Pandora doing things that don't scale

In the early 2000s, as digital music platforms were beginning to transform the industry, one startup took an approach so labor-intensive and meticulous that it seemed almost absurd in the fast-moving tech world. Pandora, founded by Tim Westergren, Will Glaser, and Jon Kraft in 2000, built its recommendation engine on a foundation of manual music analysis—literally using pencil and paper to dissect songs note by note. This painstaking approach, which would eventually become known as the Music Genome Project, represents one of the most fascinating examples of a startup “doing things that don’t scale” to create something truly revolutionary.

The Genesis of an Idea

Tim Westergren’s journey to founding Pandora began with his background as a musician. After graduating from Stanford University with a degree in Political Science, Westergren spent years as a pianist and composer, even scoring independent films. This experience gave him unique insight into the structure and composition of music, as well as the challenges of connecting artists with audiences who would appreciate their work.

The idea for what would become Pandora emerged in the late 1990s when Westergren was working as a film composer. He noticed that directors would often struggle to articulate exactly what kind of music they wanted for their films. Westergren found himself asking detailed questions about musical elements—tempo, instrumentation, vocal qualities—to understand their preferences. This process of breaking down music into its component parts became the conceptual foundation for the Music Genome Project.

In 1999, Westergren teamed up with Will Glaser to develop this concept further. They envisioned creating a comprehensive taxonomy of music—a “genome” that could map the DNA of songs and use that information to make personalized recommendations. The name was inspired by the Human Genome Project, which was mapping human DNA at the time. Just as the Human Genome Project was cataloging the genetic makeup of humans, the Music Genome Project would catalog the “genetic makeup” of songs.

The Manual Analysis: Pencil, Paper, and Passionate Musicians

What made Pandora’s approach truly unique was its commitment to human analysis in an era when automated solutions were becoming the norm. Instead of relying solely on algorithms or user behavior data, Pandora hired trained musicians—many with degrees in music theory or composition—to analyze songs manually.

When the Music Genome Project first began over 20 years ago, the process was remarkably low-tech. Music analysts would listen to songs with pencil and paper in hand, manually noting various attributes. They would physically rip CDs into their ingestion systems to get the music into their database. Each song would be analyzed across hundreds of musical attributes or “genes,” including melody, harmony, rhythm, instrumentation, orchestration, arrangement, lyrics, and vocal performances.

This analysis was incredibly time-consuming. A single song could take 20-30 minutes to analyze completely. The analysts would score each attribute on a scale, creating a detailed musical fingerprint for every piece of music in their catalog. This process was not only labor-intensive but required significant musical knowledge and training.

The company eventually developed a more sophisticated interface for their analysts, but the core approach remained the same: human experts listening to and deconstructing music. This manual process allowed Pandora to capture nuances that automated systems of the time couldn’t detect—the emotional quality of a guitar solo, the distinctive characteristics of a vocalist’s vibrato, or the subtle complexities of a jazz composition.

Building the Business: 348 Rejections and Counting

While the Music Genome Project was intellectually fascinating, turning it into a viable business proved enormously challenging. Westergren and his team initially conceived of their technology as a B2B recommendation engine that could be licensed to music retailers or other businesses. However, finding customers proved difficult, and the company struggled to generate revenue.

The early 2000s were a particularly challenging time for music-related startups. The industry was in turmoil due to piracy concerns and the disruption caused by digital distribution. Investors were wary of music tech companies, and Pandora (then called Savage Beast Technologies) faced rejection after rejection.

Westergren’s persistence during this period became legendary in startup circles. He pitched the company to investors 348 times before securing the funding that saved the business. During this grueling fundraising process, Westergren accumulated nearly $150,000 in personal debt and maxed out 11 credit cards to keep the company afloat and continue paying his employees.

The situation became so dire that at one point, Westergren asked his employees to work without pay for two years, promising to make it up to them if the company survived. This extraordinary commitment from both Westergren and his team demonstrated their belief in the value of what they were building, despite the seemingly insurmountable challenges.

The Pivot to Consumer Service

The turning point for Pandora came in 2005 when the company pivoted from its B2B model to a direct-to-consumer service. Rather than trying to sell their technology to other businesses, they launched Pandora Radio, a personalized internet radio service that allowed users to create stations based on their favorite songs, artists, or genres.

This pivot was inspired by the realization that the real value of their technology was in creating personalized music experiences for individual listeners. The new service was simple: users would enter a song or artist they liked, and Pandora would create a radio station playing music with similar attributes based on the Music Genome Project’s analysis.

