It has been pointed out that I haven’t written anything advertising related in quite some time. In my mind, understanding STEM is crucial to advertising, but I get that most of what interests me requires a chasm leap to convert from just interesting physics information to actual advertising. As such, my new year’s resolution is to work harder to bridge the two elements in the bloomfield knoble blog. Like all resolutions, this one probably won’t last very long either, but it’s worth a shot. I am also starting now, because (if by the time it hasn’t already been published), bloomfield knoble is updating our website, and it seems like a great time to launch.
With that introduction out of the way, let’s talk about the science of advertising. There are really several paths that science of advertising can take. Today’s blog is about using science to better target promotions and personalize customer service. A recent article in MIT Technology Review highlights the effort by IBM to use social media, specifically Twitter, to guess at people’s core psychological traits with the goal of offering personalized customer service or better-targeted promotional messages.
“We need to go below behavioral analysis like Amazon does,” says Michelle Zhou, leader of the User Systems and Experience Research Group at IBM’s Almaden Research Center in California, which developed the software. “We want to use social media to derive information about an individual—what is the overall affect of this person? How resilient is this person emotionally? People with different personalities want something different.” Zhou’s software develops a personality profile based on a person’s most recent few hundred or thousand Twitter updates. That profile scores the“big five” traits commonly used in psychological research: extroversion, agreeableness, conscientiousness, neuroticism, and openness to experience. It also scores the person on measures of “values” (for example, hedonism and conservatism) and “needs” (for example, curiosity and social harmony).
Zhou says she is working with several IBM customers to test how the technology might help their businesses. She declines to identify the companies but says they might use the system, for example, to tune marketing messages sent by e-mail or social media, or to select the promotional content displayed when a customer logs in to his or her account. A crucial part of the pilot program will test whether messages targeted with the technology’s help perform better than others. “Our hypothesis is that the conversion rates will be quite high,” says Zhou. At least, she expects them to be higher than is typical: e-mail marketing messages ordinarily have a response rate of just 0.34 percent, she says, and phone marketing calls achieve about 13 percent.
Zhou says that having a rough idea of a person’s personality could also help in call centers or other customer service settings, such as when an airline must break the news that a flight has been cancelled or delayed. “Studies show people that are extroverted want a reward and recognition, like 10,000 [frequent flyer] points,” she says. “Conscientious people want efficiency, to know their new flight right away.” In a call center scenario, a customer’s personality profile might advise a customer service agent whether to efficiently provide “just the facts” or to try to be more engaging and supportive, says Zhou. Many businesses already make use of software that analyzes social-media activity. However, it is aimed either at helping corporate representatives interact with customers or at summarizing the overall volume and tone of a discussion, not at profiling individuals.
IBM’s software was developed by recruiting people to answer psychological questionnaires and comparing the results with their Twitter activity. Machine learning software then looked at how different patterns of word use matched with psychological traits. Those correlations were used to derive models that can create a profile from a person’s tweets alone. In a study where 300 people had their Twitter profiles processed by the software and also took psychometric surveys, the results were “highly correlated” more than 80 percent of the time, says Zhou. However, she notes that when people use Twitter in a specialized way—for example, journalists discussing their beat—their tweet-derived profiles may not be so representative.
Still, Zhou argues that since the methods companies currently use to target and understand their customers are relatively imprecise, IBM’s software doesn’t have to capture a person’s personality completely to be useful. She also says it should be possible to adapt the software to use other sources of data, such as call center transcripts or online customer service chats.
Software like Zhou’s that relies on language use should be able to usefully capture something of a person’s personality, says Andrew Schwartz, a researcher at the University of Pennsylvania, who recently published a major study of how personality traits show up in Facebook activity. He says previous research has shown that measured personality traits can predict future actions, such as the number of sick days or doctor’s visits a person will report. “It seems reasonable that personality would be useful for presenting ads that resonate better with the recipient,” says Schwartz. “The ad-targeting application has been talked about for a few years now, but I think language-based measures of personality are just now becoming reliable enough to see it happen.”