The music streaming company Pandora isn’t doing so well. The company lost at least $250 million in 2016, and despite being publicly traded for five years, it has never turned a profit. Rumors that the company was angling to sell itself to Sirius XM were squashed when Sirius’s CEO dismissed an acquisition as “not very likely.” One analyst wrote that “The Pandora empire has crumbled.”

There are many reasons why Pandora is struggling. Music royalties are still expensive and hard to navigate. Other services let people listen to specific songs and albums whenever they want, while Pandora only has radio stations.

Pandora also has a much smaller catalogue of music — but that part is by design.

From the beginning, the company has banked on a proprietary music rating system called the Music Genome Project. It involves rating songs on hundreds of attributes, from “aggressive drumming” and “jazz influences” to “angry lyrics” and “ambiguous soundscapes.” These ratings are given by trained musicologists — people who are musicians themselves, many of whom have advanced degrees in music — and it takes about 12 to 15 minutes per song.

In Silicon Valley, this type of approach would normally be considered unscalable. It’s why the Music Genome Project still has fewer than 2 million songs, while other music streaming services advertise more than 30 million. These other services rely more heavily on algorithms to process songs and make recommendations, which is why they are able to add songs so much faster.

We wanted to know: Was it worth it? Or should Pandora cut its losses and concede that computers are cheaper, faster, and good enough?