A researcher at IBM’s Haifa Laboratories has developed a program that monitors the behavior of online shoppers over time.The Internet has made shopping as easy as clicking the mouse, but not everyone who surfs a site will buy something: there are those who put items in their online basket but abandon it before they get to the checkout, and some even change their mind just before confirming the purchase. Others buy regularly and would purchase more if given the right incentive.
An Israeli researcher has developed a method to predict which customers will bring in the most value for your company – whether you are selling physical items, financial services or other goods – so you can offer them better service and reap even more revenue.
“What we are dealing with is a model for Customer Lifetime Value,” says Amit Fisher, a researcher at IBM’s Haifa Laboratories. “Normally customer value is calculated by looking at the purchases up til now and assuming that that is what they will carry on doing. It is very simplistic. But you can’t assume that what happened in the past is what will happen in the future.”
Fisher decided to take algorithms from the field of data mining, operations research and artificial intelligence and combine them with economics in order to take a more complex view of predicting how human beings are likely to behave in the “long term” – a phrase which to a site like Amazon.com may mean ten years, but to another website could mean one year.
Fisher took a popular Israeli online auction site as his first test case and, in conjunction with the site’s owners, defined the relevant variables: “What is relevant are not just purchases but, say, putting a bid in. It’s whatever is important to the particular domain,” he told ISRAEL21c. All the variables are behavioral, such as how many times someone visits a site over a certain period, how many bids they put in, how many items they win, how much they spend – it’s about what the customer does, not whether they are male or female, under 30 or over 45. Everything is done anonymously, so there are no privacy worries for customers.
The site had 10,000 users, which Fisher’s technology automatically divided into groups, ten groups in this case, according to customer behaviour and other variables, such as those people who put in ten bids and didn’t win anything but logged on to the site 100 times, those who won several items that they bid during a one-month period, or people who logged in but didn’t bid on anything. He collected data daily for a year, during which time customers made over 70,000 purchases worth over $18m in total.
Fisher then used a mathematical method using in artificial intelligence called Markov Chain Theory to calculate the probabilities of a customer from a particular group carrying out a certain action, combined with how much each action was worth to the auction site in monetary terms.
Fisher found that with the most important group, the one containing the largest number of people, he could predict their future behavior almost perfectly. And this can not only provide companies with information, but they can actually improve their revenue by targeting these customers and providing preferential service. A company could also use the method to target the less significant customers and “consider people who haven’t made any transactions but can give them the chance to do that,” he says.
The next step is to prove the concept further, which Fisher says will probably done within IBM and some of IBM’s clients, customized to what they need. IBM Israel is one of the parent companies largest foreign subsidiaries, employing 2000 people.
Fisher’s method is already being integrated into other business solutions being developed within IBM worldwide. It could also be applied to real-life “brick and mortar” stores, says Fisher, “anywhere that can give you the data, supermarkets or banks, etc…” This could mean better service for all.