers in marketing and familiarise the qualitative researchers with the innovative tools, often used in real-world marketing practice.

Oxford: Felipe Thomaz, Natalia Efremova, Andrew Stephen, Cammy Crolic

Marketing image

Valuation models of sustainability

In this study we evaluate the changes in consumer perception and values associated with Brands' statements of purpose (general, social, environmental) over time. We also construct an augmented Capital Asset Pricing Model that incorporates this information to show that sustainability has value implications for the firm (stock/equity valuation).


Oxford: Felipe Thomaz, Jason Bell, Andrew Stephen
External: Kantar, WPP 

Marketing image

Predicting Influencer authenticity

We develop hypotheses about influencer authenticity and test them with YouTube data. We also create a predictive model of authenticity which can be applied at scale.

Oxford: Andrew Stephen, Jason Bell, Gillian Brooks
External: Edelman 

Marketing image

How technology is reshaping the customer journey

We revisit the customer journey to explore how using technologies for detecting and interpreting such signals is fundamentally reshaping it. We identify three distinct zones based on a focal firm’s insight into customer activities and the ability of competing firms to engage the customer: Blind-Spot, Latent, and Active Zone. In addition to investigating how marketing technology may affect the customer journey at different points throughout each of these zones, we examine how new technology may shift the boundaries between these zones.

Oxford: Felipe Thomaz, Andrew Stephen
External: David Schweidel, Yakov Bart, Jeff Inman, Barak Libai, Michelle Andrews, Ana Babic Rosario, Inyoung Chae, Zoey Chen, Daniella Kupor, Chiara Longoni

The abnormal structure of illegal digital marketplace communities

We examine the social network structure of a community in a dark web black market and compare it to the structure of communities in legal counterparts (amazon marketplace and eBay). We show how these differ as a function of the criminal marketplace, and for the purpose of protecting users in a privacy enhancing way.

Oxford: Felipe Thomaz
External: John Hulland (University of Georgia)

Dark web

Designed to go dark: an examination of Incentives for digital black markets to self-terminate

We consider the financial structure of Dark Web Markets and develop a mathematical proof showing that, unlike legal counterparts, they have a built-in rational expectation to cease operations. In essence, as they grow, it becomes more likely that operators will choose to close their market and steal money from customers and vendors (termed an 'exit scam'). They are, then, designed to go dark from the onset of operations; when fee structures and transactional norms are established. The introduction of varying levels of law enforcement pressure only serves to accelerate the process. 

Oxford: Felipe Thomaz, Greg Clark
External: Alexander Wiedemann (University of South Carolina)

Isomorphic and competitive homogenisation, and the buffering effect of strong brands in the face of new legislation

Evidence from the introduction of Sarbanes-Oxley

We study the change in marketing strategy given the introduction of new regulation. We use an accounting specific law (Sarbanes Oxley 2002) to show that marketing shifts in response to broad regulatory movements and the strategic and competitive nature of the firm shifts.

Oxford: Felipe Thomaz
External: John Hulland (University of Georgia), Leonce Bargeron (University of Kentucky), Chad Zutter (University of Pittsburgh)

Marketing image

Memescapes / meme culture

We aim to define and measure the structure of memes and its surrounding culture as a means by which customers share complex information and meaning with others, as well as a tool for creating and shaping culture. The role (passive or otherwise) of brands is considered.

Oxford: Felipe Thomaz
External: Ana Babic Rosario (University of Denver)

User-generated service and satisfaction

This project considers the role of user generated service (i.e., peer technical support) and the choice of the venue where these interactions take place on consumer satisfaction with the associated brand. The study combines experiments and data gathered from service forums (>700,000 user interactions) to show that consumers evaluate service received in non-brand owned environments more highly, and that non-branded environments contribute to brand satisfaction, while firm-controlled venues do not.

Oxford: Felipe Thomaz
External: Sotires Pagiavlas (University of South Carolina)

Natural language processing for brand - channel customisation

The goal of the project is to get contextual and cultural queues from content on social media to better understand the audience in the particular social media channel. To do this, we analyse text from content creators (companies and individuals) to automatically detect the sentiment and degree of affect and personal queues (social, family, cognition related). 
We aim to provide a robust framework for understanding advertisement text fit for a specific social media channel. We use Facebook as an example of the channel.

Oxford: Natalia Efremova, Felipe Thomaz
External: Yakov Bart