We live in a world where the ‘brand’ is king and, inevitably, this results in a reduction in the choice of competing products available to us. Fancy a cola? Your choice is between Coke and Pepsi.
Want to buy a pair of trainers? Well if you’re not a Nike, Reebok or Adidas guy you’re just not at the races. Once there were hundreds of PC makers. Now there are about a dozen, at least in this market anyway, and that number is shrinking fast.
It’s not just commercial pressures. Peer pressure to conform is vast as well. What football team do you support? Chances are, if you have an Irish accent, you’re a LULA—a devotee of Liverpool, United, Leeds or Arsenal. Why on earth would anybody want to support anyone else?
Faced with this combination of commercial and peer pressures to conform it is only natural that the individual will want to break out and do their own thing, make purchases that have their own personal stamp, choose their own reading material, work towards common goals in their own way.
Enter the area of personalisation which attempts to go some way towards making people feel that they are individuals in an increasingly homogenised world. Personalisation aims to utilise the power of computing and databases in an attempt to make the World Wide Web a less sterile and anonymous place.
It has two aspects: allowing an individual to customise the way they receive information and carry out their work; and allowing an organisation to treat visitors to its Website as individuals based on the information that it can gather about them from their visit.
Personalisation is realised through the deployment of portals, and these have evolved in complexity as the Internet has grown.
According to Mark Ellis of Computer Associates: ‘Portals grew from two roots — the concept of personalising an Internet site, and the concept of displaying corporate information through a Web browser. In simple terms, this is exactly what the first portals did, before realising that they needed to do a little more to provide real business benefit. In the last four years we have seen additional security layers added, single-sign on, XML handling, LDAP compliance and a vast array of features to essentially cater for that initial concept – the single, personalised aggregation point for all corporate information.’
In terms of customising a commercial Website for the visitor, the simplest and most traditional technique has been to let the customer set their own preferences manually the first time they arrive at the site.
For example, a customer could select the stocks they wanted to watch, or set the categories of information updates and articles they wanted to view, and when and how they wanted to be notified. A disadvantage with this approach is that it places the burden of defining the preferences on the customer.
A more sophisticated example of personalisation is suggesting purchases that a customer can make based on business rules. These tend to make simple suggestions based on what a purchaser has bought, and not based on any additional information about the individual customer. A real world example would be suggesting to somebody who has just bought a pair of shoes that they should consider buying some shoe polish in the same colour. Whereas this can sometimes be useful, it is frequently intrusive and unhelpful.
The next stage is collaborative filtering which automates the process of personalisation by matching visitor profiles with profiles of other customers who exhibit similar behaviour, share similar likes and dislikes, and have similar demographics. Collaborative filtering is currently the most common personalisation technology but it has its own drawbacks.
It needs a lot of computation power, so it is forced to make simplifying assumptions. Collaborative filtering does not scale well. When collaborative filtering technology is applied to high-volume Web sites with large numbers of Web visitors/customers, it tends to fail due to the computational burden it imposes. Consequently, collaborative filtering makes simplifying assumptions to reduce the data complexity and volume, and groups individuals into ‘typical’ customers.
This can defeat the whole purpose of personalisation by appearing to be presumptuous instead of intuitive, and frequently presuming incorrectly. For example this writer once purchased from a Website a CD of music by a well respected female Canadian singer songwriter. Next time I visited the site, it suggested to me that I might like to buy an album by Celine Dion.
I declined.
In the quest to build up a clearer picture of their customers, many Website designers attempt to gather as much information about their visitors as they can. This includes tracking every click a Web visitor makes to and from the site, and building cross references to other data the company may hold on the customer, from information the customer has given freely to bought-in information from other parties. However, without the proper tools to analyse all this data, it can frequently be difficult to separate the true meaning of the information from the noise.
Ellis of CA maintains that the inevitable consequence of traditional personalisation approaches is information overload. ‘If I consumed all of the information I had at my fingertips, then I’d be informationally obese,’ he says, ‘unable to move from my desk with the weight of knowledge and the quantity of pending information. I have so many roles and interests that my personal information stream is enormous.’
That is why, he continues, that the future requirements of portal technology will be to act as a personal assistant that will increase the value and not the volume of information delivered.
‘I need some form of technology that is adept at scanning vast amounts of information for those bits I need to help me perform better. I need something that will use neural intelligence to learn from the behaviour of my peers and use that to my advantage, something that will make sense of the vast amount of data at my fingertips.
Portals are just entering into this arena now, with their proven background scouring data sources for information, and delivering huge returns on investment. Predictive analysis technology is going to be the most vital aspect of any portal environment.
Acting as a virtual PA, my portal is going to suggest reading material, contacts, Websites, research notes and workplaces which will streamline the flow of information to my desktop, or even via wireless links to my mobile phone or PDA.’
In general, popular or high-volume products tend to be commodities with tight profit margins. It follows that the effort and expense needed to build up detailed data on an individual’s preferences would be largely wasted if they were directed to purchasing such low-margin products.
Personalisation works best when it can identify a customer’s niche interests and assist them in the purchase of low-volume high margin products. The value of personalisation tools is based on the ability of such tools to identify such niche interests.
