Making the most of the personal touch

Susanna Ho

With personalisation an increasingly ubiquitous feature of the online shopping experience, retailers are clearly on to something. But how are they tapping into customers’ preferences, and how are they making it pay? Professor Susanna Ho is taking an innovative approach to understand more. By Stephen Green

One of the paradoxes of modern life is that while we retreat ever further from face to face interaction and conduct more and more of our daily transactions online, at the same time our online retail experiences are becoming ever more personalised.

Professor Susanna Ho has been researching web personalisation for over thirteen years, during which time she has seen the phenomenon grow from the relative simplicity of its early stages to an increasingly sophisticated example of know-your-customer marketing.

“Back in 2000 it was much less common. We were looking at a single recommendation and how we structure that recommendation to attract the user. Now we are looking at pools of recommendations and how they work. In the future I’d like to investigate how recommendations affect user activity across the entire website – because that is what online merchants are ultimately concerned about.”

Whilst a purchase recommendation is a familiar and apparently simple idea, within cyberspace it is an extremely complex process. In the first place, you need a basis for your recommendations – how are you learning about your customer – what are the means at your disposal? Secondly, how do your present your recommendations so that they will encourage your customer to do what you want – ultimately of course, to make a purchase. Then there is how you can interpret your customers’ behaviour online – what is that telling you about how they are reacting to your recommendations and your website in general?

As web personalisation has become more complex, so it is now demanding more sophisticated analysis. Susanna’s latest research uses an innovative approach which combines two existing theoretical models to try to make personalisation research more valid as a practical guide by widening the scope of the effects that are measured.

“The practical problem is more than why a user clicks on something. It is how many users click on something, and then after they click, do they really process the information. That is why we have integrated two theories. One looks at how many items a user clicks on, and the other on how the consumer processes the information afterwards.”

Building a theoretical platform from two existing approaches is always a challenge but particularly so in this instance, because the two theories are drawn from different academic traditions.

“We use the Elaboration Likelihood Model (ELM), which is drawn from cognitive psychology, and Consumer Search Theory (CST) which is really a set of mathematical models. The two theories provide different perspectives on users’ information processing. ELM tells us how deeply a person processes a stimulus: depth of processing; while CST tells us about the extent of information-seeking going on – in other words how many recommendations you will click on before you make a purchasing decision. It took us three months to come up with a logical, convincing way to integrate these two theories.”

The Elaboration Likelihood Model is born out of research into how people process a stimulus and the factors governing the depth of their mental processing that might then lead to a decision:

“Some people like thinking,” explains Susanna, “their elaboration level is likely to be high. For others it is low, and they are more likely to rely on something peripheral to make their decision. For example, when students do a teaching evaluation, if they have high ability and are highly motivated, they will carefully evaluate course structure, its relevance to practice and the depth of classroom discussions. But for students who don’t like thinking so much, their evaluation might be based on whether the PowerPoint presentation was good, or lecture notes are available online – they base it on something peripheral.”

Consumer Search Theory applies mathematical models to describe sampling behaviour where a potential buyer is searching a range of different purchase options. In the integrated model, the researchers developed a construct called ‘attitude confidence’ which hooked the two theories together. This is based on the finding from CST that consumers actually build up confidence in their decision through the search process.

“For example,” Susanna explains, “if a manager is interviewing for a secretary. The more candidates that he or she sees – the more interviews he or she conducts – the greater the confidence he or she will have about the quality of the job applicants. In CST this is a mathematical measurement, but we frame it as part of the consumer’s attitude – it is more like a psychological construct. The resulting model can then explain the effect of web personalisation on user information processing AND on the extent of search.“

Existing research has not attempted to explain how consumer attitudes are formed towards the online merchant (or ‘personalisation agent’) or how these attitudes relate to two kinds of consumer behaviour – item sampling and item selection. There has also been a tendency to treat depth of processing and breadth of search as just a single dimension – effort of search. The integrated theory should make it easier to analyse these factors side by side, whilst distinguishing the effects of each.

