Morphing the Online World, One Consumer at a Time
The emergence of the online universe has produced tools that can fundamentally change how businesses communicate with consumers. Each consumer is unique and can be addressed
personally, but online advertising still widely employs the traditional ‘one-to-all’ and ‘best on average’ approaches. Dr. Gui Liberali from the Rotterdam School of Management (RSM)
and member of the ERIM Marketing programme applied his research in the development of Morphing, a family of methods that can have advertisements automatically adjusted to individual
“The companies that employ Morphing have, thus far, experienced substantial improvements in the use of online marketing instruments that lead to a superior economic performance by recognizing user differences.”
– Dr. Gui Liberali
Each consumer is a unique – and addressable – individual producing a rich clickstream. However, online advertising still widely employs the traditional ‘one-to-all’ and ‘best on average’ approaches and vastly ignores the potential benefits that personalisation and individualisation can offer. Meanwhile, methods that do acknowledge some degree of individual differences, such as behavioural targeting, still treat consumers as if they were part of homogeneous segments.
A growing stream of innovative research conducted at the ERIM Marketing programme has been developing methods along with colleagues at MIT to perform full-blown individualisationin real time, such as Morphing. Morphing is a family of methods that increases the performance of businesses in online environments (e.g., online sales) by matching the characteristics of marketing instruments (a ‘morph’) to each individual consumer’s cognitive style. For example, a morph may be a website design with primarily detailed data content, which may be most appropriate to a more analytic cognitive style of consumer, whereas another morph may be a website design with only concise top-line figures, which may be most appropriate to consumers of a more holistic cognitive style. Morphing unobtrusively estimates a consumer’s cognitive style based on a Bayesian analysis of past clicks across different webpages made by current and past website visitors.
While deciding which morph to use when communicating with a consumer of a specific cognitive style (e.g., an analytical consumer), morphing solves a tricky problem – how to balance the gains from a myopic policy with the gains from a forward-looking policy. A myopic policy would exploit the current information about the matching between morphs and cognitive styles and serve the best morph given what is currently known about this matching (for example, that high-content morphs are best for analytical consumers). A forward-looking policy would promote further exploration and learn more about this matching (for example, exploring whether analytical consumers may prefer top-line morphs if ever exposed to them). Solving in real time exploration-exploitation problems such as this one was considered technically impossible until recently. Morphing builds on new developments in the dynamic programming literature in operations research to find the optimal solution after every click, with no noticeable latency even in high-traffic websites.
Through cognitive styles and matching, morphing is able to serve each consumer the best morph for her at that specific time, instead of serving the best-on-average for all users. This concept of real-time optimal individualization has a huge impact on current business practices in the online world.
Gui Liberali (RSM), Glen Urban (MIT), and John Hauser (MIT) have consistently found that morphing performs extremely well in practice. Morphing dramatically increased click-through and conversion rates in several research collaborations with large companies. For example, they obtained a lift of up to 20% conversion rate for the British Telecom Group,1 a 93% lift in the click-through rate for CNET.com,2 and a 66% gain in the click-through rate for General Motors Corporation.3 In empirical simulations using experimental data from the British Telecom Group, the authors showed that morphing the group’s broadband sales website can increase online revenue by up to US$80 million. These results directly translate into noteworthy economic impacts for the online segment and correspondingly outweigh the increase in fixed costs associated with morphing.
ERIM has supported the growth of full-blown individualisation methods, such as morphing, by developing the e-Code, the Erasmus Centre for Optimisation of Digital Experiments. The e-Code is a centre created to develop and disseminate methods, concepts, and tools that help firms improve the way they use the Internet to relate with their consumers. This includes new ways to design and optimise digital experiments, new multi-armed bandits algorithms (such as website morphing), and A/B testing. The e-Code hosts an online portal to facilitate the interaction between firms and marketing researchers. This portal is home to online field experiments using morphing and other methods and has been used to coordinate online experiments involving thousands of data points (clicks) that are received through secure API connections established with companies from diverse sectors, ranging from telecom to insurance to higher education and online retailers.
Overall, the research conducted by Liberali and his coauthors paves the way to new technological discoveries, as methods such as morphing can very well re-define the way businesses conceive and practice online marketing. This lays the foundation for extensive and highly promising developments in the online world.
1 Hauser, J., Urban, G., Liberali, G. and Braun, M. (2009), Website Morphing, Marketing Science, 28, 2, (March- April), 202-224.
2 Urban, G., Liberali, G, Mac Donnald, E., Bordley, R. and Hauser, J. (2014), Morphing Banner Advertising, Marketing Science, 33 (1) 27-46.