Customer Relationship Management
Customer Relationship Management
Automating Fandom in Music Communities
Fan cultures receive much attention in contemporary media studies, and for good reason. As social and cultural phenomena, they offer researchers a chance to observe seemingly pure play—authentic and often charming self-disclosures and shared identities among enthusiastic participants. However, cultural industries increasingly are seeking to rationalize and routinize these expressions of identity and solidarity in online contexts in the hopes of reducing uncertain demand (Burkart & McCourt 2006). Since consumers face an ongoing avalanche of products in the form of recordings, videos, and texts, it is imperative that marketers steer the right items to the right consumers at the right time. Discovering affinity groups, and tapping into their searching and sharing operations, has become a lucrative business. “Word-of-mouth is an incredibly powerful discovery tool for music fans,” according to eMusic’s COO, David Pakman. “Our new ‘neighbors’ and ‘top fans’ features deliver the virtual equivalent of that. For the first time, a music service will introduce you to your musical ‘neighbors’ and kickoff a more personal way to discover new music” (Choicestream 2004).
The major record labels are tethering online spaces to their newest digital distribution channels (Burkart & McCourt 2006). The “automation of fandom” denotes their management of virtual communities through sponsored online “hosts” and automated content software that defines and controls each fan’s online “experience.” As they attempt to displace informal fan sites, their legal strategies have also hampered open file sharing and threaten the “relative anonymity and diversity of public criticism” (Bielby, Harrington, & Bielby 1999: 1) characteristic of online fanzines. The automation of fandom reduces prospects for music fans’ autonomy as it also pulls the rug out from under their self-organized communities.
The music industry anticipates that online music distribution will grow in the double digits for many years to come. Recent sales figures from iTunes and other online music service providers validate digital distribution as a preferred mode of delivery over physical CD shipments. However, the need to effectively market music becomes more acute in cyberspace as recordings shed their physicality and, in many ways, their corresponding value (McCourt 2005: 249). Customer Relations Management (CRM), based on personalization systems, seeks to build brand loyalty by creating online “experiences” tailored to customer preferences and sending personalized content to consumers. In addition to its purported ability to identify customers who have affinities for particular products, CRM is useful for cross-promoting the product lines of partner companies or subsidiaries. But CRM’s greatest strength may be its ability to identify an individual customer’s value to the company or firm. Through CRM, one bank discovered that 20 percent of its customers created all of its revenues, while the other 80 percent were “destroying value” through the labor costs required to process their transactions (London 2001: 10). As the music industry makes the transition from hard goods to digital files, the value of customer profiles traded among online portals, affiliates, and advertisers rises accordingly.
Types of CRM
Online music buyers typically access such systems through a “portal” or “My Service” interface, which allows them to customize the information they receive, including news, messages, recommendations, and billing notices. These systems recognize and track returning customers; the more the customer uses the service, the more accurate its suggestions become. Such systems assemble marketing dossiers in the process; this information, about both individual and general consumer behavior, can be used to hone in-house marketing efforts and also can be traded between corporate divisions or sold to outside interests. Direct mail firms and the U.S. Postal Service have used similar software for decades to sort information into databases that can be rented and sold. In the online music industry, personalization systems fall into three categories: collaborative filtering, which suggests content based on the user’s purchasing history and volunteered comments from the user and others; human-based genre/mood matching, in which experts classify and categorize individual music tracks into logical groupings; and “listening machines,” which analyze the actual wave forms of recordings to compare their melody, tempo, harmony, timber, and density.
Collaborative filtering is intended to serve as an automated equivalent to “word of mouth.” The navigation patterns, purchasing history, and volunteered feedback of users are compared to each other, with recommendations based on the resulting matches. An early experiment in a collaborative filtering system, a 1992 project of Xerox PARC called Tapestry, was soon followed by the GroupLens project, which tracked each item a user rated along with a score for preference (Reidl, Konstan, & Voorman 2002: 5). It then estimated which users were “good predictors” by finding similarities between user tastes, and then used these “good predictors” to find new items to recommend. The first music collaborative filtering system, Ringo, was created at the MIT Media Lab in July 1994. One user described the service:
What Ringo did was simple. It gave you 20-some music titles by name, then asked, one by one, whether you liked it, didn’t like it, or knew it at all. That initialized the system with a small DNA of your likes and dislikes. Thereafter, when you asked for a recommendation, the program matched your DNA with that of all the others in the system. When Ringo found the best matches, it searched for music you had not heard, then recommended it. Your musical pleasure was almost certain. And, if none of the matches were successful, saying so would perfect your string of bits. Next time would be even better. (Negroponte & Maes 1996: n.p.)
The founders of Ringo launched Firefly, a commercial service, in 1996. Users registered for a Firefly “passport” on the company’s website and rank-ordered artists on a list. The site then looked for those with similar tastes (which Firefly termed “trusted neighbors”) and made recommendations. In addition, users could visit the home pages of their “trusted neighbors” on the Firefly website and engage in correspondence. One writer reported, “The concept was neatly logical: users would rate and review music, building a grand cross-referenced database of musical tastes. The more you told the system what you liked, the more Firefly would be able to make specific recommendations based on what other users liked” (Brown 2001). However, Firefly was unable to develop a successful business model, and pieces of its technology were sold to Microsoft and Launch.com in 1997 (Brown 2001).
