More broadly, American is in the process of investing $2 billion in improving its products and amenities, such as nicer business-class seats, international wi-fi, and even its boarding music—which means ”pop piano” is out, and indie hits like Real Estate and The XX are in.

But what’s interesting is how American decided to change, and how it’s measuring success: in part, by monitoring and measuring what its customers are saying on social media, particularly on Twitter. ”I’m not sure why it is, but we get a lot of feedback on boarding music and arriving music,” Brian Richardson, the airline’s manager of in-flight entertainment and connectivity, tells Quartz.

American Airlines has actually won awards for how it uses Twitter for customer service—mostly helping passengers, and apologizing for delays. But Twitter has turned out to be a compelling source of information to help make and support operational decisions, too—down to the boarding tunes.

“The decision to change to the lesser-known indie music was a reaction to customers’ complaints, which included boredom,” according to an American Airlines employee newsletter article, which was posted to Twitter. “Since the switch, complaints about our cabin music have decreased, and compliments have increased, according to a social media analysis.” The company uses a range of tools—including ad-hoc searches, a proprietary tool for listening and response, and commercial software called Netbase for research.

Listening to customers and making changes is nothing new, of course—that’s simply good business. (Twitter’s public nature makes it harder to ignore.) But Twitter’s broad scale and real-time speed means it has also become useful for analyzing bigger trends, and even making predictions—an area several companies are now building real businesses around.

IBM, for example, touts more than 100 “early client engagements” for services that help companies make decisions based on Twitter data. Alistair Rennie, general manager of analytics for IBM, says companies are using Twitter data for a broad range of applications, from a fashion retailer trying to sense color and pattern trends to a service company trying to match customer churn with the weather. It’s ”going well beyond typical sentiment analytics,” Rennie tells Quartz, and is “really becoming a predictive tool.”

Another firm mining Twitter data, Dataminr, just raised $130 million in financing to build out its tool that helps traders, governments, and news agencies make sense of the stream. And selling access to its feed is also becoming a meaningful business for Twitter itself. The company’s “data licensing and other” sales reached $147 million in 2014—more than doubling over 2013, and now roughly 10% of overall revenue.

📬 Sign up for the Daily Brief

Our free, fast, and fun briefing on the global economy, delivered every weekday morning.