For the empirical part of their research Susanna and her colleagues set up both laboratory and field testing which made use of an ongoing collaboration with Hong Kong-based online digital music provider, EOLAsia. The researchers had been working with the company since 2002 and had built up a relationship of trust and mutual benefit through some small-scale projects which made it possible for them to recruit registered EOLAsia customers for the current study. As Susanna says:

“Having their customers for our experiment would really enhance the quality of data collection, because these were real users of the system.”

In the first place, laboratory testing was carried out using students as participants. This enabled the researchers to control the environment, separate out particular effects they wished to test whilst minimising ‘noise’ and confounding factors and obtain a rich set of data which would increase the internal validity of their findings.

“We wanted to see whether the users chose the personalised recommendations or if they in fact ignored them and did their own search in the general categories. And then, at the end of the study they have downloaded a song that they can keep. I think that is one of the strengths of our study. It motivates the students to make a more careful selection because they are actually getting something.”

Whilst a laboratory study enables researchers to control for complicating factors, it is of course an artificial environment, and so combining this with a field study, in a more natural environment similar to real websites was a means of balancing that effect which gave ‘external validity’ to the findings. Although participants were existing EOLAsia users, the test website was deliberately designed differently to the EOL website to “control the effect of familiarity with the merchant – we wanted to eliminate that,” says Susanna.

Web personalisation is significant in achieving two business aims for the merchant: increasing advertising revenue and increasing sales revenue. Advertising revenue is generated through users clicking on third-party content on a site, so this is related, although not in an entirely linear fashion, to item sampling, the ‘breadth of search’ dimension – in other words, number of clicks. Conversely, sales revenue is all dependent on inducing the customer’s intention to purchase. But how do these behaviours correlate to attitude formation and how the customer reacts to personalised material? The study reveals some interesting insights:

“We wanted to look at the quality of the recommendations and how varying the recommendations would affect the depth of processing (how much the customer thinks about it) and the breadth of searching (how many they click on). What we found was that if the items are more varied, the quality is very different, and maybe the items are from different categories, it meant the user would click on more items – they would have a general look at all the recommendations. If a recommendation is more relevant to the user’s initial aim they will click on that and think about it more.

“We also found that if the user clicks on the recommendations more they would sample stock items – outside the personalised categories – less. This is very interesting because it means that with web personalisation users will actually click on the website less overall. And as a general rule, if they click on the website less it will mean that the website will get less advertising revenue. But if they click on a recommendation they are more likely to buy it. So it means that the sales revenue and the advertising revenue are actually competing with each other. Online merchants therefore need to think about whether they want to attract the customer to buy, or to click on more items to increase the advertising revenue.”

Whilst the study provides useful indicators about user behaviour within a single visit to a website, it has implications for the merchant-customer relationship over the longer term – for encouraging repeat visits and building trust and brand loyalty.

“Online merchants are interested in breadth of searching because the number of items that a customer browses will affect the likelihood of a purchase,” says Susanna. “But depth of processing is meaningful because it will implant a message in the memory. Someone may not buy the product in this transaction, but because they thought more deeply about it, they may come back after three months and then make the purchase.”

This is, of course, one of the reasons that established brand leaders continue to advertise – to maintain the company’s position in customers’ memories. It is not just about encouraging a purchase decision now, but about raising the likelihood of purchases in the future.

Technology continues to enhance firms’ ability to draw useful conclusions about customer attitudes and preferences and take advantage of them. Each time a customer buys, or browses online, their personal market place becomes better defined for the merchant. As well as continuing to investigate the impact of personalisation on consumers’  website use, Susanna now hopes to extend her research to examine how personalisation and the use of mobile devices is being exploited to encourage impulse purchases.

Updated:   14 October 2014 / Responsible Officer:  CBE Communications and Outreach / Page Contact:  College Web Team