Many software companies offer personalisation tools, some as part of a suite of e-business or integration products; others based on data-mining or business-intelligence tools. Essentially a personalisation tool is based on a ‘recommendation engine’ which makes suggestions to the customer based on one or a combination of the criteria described above.
Oracle, for example, has a personalisation option as an add on to its Oracle 9i Application Server product. It is a real-time recommendation engine deployed via the Application Server product.
Oracle9iAS Personalization provides real-time recommendations and answers to questions such as: Which items is this person most likely to buy? How likely is this person to buy or like this particular item? Which other items is this person most likely to buy or like, given that they like or are buying a particular item now? Which other items are people who bought this item likely to buy?
Oracle9iAS Personalization dynamically serves personalised recommendations (such as products, page content, banner ads, and navigational links) in real time based on a registered customer’s or anonymous visitor’s explicit (transactions, purchases, ratings, and demographic data) and implicit information (mouse clicks, pages visited, and banners viewed).
DBAs and Webmasters use Oracle9iAS Personalization’s Recommendation Engine API to instruct a Web site to tag or capture a visitor’s ‘clicks’ and to request real-time recommendations. This API eliminates the need to sift through mountains of noisy ‘clickstream’ Web log data. Oracle9iAS Personalization’s ‘click’ data are combined with historical data, if available within the Oracle9iAS Personalization schema, and are passed to the Oracle9iAS Personalization Recommendation Engine.
The Recommendation Engine searches for ‘rules’ or recommendations that best fit the current session and historical data scenario and passes the recommendations to the Web application in a fraction of a second.
Oracle9iAS Personalization’s ‘personalisation index’ allows e-tailers to specify the amount of ‘individuality’ desired. Recommendations can either be widely applicable, but perhaps perceived to be of lesser value, or more ‘individualised’ and hence of higher perceived value.
Oracle9iAS Personalization handles ‘ratings’ data, or the measure of how much a customer ‘likes’ or ‘dislikes’ something. Ratings data can be used both to collect detailed information from customers and to make predictions of how much they might enjoy a recommendation. Asking a visitor to rate how much they liked past films, music CDs, or restaurants or hotels allows Oracle9iAS Personalization to make recommendations that also provide a degree of anticipated customer satisfaction.
IBM too has a personalisation tool built into its WebSphere suite of integration products and application servers. According to Robert Webster, sales manager for IBM’s application and integration middleware, the recommendation engine is based on Macromedia’s Like Minds personalisation server.
‘A site can gather information about a user from a variety of sources,’ he said. ‘When the user logs on they can be asked to list their own preferences and these will be logged in the user’s own profile. The recommendation engine will then make references based on a number of criteria. You can use different models to make the recommendations. Some models look at the pages you have viewed, some take account of what you have bought. Clearly, you can let this get out of hand very quickly. You need to limit the decision making to a certain set of business rules.’
Similar to Oracle, IBM’s WebsSphere Personalization tool is an add on to its Websphere Application Server. Its Personalization Workspace provides nontechnical users with an easy-to-use browser-based interface allowing them to control the personalisation strategy of a Website. Included in the Workspace is a campaign management tool that allows the site manager to define content and personalise e-mails targeted at a specific segments of the site’s audience.
Implicit profiling lets an organisation develop site visitor profiles and customise the site based on the content a visitor views or the actions taken by a visitor while viewing the site.
Generated business rule effectiveness reports are integrated with WebSphere Site Analyzer to highlight the success of campaigns in achieving business objectives. A rules engine is used to decide what content is displayed to each site visitor based on predefined business rules.
The recommendation engine provides collaborative filtering to make content and product recommendations to site visitors. Finally, the resource engine includes a set of Java-based application programming interfaces (APIs) which allow the personalisation system to query profile and content databases at runtime to assemble customised pages.
According to Webster, Irish-based businesses making use of the IBM personalisation tools include Belfast Crystal and the Belfast Giants ice-hockey team.
An example of a specialist personalisation company is the Belfast-based Lumio, formerly Mine-IT. According to chief executive Maurice Mulvenna, Lumio concentrates on the back-end processing and data analysis that makes personalisation possible and leaves the actual presentation functions of the Website to its customers.
Lumio’s product suite is called Re:Cognition and it comprises modules for data collection (Re:Collect), data analysis (Re:Search) and deployment (Re:Action).
Re:Collect is made available for a monthly licence fee. It’s a piece of code that is inserted on every public page of a company’s Website and is used to capture data on visitors to that page. It can gather information about the IP address of the visitor and stores information about the pages the visitor has accessed in a log file.
Re:Search uses an intelligent algorithm to deliver predictive and descriptive models that make use of e-metrics, promotional effectiveness, process analytics, segmentation, sequence discovery and other knowledge-based measures of visitor and customer behaviour across on-line and off-line information, product and service delivery channels.
Knowledge generated by Re:Search is represented in PMML, an open standard for representing knowledge discovered by data mining algorithms. According to Mulvenna, the advantage of the Lumio approach is that instead of providing reams of low-quality data from which imprecise conclusions can be drawn, site managers are getting smaller amounts of high-grade information.
It is probably true to say that nothing beats the personal touch of a small shopkeeper or business person who knows all their clients and their preferences intimately, but modern technology is helping to make shopping on the Web a more personal experience.
Ltd
