If you want to do good design, it has to make good business sense, or else it is not going to happen.
Seth Orsborn can tell you what you want before you even know it yourself. The assistant professor of management has developed a tool to better understand what consumers like. It's based on quantifying preferences in a way that enables him to extract a lot of information from a minimum of input.
From the consumer's perspective, the method is remarkably quick and easy. Orsborn gives subjects a 25-question survey, asking them to compare three pictures of an object that differ in one aspect. The pictures might show backpacks in three different colors, for example.
"Keep in mind that we are asking 25 questions and trying to understand preference among 13 million colors," says Orsborn, who has a background in engineering and graphic art. "We are able to map the preference curve between the ones they prefer and estimate which colors they would like that we don't actually show them."
A second survey is then given to refine those estimates. This time, the subject sees backpacks, for example, in new colors that were not shown in the first survey. These new colors are selected based on the model's interpretation of the first survey. If the model's prediction of how well the customer will like the second set of colors holds true, then Orsborn's algorithm has done its job.
Orsborn's technique is currently about 85 percent accurate in predicting a customer's preference. That's not bad, considering that the average success of similar products that get to market is only about 50 percent. One challenge he is working out is automating the transition from the first to the second survey. Currently the subject has to wait about 15 minutes for Orsborn to run the process before taking the second survey.
Orsborn can incorporate any attribute that can be represented numerically into his model, from shapes to colors and even including texture, which is otherwise known as reflectivity, a numerical value. Figuring out how to represent non-visual attributes in the computer-based survey, however, is another matter. How will people decide which texture they prefer, based on a picture on the computer screen? Orsborn is trying different ideas to solve that challenge. He also is working on algorithms to incorporate several different attributes at once, without having to ask subjects too many questions.
"Eventually," he says, "I would like to cover the whole gamut of all the various attributes that go into a product and bring them all into one big ugly mathematical function that describes what people like."
Posted Sept. 27, 2010
The following links are virtual breadcrumbs marking the 27 most recent pages you have visited in Bucknell.edu. If you want to remember a specific page forever click the pin in the top right corner and we will be sure not to replace it. Close this message.