Case: “Polyphonic HMI: Mixing Music and Math”.
POLIPHONIC HMI: MIXING MUSIC AND MATH
In late 2003, the management team of Barcelonabased Polyphonic HMI was preparing to launch an artificial intelligence tool that they believed had the potential to create tremendous value for the music industry. The technology, referred to as Hit Song Science (HSS), analyzed the mathematical characteristics of music (by isolating aspects such as melody, tempo, pitch, rhythm, and chord progression) and compared them with characteristics of past music hits, making it possible to determine a
song’s hit potential. Mike McCready, the CEO of Polyphonic, explained: The music industry has always used two criteria to determine if a song will be a success. One is that it sounds like a hit. They have professionals at the music labels who are paid to determine if a song sounds like a hit. And two is that they have an idea how they can bring the artist and the song to the market. The problem is that the industry has about a 10% success rate: only one in 10 songs that get promoted ever charts. We add a third criterion—that it has to have optimal mathematical patterns—and significantly increase success rates. “This piece of technology is truly special,” he raved. “In one of our early tests, HSS generated unusually high scores for Norah Jones, a jazz singer who most industry insiders expected to have limited commercial impact but whose album later rose to the top of the charts. We also correctly predicted each of the hits of rock band Maroon 5.” Nevertheless, Polyphonic was having its share of problems. Initial sales pitches had met with resistance. “When we tell music executives about the concept, they typically look at us with glazed eyes, check their watch, and think of an excuse why they need to leave as soon as possible. Many people simply cannot imagine that science might play a role,” said Tracie Reed, Polyphonic’s vice president. In addition, the company had to abandon efforts to market a music recommendation system based on the same technology to retailers when hardware partners could not be found. Funds were running very low as a result—McCready himself had agreed to forgo his monthly paychecks in exchange for company stock. At this point, Antonio Trias, the creator of much of the HSS technology, provided Polyphonic with a personal investment of $150,000. Armed with what McCready called a “shoestring budget,” the Polyphonic team was faced with several pressing questions. What was the best target market for this piece of technology—record companies, producers, or unsigned artists? And what was a suitable marketing plan? Specifically, how should the product be positioned, priced, and marketed? Company Background Polyphonic HMI was a subsidiary of Grupo AIA, a company that used its expertise in the area of artificial
intelligence and the natural sciences (such as mathematics and physics) to solve complex business problems. Founded in 1988, AIA was headquartered in Barcelona but also had operations in Mexico and the U.S. It generated just over $5 million in revenues in 2002. Many of the approximately 50 employees, including four of the five top managers, had Ph.Ds in fields such as mathematics, physics, and engineering. “AIA is known for approaching business problems from different scientific perspectives,” said Regina Llopis, AIA’s CEO. “Our Ph.Ds are often working on one specific business problem, each from a different point of view. Together, they come up with a solution.”
AIA operated in a wide range of industry settings, ranging from energy and finance to telecom and e
business. For example, its product portfolio covered a planning and monitoring tool for electric grid network operators, a moneylaundering detection system for financial institutions, and a network interconnections management system for telecommunications providers. All products were artificial
intelligence and naturalscience applications. Its list of clients was dominated by several of the main Spanish companies in banking (such as Caja Madrid, Grupo Santander, and La Caixa), utilities (such as Red ElÈctrica de EspaÒa, Endesa, and Gas Natural), and telecommunications (like Auna) but also included major players in those sectors in Europe, Latin America, and the U.S. Polyphonic HMI, founded in 2002, was AIA’s first foray into the entertainment sector. As the name implies, Polyphonic was established specifically to market Grupo AIA’s artificialintelligence tools to the music industry. Its top managers, consisting of Jimena Llopis (executive chairman), McCready (chief executive officer), and Reed (vice president for North America) each had a background in music. They had assembled an experienced advisory board, which included Thomas Mottola, one of the music world’s most prominent artist managers who previously had served as chairman and CEO of Sony Music
Entertainment, and Ric Wake, a leading producer who scored numerous hits with artists such as Taylor Dane, Mariah Carey, and Celine Dion. Brief biographies for the management team and advisory board are provided in Exhibit 1. Polyphonic’s management team worked with a small group of dedicated scientists and other staff members and had access to experts in the Grupo AIA parent organization. The annual fixed costs of operating Polyphonic were estimated to be around $500,000. Basic Technology The technology underlying HSS found its origins in an extensive analysis of millions of songs. This covered nearly all music released by music labels since the 1950s, and the database was updated weekly with new releases. Polyphonic devised a way to “listen to” a piece of music and isolate particular patterns. The process, referred to as “spectral deconvolution,” considered over 25 characteristics in total, including melody, harmony, tempo, pitch, octave, beat, rhythm, fullness of sound, noise, brilliance, and chord progression. Based on its mathematical characteristics, each song was then mapped onto a multidimensional grid called the “music universe.” Exhibit 2 provides some details. Songs with mathematical similarities are positioned very close to one another in this universe. According to McCready, the technology organized music in ways that sometimes seemed counter intuitive: Our technology reflects the mathematical patterns in the music—not necessarily the sound of music. For example, as far as the melody pattern is concerned, one composition by Beethoven could fall on one end, and another composition by Beethoven on the opposite end of the universe. Similarly, one Beethoven composition could fall very close to a song by rock band U2 or pop singer Mariah Carey. Music Recommendation System Polyphonic had initially used the technology to develop a music recommendation system. The idea was to develop a device placed in music stores that provided recommendations to shoppers, thereby helping retailers to increase sales. McCready explained: Many people walk into music stores looking for something new, but they just don’t know how to look for