The consumer service launched with a free, advertising-supported model, with a subscription option for listeners who wanted to avoid ads. This approach allowed Pandora to reach a much wider audience while generating revenue through advertising and subscriptions.

Growth Through Word of Mouth

Despite having minimal marketing budget, Pandora experienced remarkable growth in its early years as a consumer service. The company relied heavily on word-of-mouth marketing, with satisfied users telling friends and family about the service.

What made Pandora so compelling to early users was the quality of its recommendations. Because of the detailed manual analysis that went into the Music Genome Project, Pandora could make connections between songs that other services couldn’t. Users were often delighted to discover new music that matched their tastes in unexpected ways.

This word-of-mouth growth was particularly important because Pandora launched before the era of social media marketing and viral growth hacking. The company had to rely on the genuine enthusiasm of its users to spread awareness of the service.

Scaling the Unscalable

As Pandora’s user base grew, the company faced the challenge of scaling a system that was fundamentally built on unscalable human analysis. Each new song added to the catalog required the same time-consuming manual process, creating a potential bottleneck as the service expanded.

To address this challenge, Pandora developed a hybrid approach that combined human analysis with technology. While they continued to have trained musicians analyze songs, they also built systems to help streamline the process and make the analysts more efficient. They created a specialized interface for music analysis that replaced the original pencil-and-paper approach, allowing analysts to work more quickly while maintaining the quality of their assessments.

The company also began using the data from their growing catalog to develop algorithms that could predict some attributes based on others, reducing the number of characteristics that needed to be manually assessed for each song. This approach allowed them to maintain the human touch that made their recommendations special while increasing the efficiency of their analysis process.

The Mobile Revolution and Beyond

Pandora’s growth accelerated dramatically with the rise of smartphones. In 2008, when Apple launched the App Store, Pandora was one of the first music services to release an iPhone app. This mobile presence helped the company reach new users and increased engagement among existing listeners who could now access the service on the go.

By 2011, Pandora had grown enough to go public, with an initial public offering that valued the company at $2.6 billion. At this point, the service had over 80 million registered users and was streaming billions of hours of music annually.

Throughout its growth, Pandora continued to refine and expand the Music Genome Project. What began as a simple system of musical attributes evolved into a sophisticated taxonomy with over 450 musical characteristics spanning multiple genres. The company expanded beyond its initial focus on popular music to include classical, jazz, hip hop, electronic music, and more, each with its own set of relevant attributes.

In recent years, Pandora has evolved its approach further with the development of MGP2, a new system that replaces the original numeric scoring with a more flexible, tag-based approach. This system allows for more nuanced analysis and can better accommodate the increasingly genre-agnostic nature of modern music.

Lessons from Pandora’s Unscalable Beginning

Pandora’s story offers several valuable lessons for entrepreneurs:

1. Sometimes the best solutions aren’t the most efficient ones. The manual analysis at the heart of the Music Genome Project was incredibly inefficient by conventional standards, but it created value that automated systems couldn’t match at the time.
2. Expertise matters. By hiring trained musicians to analyze songs, Pandora brought domain expertise to a technical problem, resulting in better recommendations than purely algorithm-driven approaches.
3. Persistence is crucial. Westergren’s 348 pitches before securing funding demonstrates the importance of believing in your vision even when others don’t see it.
4. Be willing to pivot. Pandora’s shift from a B2B technology provider to a consumer service allowed them to find a business model that worked for their unique technology.
5. Start with quality, then scale. Pandora focused first on creating high-quality recommendations through manual analysis, then gradually developed systems to make that process more efficient as they grew.

The Legacy of Doing Things That Don’t Scale

Today, Pandora (now part of SiriusXM) continues to use elements of the Music Genome Project in its recommendation system, though the technology has evolved significantly from its pencil-and-paper origins. The company’s approach has influenced the entire music streaming industry, with competitors developing their own sophisticated recommendation systems.

What makes Pandora’s story particularly compelling is that its success was built on a foundation of doing something that fundamentally didn’t scale—having humans manually analyze music. This approach ran counter to the prevailing wisdom in Silicon Valley, which often prioritizes automation and efficiency from the start.

Pandora demonstrated that sometimes the path to building something valuable requires taking the long, labor-intensive route—doing things by hand before figuring out how to automate them. By investing in this manual process, Pandora created a unique asset that differentiated it from competitors and delighted users with the quality of its recommendations.

In an industry increasingly dominated by algorithms and automation, Pandora’s origin story reminds us that human expertise and painstaking attention to detail can create value that technology alone cannot replicate. Sometimes, the things that don’t scale are precisely what make a product special.