Algorhythm and Blues

They got the cash at the end of March 2000. Weeks later, the dotcoms marched off the cliff.

By early November 2002, only four full-time staffers remained of the 45 who'd been there at Pandora's peak. Westergren came to work to find an eviction notice posted on the office door. Several former employees were suing for back pay. "I was petrified," he says now. "I was waking up at 4 a.m. in a cold sweat. I had chest constrictions. I started losing my hair."

Even so, "I always thought this was a big idea. Everybody I demo'd it for thought it was cool. That was oxygen for me."

He needed all the oxygen he could get. Westergren would pitch Pandora 348 times before AOL and Best Buy finally built the service into their consumer operations in early 2003, delivering a revenue stream that kept the company afloat. In March 2004, he secured a $7.8 million round from Walden VC; that June, he was finally able to hire Kennedy as CEO.

Inside Pandora's Box

The genome project itself is Westergren's brainchild, but the brain that shaped it was largely Nolan Gasser's. An earnest, friendly guy with a head of curly brown hair, Gasser was employee number six when Pandora launched in 2000. He is Pandora's answer to Craig Venter--and the man most likely to win a Nobel Prize for parsing the gene for Fitty Cent.

Gasser also happens to be that rarest of musical prodigies: a PhD in musicology with an equal passion for renaissance music (he has conducted an early-music ensemble), jazz (he's jammed with John Handy), and pop (he's backed Steve Miller). He deployed that eclecticism to crack the DNA of just about every musical genre, deconstructing the hundreds of traits that give each song its unique signature, from basic building-block genes like the vocal timbre and harmonic language that power rock to mutant chromosomes like rap's vocal stuttering and kick-drum patterning. Pandora currently has some 15,000 artists--several hundred thousand songs, representing a century of music--in its database.

That kind of vivisectioning is arduous work--the decoding process typically takes about 20 minutes per song (longer for dense rap lyrics, five minutes for death metal)--and Westergren knows that as well as anyone. But, he points out, the goal is not to catalog all music, just the good stuff.

"Ironically, I found over the years that the fact that we couldn't go fast was a big advantage," he says. "The problem that needs solving for music is not giving people access to 2.5 million songs." The trick is choosing wisely.

Which brings us to the company's boiler room, a back corner of the cavernous office that looks more like backstage at Lollapalooza. There, five days a week, 32 music-addicted miners pick apart stack after stack of CDs, analyzing each track against a digitized checklist of characteristics. They are all professional musicians--most moonlight in Bay Area bands--and all have studied music theory and been put through an initial 40-hour genome training session (there are training updates as new genomes launch).

"This job's awesome for a musician," says Danny Eisenberg, a spiky-haired blond keyboard player for Grammy-nominated singer Tift Merritt. "You can go on the road and come back, plus you're using your ears in your work." (Still, after a hard day's slog through the collected works of Britney Spears, even the most dedicated gene splitter needs a break: "I listen to a lot of NPR now," admits analyst Jeff Anthony.)

No one is more crucial to the process than Michael Zapruder, Pandora's music buyer, whose hunting grounds run from the Billboard charts to fanatic Web sites such as Pitchforkmedia.com and Insound.com to the catalogs of Forced Exposure in Boston and Other Music in New York. He typically buys between 250 and 350 new CDs a week, or 10,000 a year, generally fewer than 10% of which are on major labels.

"I have to keep the analysts fed," Zapruder says. "If the bins are empty, I walk through the office shamefully."

Dotcom Chorus: Caution

It was only in late August that Pandora moved from a B2B model to a consumer one, expanding beyond AOL's music-recommendation service and Best Buy's music-discovery kiosks to unveil its Web-based consumer interface. Since then, the company has relied on Web buzz to get the word out. And so far it's working: By early October, a growing blog-driven fan base led to a story on the nerd news site Slashdot and deluged the site with traffic. In less than two months, Pandora had been mentioned in more than 1,000 blog postings, most of them of the "holy s--t, this is awesome" variety.

But Ted Schadler, an analyst at Forrester Research, says "the real challenge will be to get the service in front of consumers, not just the cognoscenti who will jump through fire hoops to get cool stuff."