it. We believe people don’t just like a certain genre of music but that they like specific mathematical patterns that transverse music genres—genres are just marketing terms. Our recommendation system locked onto that idea: we asked what music a consumer liked, matched that to our universe of music, and
used that information to recommend other music. Polyphonic planned to license the technology to retailers such as Best Buy, HMV, and Virgin Megastores for up to a quarter of a million dollars a year. It had developed the software needed for the system and counted on partners to develop the accompanying hardware. Unfortunately, the difficult economic situation—annual music sales were sharply decreasing—caused reluctance among retailers and hardware
providers. “The sales cycles with the hardware providers proved to be extremely long. We are a small, tenuously financed company, and we could not afford to ride this out,” said McCready, “so we went back
to the drawing board. We knew we had to come up with another application of the technology.” Hit Song Science That new application became HSS. McCready explained how the idea emerged: Antonio Trias, the vice president for innovation at AIA, figured out that hit songs had common mathematical properties. He had gone back to the music universe and had initially focused on songs that had made it to the Singles Top 40 of the leading music chart, the Billboard Hot 100,in the past five years. When he found that there were only about 50 to 60 hit clusters—not an infinitely large number—we realized the potential of the idea. McCready and his team felt the existence of clusters was evidence for the view that hit songs share certain
mathematical patterns. Consequently, the extent to which new releases “fit” those clusters should indicate their hit potential (see Figure A). McCready said: We can take an unreleased album and examine how the songs on that album map onto the clusters. If a song falls within one of these clusters, we can’t necessarily say that it will be a hit. We just know it has the potential. The song has to conform to a couple of other criteria in order to become a hit: it has to sound like a hit, be promoted like a hit, and be marketable. But if a song falls outside of the clusters, we know it will probably not become a hit. The closeness to a cluster was indicated with a “Hit Song Science score” on a scale of 1–10, with higher scores indicating a great hit potential. The clustering technique also allowed Polyphonic to provide insights into the coherence of an album, that is, the extent to which songs fall into the same or nearby clusters, and a list of songs with similar mathematical properties. Additional testing led to further refinements. For example, by giving higher weights to songs in the universe with higher sales levels, Polyphonic scientists could better pinpoint the significance of a cluster. Similarly, by focusing on more recent successes among the songs in the universe, they could incorporate the extent to which tastes might change over time (see Figure B). As of late 2003, AIA and Polyphonic had spent about $600,000 developing HSS. Now that the technology was in place, generating reports for clients was relatively inexpensive. McCready estimated it took about two hours, at a total cost of $300, to analyze an album with 10 songs The Music Industry
The worldwide market for recorded musicwas worth over $32 billion at the retail level in 2002 (see Exhibits 3a and 3b for a breakdown of sales by region and country). The number one market, the U.S., was a strong force in the world’s music business, both as a place of origination of new music and as a consumer of music products. In 2002, about 30% of all the recorded music was produced there, and the size of the retail market was well over $12 billion, or 39% of worldwide sales (see Exhibit 4 for manufacturers’ shipments). European countries accounted for 34% of worldwide sales in 2002, with the U.K., Germany, France, Italy, and Spain being the largest markets. Most major regions of the world had endured several consecutive years of falling music sales, which many industry insiders attributed to online and offline piracy. Although legal online downloads were still a small fraction of total sales, the rise of music distribution via the internet was beginning to affect music consumption. Exhibit 5 provides insights into consumption patterns. Recording Music Artists The music recording process typically started with the artists who wrote song lyrics, composed music, and performed music. Recognized talent and some new artists typically had contracts with a “label” within a record company that stipulated the terms under which they were to deliver one or more albums. They were supported by legal advisors, managers, and agents who helped them negotiate contracts, book concerts, and schedule recording sessions, among other things. McCready estimated that there were about 10,000 artists with a record contract in the U.S. and Europe, but only several hundred with some name recognition and commercial success. Tens of thousands—if not
hundreds of thousands—of artists were hoping to secure such a contract. “Every high school has a band who think they will be the next big thing,” said McCready. Unsigned artists often used “demo” (demonstration) recordings to attract a music publisher’s interest. “Record companies are always on the lookout for good, new material, and labels may receive three to four hundred demos a week,” mentioned McCready, “but only a small fraction of those demos lead to record contracts.” When labels signed artists,
