Can You Improve Your E-Mails by Analyzing Recipients’ Personalities?
Startup Crystal claims it can help you write better e-mails by mining recipients’ online data for clues to their personality.
The websites and apps we use collect ever-growing mountains of data about us that are being used to tailor all kinds of content to our preferences.
It can be hard to figure out just what to say in an e-mail to someone you don’t know very well. A startup wants to make this easier by correcting messages as you type, suggesting changes that may make the recipient more receptive to what you’re saying. These suggestions are gleaned from data gathered about the recipients online.
Crystal, which launched in an invite-only beta in March, attempts to show you the best and worst ways to converse with people, in messages and in person, by scrutinizing publicly available data from LinkedIn, Twitter, blogs, and other online sources. The startup lets users look up people’s personality profiles on its website for free; for $19 per month, you can access a Gmail plug-in for the Chrome Web browser that offers specific real-time suggestions about word choice and punctuation, depending on whom you’re writing to. A mobile app is also in the works.
While Crystal might sound creepy, at its core it’s not all that different from what huge technology companies like Facebook and Netflix already do when mining your user data to figure out what ads to show you or movies to suggest.
“I could see why people are put off at first, as a small segment of people are,” says Crystal creator Drew D’Agostino. “They see it as an invasion of privacy, but it’s just using public data.”
And if it can accurately depict people’s personalities, it could be helpful for interactions ranging from sales and recruiting to dating.
When you type in a name on Crystal’s website, Crystal looks it up in an existing database of profiles based on data pulled from sites like LinkedIn, Twitter, Yelp, and startup information site CrunchBase, as well as from reviews on Amazon.com. The service aggregates the data, uses several algorithms to build a personality score based on what it finds, and matches that score with one of its 54 personality types (D’Agostino says these types are determined by using a few existing personality assessment tools).
After choosing a personality type, Crystal shows you the results—a quick personality summary, along with advice on how to speak to that person—as well as a score to show how confident Crystal is that that the results will be accurate (the more data it has to munch on about a person, the higher the confidence score).
When I look up Barack Obama, for example, Crystal deduces that the president is “friendly, casual, and extremely perceptive, ‘connecting the dots’ more quickly than others but occasionally rambling in conversation.” It suggests that when I e-mail him, I use an emoticon and “appeal to his feelings to drive him to action.”
In addition to the president, I looked up some coworkers and family members on Crystal to see how accurately it captured their personalities. Then I wrote a few e-mails with Crystal’s Gmail plug-in to see what kinds of suggestions it gave me for crafting notes to different kinds of people. It properly described my older brother as “ambitious, critical, and pragmatic” and my little brother as “persistent and results-driven,” but like the characteristics associated with any astrological sign, those descriptions could apply to a lot of people.
Crystal was more specific when it came to e-mails; in a note to one brother it suggested I cut a few characters from the subject line, predicting he wouldn’t like a rambling header, and it flagged my use of the word “basically” in a note to an editor, recommending I leave it out or, if necessary, “replace it with something like essentially or fundamentally.” For another editor, it reprimanded me for using two question marks in a row, saying she would not appreciate them, and said I should cut the phrase “I’d love to” and instead say something more like “I want to.”
It seemed most accurate when offering tips for writing me an e-mail, perhaps because Crystal had more data about me to analyze than it did about other people I looked up. Its best tip? “Rachel only focuses on things that are immediately interesting and tends to procrastinate on the rest, so don’t use boring, formal language or make small talk.”
Jennifer Golbeck, an associate professor at the University of Maryland, College Park, who has studied how to deduce personality traits and relationships from social media, says that such personality inference technologies tend to be about 75 percent correct. The biggest challenge, she says, is getting ground-truth data about people’s actual behaviors (such as how people respond to certain phrases used in e-mails) so the tool can learn to improve the conclusions drawn from the information—something Crystal is attempting to do by asking users to add information about themselves and people they know.
As for whether Crystal is handy or creepy, Golbeck says it’s most likely both. While people might be used to seeing websites target ads and suggest content based on their online data and activities, the idea of individuals using a similar tool to craft messages may feel too personal, she says.
Keep up with the latest in machine learning at EmTech MIT.
Discover where tech, business, and culture converge.
September 11-14, 2018
MIT Media Lab