personalized | Consumer Value Creation

Recommendations: who has not seen them? Whenever you go online, different recommendations appear for you: with similar products, with different products, based on your past purchases, or based on what other people viewed. But did you know that all these types of recommendations have different names and different effects?

Li and Karahanna (2015) review 40 empirical studies between 1990 and 2013, that focussed on the understanding of online recommendation systems (RS). An RS is basically a web-based technology, that has the ability to advise and offer a certain product that would satisfy the individual users’ needs.

Based on past literature, three stages in this so called recommendation process have been found. Stage 1 involves the understanding of the consumer (including the collection of consumer data and creating a consumer profile), as well as the delivery of recommendations to this consumer (which are match making approach and the recommendation system presentation). This is followed by a personalized recommendation (stage 2). In stage 3, the impact of the recommendation system is assessed. Stage 3 ‘flows’ back to Stage 1 in the form of feedback.

I think especially the recommendation system presentation and its effect are particularly interesting. Multiple types of RS are discussed within academic literature, such as content-based, visual, collaborative-based and social-network based recommendations. According to Li and Karahanna (2015), these types often overlap in practice, creating hybrid RS.

The content-based recommendation takes into account a consumer’s preferences, as well as his past search and purchase behaviour. A collaborative-based recommendation system does the same, but also takes into account other customers (Other customers also bought…). One main difference is that for the latter, much more data are needed, since you need data on not only one, but more customers.

An example of recommendations on Amazon

While collaborative recommendations have as a disadvantage that new products do not have such links yet and some customers have atypical behaviour, collaborative RS are often used when it comes to alternative-based and cross-sell recommendations. The latter means that recommended items are generated across multiple, different categories, whereas the first is are mostly based on multiple customer ratings and purchases. The algorithms used for alternative-based recommendations are further based on a bunch of different customers’ clickstream data to detect preferences.

An example of a content-based recommendation is the visual recommendation. While content-based recommendations take past behaviour into account, visual recommendations do not. As expected, this type of RE shows products that are similar to another product a consumer has viewed.

So, which recommendation do you think is most effective?

That is up to you to find out (if you are still looking for a thesis topic)! While a lot of research has been done on the types of RS, limited empirical research exists on which strategies to implement to optimally use the different types of recommendation systems.

Based on some other papers and past theses I have read, I think that the visual recommendation works the least well – it will not increase the sales of one consumer, but I believe it rather shows alternatives to something they were thinking of purchasing (black dress 1, 2 or 3: that is the question). Further, while it might be nice to know what other consumers bought or viewed, I often find it irrelevant to myself. I’d rather shop-the-look if a complete outfit is shown on a model for example. However, with products other than clothes (such as books or videos) it might be different. Hence, go ahead and pick a nice thesis topic regarding these recommendations in different product categories!

Li, S., & Karahanna, E. (2015, February). Online Recommendation Systems in a B2C E-Commerce Context: A Review and Future Directions. Journal of Association for Information Systems, 72-107.

If you go to any website, or online store specifically, your behaviour is tracked. Landing page, time spent, clicks, exit page: you name it, it is tracked. But even when you leave a page, a company does not really leave you: they saw what you clicked on, and based on your browsing behaviour, they retarget you: they show you a (sometimes personalized) advertisement on another channel, hoping you will come back and purchase the product you viewed.

Retargeting can either be done during or after a website visit, and is done based on a customer’s visit. When showing personalized recommendations, for example, it is important to take into account the quality of the recommendation, the level of personalization and the timing. This is what Bleier & Eisenbeiss (2015) looked at: what should they show, when should they show it, and where should they show it.

As with any academic article, past literature is analysed and hypothesis are developed. In order to test the what, when and where of personalized online advertising effectiveness, Bleier & Eisenbeiss conduct two large-scale field experiments and two lab-experiments. The first field experiment looked at the interplay of degree of content personalization(DCP), state, and the time that has passed since the last online store visit, at a large fashion and sports goods retailer, who carries over 30000 products. The second field experiment, conducted at the same retailer, looked at the interplay of placement and personalization. Based on the results from these two field experiments, two lab experiments were designed: one focussing on web browsing in an experiential model, the other focussing on goal-direct web browsing.
Within this paper, thus, many things are studied and confirmed. The papers shows the importance of how to determine the effectiveness of online personalization’s, and which one works best when. When a customer sees a personalized ad right after his/her website visit, the ad becomes more effective. This is mainly because preferences are not constant: they can over time. Thus if you liked a shirt 5 minutes ago, you will mostly still like it now. Thus if a company is able to directly respond to a consumer’s behaviour, the CTR is expected to be higher.

While the effectiveness of recommendations decreases over time, the level of personalization plays a moderating role. This means that high-level personalization in later stages of the decision making process have lower effectiveness, because of changes in customers tastes’ and preferences. The personalized ad is therefore not applicable anymore. Thus, the more personalized an ad, the sooner after a website visit it should be sent. Moderate personalized ads are thus more effective over time, as they take into account these changes in preferences. As visual recommendations are often highly personalized, these type of recommendations are more relevant shortly after a visit. Cross-sell recommendation, which is a more moderate recommendation type, performs better later in time

So what does this all mean? When retargeting customers and showing them personalized ads, it is important to keep in mind how long ago they visited a website. Given that this research was performed at a large fashion/sports retailer, it would be interesting to see whether the same conclusions hold for other settings. What do you think? And when do you consider (personalized) ads target to you most effective?

Bleier, A., & Eisenbeiss, M. (2015). Personalized Online Adverstising Effectiveness: The Interplay of What, When, and Where. Marketing Science, 669-688.

Create Value "for" and "from" Consumers