they usually required an artist to deliver several albums over a certain period and work exclusively for that label during that time. Record companies There were five big companies, “the majors,” that dominated the recorded music business: BMG Entertainment, EMI, Sony Music, Universal Music Group, and Warner Music Group (see Exhibit 6 for their 2002 market shares). Each of those companies had various labels and music publishing companies under their umbrellas. For example, Universal Music Group incorporated at least a dozen labels aimed at the U.S. market, including Motown, Interscope, Geffen, MCA, Universal Classics, Universal Records, Universal South, and Universal/Island, and three times as many labels aimed at international markets. Some labels covered all music types, whereas others specialized in certain music genres. In the U.S. alone, there were also tens of thousands of small and midsized record companies. Figure C depicts how labels are commonly organized. When labels signed an artist, they typically had an inhouse producer, known as an artistand repertoire (A&R) person, guide the project. “A&R executives are usually young people who really like music and who have convinced somebody that they have ‘good ears,’” said McCready. Quique Tejada, an A&R director at Spanish record company Vale Music, explained his role: “I prepare the launch of new artists. I look for talent, find songs that they can record, select the best producer for each project, participate in the recording process, and check the master recordings. I am responsible for 20 artists.” “An A&R executive is shepherding a band he has discovered through their early career development,” added McCready. ”He
wants to make sure that their music is successful, because his career rides on that success. Most A&R people have a career span of three to four years. Those with a longer career span really become somebody
in the music industry—people know who they are.” Nurturing talent was a critical activity for record companies. The lion’s share of a record company’s revenues was generated through its established artists—not new artists. For Warner Music Group, for example, established artists accounted for almost 90% of the $1.7 billion in revenues from newly released titles in 2003. Producers In addition, artists often worked with independent producers who, much like the A&R people, helped the artist select music and develop a music style, oversaw recording schedules, recruited engineers, and watched over recording budgets. “A producer can be responsible for a lot of things—
writing the song, recording the song, mixing and editing it, making sure it has the right vibe. The producer is the pivot between the label, the publisher, the manager, and the artist,” explained independent producer Wake. “A good producer can see what an artist is supposed to sound like and come up with the right sound, or take an artist who has a vision and help him or her realize it.” He added: “I once saw a performance by a girl named Leslie Wonderman and just had a strong feeling, a special sensation, that she could be a star. I turned her into Taylor Dane, found a sound that worked for her, and helped her score several huge hits.” According to Wake, there were only about 20 to 30 top producers who were responsible for the majority of successes, a larger group of a few hundred producers who had a hit once in a while, and thousands of people trying to establish themselves as producers. Deals among Artists, Producers, and Record Labels The deals among artists, producers, and labels could be structured in a number of ways. In some cases, the label signed the artist and hired an inhouse producer, compensated through salary plus perhaps some royalties, to handle the project. In other cases, the label retained an independent producer or company to deliver a master recording for an artist under contract. In a third variation, independent artists and producers made a master recording and then tried to sell the master to a label. Income from music came from three different royalty streams. Mechanical royalties were mostly derived from the sale of music recordings. Performance royalties were earned each time music was performed (by
radio stations, orchestras, nightclub singers, and so on). Synchronization royalties were paid for music played to visual images and soundtracks. A record company typically paid an artist recording a new album an upfront fee and, once sales had passed a level that allowed the company to fully recover its costs, also a salesbased payment. The latter depended on the artist’s track record and stature but usually varied between 5% and 15% of the record’s
suggested retail price. Industry contracts generally dictated that most of the costs of making a record were to be repaid out of the artist’s upfront fee and royalties. “A record contract is essentially a loan from the record company to the artist,” clarified McCready. Managers and talent agents alone extracted between 25% and 40% of a performer’s income. Outside producers usually received a production fee and often also negotiated a royalty of 1% to 5% of the suggested retail price. Production fees varied dramatically, according to Wake: “A Mariah Carey record may be close to a million dollars in production fees, but a first record by a new artist might have only cost $100,000.” Marketing and Distributing Music The recording process ends with the delivery of a completed master. At this point, record companies start
planning a marketing campaign for the album (or individual singles) and establish a timetable for the physical production and distribution of records. Although practices differed across labels, the record label president and marketing director generally made final decisions about marketing strategies, but the A&R person also often played a role. “Our marketing department develops and executes a marketing plan for each of our artists, but I participate in that process as well,” remarked Tejada about the procedure at Vale Music. “In my role as the A&R director, I am most frequently in contact with the artists and know what strategies will work for them. At our weekly marketing meetings, which are led by our president and attended by all 12 of our executives, I often speak out on marketing issues.” One of the most important decisions for a label usually was which single of an album (with on average of 10 songs) to release first. “Virtually all mainstream albums released nowadays are accompanied by a single,” said McCready. “Radio airplay is the primary advertising vehicle for popular albums—and you need a single to get on the radio. The same is true for music television, which is also an important advertising channel—you need a strong single to get on MTV.” Particularly for new artists, the first single was often a makeorbreak situation. “I have heard of many situations where albums were left to ‘die’ when the first single underperformed,” said McCready. “When that happens, labels commonly decide not to release any more singles by that artist and essentially give up on the advertising campaign. It could be the end of the artist’s career.” Marketing music was expensive. “The release of a single usually involves at least $300,000 in marketing expenditures—and that is just what they will spend on the first single for an unknown artist,” said McCready. “Labels may spend in excess of $1 million to promote the single of an established star like Mariah Carey.” Often, labels did not expect to make that money back on single sales, according to McCready: “The album, not the single, will often deliver the lion’s share of revenues.” In the U.S., the first
single and the album were typically released at the same time. Marketing campaigns involved anything from the production of music videos to concert tours, cooperative advertising with retailers, instore merchandising materials, radio and television commercials, and press kits. To promote the song and get it played on the radio, free records were often sent to hundreds of radio stations. Because radio was a significant factor in introducing new artists and songs to consumers and because popular music stations were able to add at most three or four new cuts per week to their lists, competition for airplay was intense. After bribery (socalled payola) of radio stations had been outlawed in the 1960s, record labels had turned to hiring independent promoters to ensure airplay for new songs or paying “advertising” fees for a song’s first spins on the radio. Generating significant airplay for a song could easily cost $100,000 in promotion fees alone. The lion’s share of recorded music was distributed in a physical form, on a compact disc. Each year, approximately 30,000 new full (album) CDs were brought to the market. Not all of those albums were aimed at a broad, mainstream audience—McCready estimated that about 2,500 of the albums released each year were accompanied by one or more CD singles. The distribution arms of the five majors accounted for the large majority of shipments of all discs. The suggested retail price to consumers of a full (album) CD was just under $17 in the U.S.; music retailers paid about $10.50 for a CD. As contractual terms among labels, artists, and other parties often differed substantially, the breakdown of costs for labels could vary widely. Record labels typically sought
a margin of at least 30% of the price to retailers, after paying royalties to artists, fees to the publisher (about 5% of the price to retailers), manufacturing and distribution costs (about 10% of the price to retailers), administrative expenses (also about 10% of the price to retailers), A&R expenses (often as much
as 15% of the price to retailers), and marketing and promotion costs. CD singles were typically priced between $3 and $5. Although the large majority of consumers continued to buy music offline and in the form of albums, the rise of online distribution channels also stimulated downloads of individual songs, which were typically priced around a dollar each. A HitandMiss Business The vast majority of music released never became a hit. Ken Bunt, an executive at Disney’s Hollywood Records, commented on the uncertainty surrounding new releases: “Releases in the music business traditionally are a big gamble. We can spend millions of dollars to put a product on the market and not be
sure whether there will be any demand for it or whether it will get any airplay.” For each label, the few successes covered the losses made on the many failed titles. For example, in late 2003, Norah Jones seemed well on her way to singlehandedly preventing a market share decline for her label, Blue Note, which was part of EMI. Exhibit 7a and Exhibit 7b list the topselling albums and singles in the U.S. in 2002. As Exhibit 7a shows, the year’s 25 highestranked albums sold just below 3 million units on average. A total of 65 albums sold over 1 million units in 2002. In that same year, just over 120 albums achieved “Platinum” status, indicating sales of at least 1 million units over their lifetime. In contrast, in that same year, only two singles reached the 1 million sales mark. It was a wellknown industry fact that less than 15% of music titles released were profitable. According to
McCready, having a single on the weekly Billboard Singles Top 40 chart was the best, albeit not a perfect, indicator of a label’s ability to recover its production and marketing costs. He estimated that fewer than 300, or about 10%, of the approximately 3,000 singles released each year made it to the Billboard Singles Top 40. Similarly, he estimated, only 10% of all 2,500 albums had at least one single that made it to the Singles Top 40. (The difference was caused by albums that had more than one successful single.) Based on information obtained from an industry expert, McCready had compiled key statistics (see Table
A) to “guesstimate” the likely revenues in each case: Table A Estimated Revenues for Singles and Albums with and without a Top 40 Chart Position “Averages are not very meaningful in the music business,” remarked McCready. “You either lose a lot of money with a dud, or you win a lot of money with a hit.” He added: “And of course a runaway smash hit
can gener
Case write-up question:
How would you react to HMI’s offering if you were a successful producer?