Amazon.com implemented “Bookmatcher” on its website in 1997. Customers filled out forms indicating their interests, and the system matched them with book lists tailored to their entries. Bookmatcher required more time from users than most were willing to give, however, and Amazon now uses implicit data from previous purchases, which feed into its “New For You” and “Recommendations” features. Recommendations result from algorithms based on customer ratings as well as the buying patterns of customers who placed similar orders. To reduce errant data, users can exclude purchases from their “dossiers” (Stellin 2000: C8). For example, a purchase of Boxcar Willie for your sister-in-law can be excluded from your Goth or techno-laden profile, if you do not want the system to characterize you as an admirer of “the world’s favorite hobo” (Willie 2004).
An ambitious example of the human-based genre/mood matching systems, at least to judge from its name, is the “Music Genome” project of a software company called SavageBeast (as in, “music soothes the…”). Its music experts sorted thousands of recordings according to a variety of attributes, or “genes,” such as rhythm, lyrics, and instrumentation (Clark 2000: B1). On October 31, 2001, SavageBeast introduced what it called the “Celestial Home Jukebox,” a system for organizing files, building playlists, and soliciting recommendations. Among other things, the “organization” function scans one’s collection, renames and retags files with metadata, and places them in playlists according to genre, instrumentation, mood, rhythm, tempo, vocal style, etc. Users can also select “seed songs” from which the program will create playlists. The program also provides links to retailers (Music Industry News Network 2001).
User interfaces vary among human-based services, but most make recommendations based on reactions to a series of song clips. Users of failed startup Music Buddha created a “musical fingerprint” by choosing for each clip a recognizable genre such as jazz, rock, or classical, which they then narrowed into “smooth jazz” or “heavy metal.” They then chose from a limited number of lifestyle or mood options offered on a menu, such as “tattoos and pool cues” or “celebration of women.” After auditioning a selection of eight- to ten-second song “hooks,” users indicated whether these clips matched their preferences. At the end of this process, Music Buddha produced names of recordings the user might like, and opportunities to buy the recordings or add them to a “favorites” list (Brown 2001: n.p.). The CEO of Music Buddha likened the result to “the mix tape that the boyfriend gives to the girlfriend [….] That is the oh-wow moment we are trying to replicate” (Clark 2000: B1). Other systems, such as Media Unbound, iTunes, and MusicMatch, combine collaborative filtering with “expert” classification. MoodLogic.com allows users to “search by characteristics such as ‘romantic R&B songs from the 1970s,’ and visualize them graphically with elements called ‘mood magnets’” that have been collaboratively identified by its users (Clark 2000: B1).
Listening machines, on the other hand, are fully automated. In the CantaMetrix system, for example, a computer analyzes a source track’s waveforms for melody, tempo, harmony, timber, and density, and then retrieves similar recordings. Listeners may select a mood on the interface, such as “happiness,” and stream preselected tracks. Such systems have long precedents. According to Joseph Lanza (1994), Muzak president Waddil Catchings originated the idea of coding each song in the Muzak library according to rhythm, tempo, instrumentation, and ensemble size. The results were then piped into munitions factories during World War II to flatten out work efficiency curves.
These systems are increasingly integrated with industry-created virtual communities, which can include message boards and chat as well as “lighter” interactive options such as list sharing. For example,
Playlist Central is an online Rhapsody community where you can find playlists to listen to and add to your collection, or publish your own playlists to share with the world. E-mail your Playlist Central finds and creations to your friends, or post them on your blog. Friends can download Rhapsody and enjoy 25 free plays a month as well[….] My Rhapsody gives you a central place for personalized recommendations from the Rhapsody editors, once you’ve entered your favorite genres. (Rhapsody 2005: n.p.)
Rhapsody allows subscribers to email their playlists to others; if the recipient also is a Rhapsody subscriber, he or she can click on an attachment, which will play the recordings from the sender’s playlist. iTunes introduced a playlist function in October 2003, when it solicited playlists from musicians, and also offers “iMix,” in which users publish playlists that can be rated by others, as well as “Party Shuffle,” which automatically chooses songs from a user’s library. Yahoo Music promotes its “personalization & community” features in the same fashion:
Leveraging the more than three billion song, artist and album ratings of Yahoo! Music fans, Yahoo! Music Unlimited will generate personalized homepages and recommendations based on perceived tastes and subscriber community ratings. As users rate more songs and add to their digital collection by importing music from CDs, downloads, etc., subscribers discover new music through recommendations made by other Yahoo! members who have similar music tastes. (Yahoo! 2005: n.p.)
Problems with CRM
CRM is available commercially on various platforms, and many companies “roll their own” or augment outsourced applications with their own personalization software. When CRM techniques become standardized, they will probably develop variations of Nielsen and Roper reports with detailed, real-time information on consumer behavior. Although CRM is improving its surveillance, collaborative filtering, and reporting (“business analytics”) functions, however, these systems have failed to realize the success predicted by their developers. A 2003 study by Forrester Research reported that “[o]f the 30 million people who used a personalized retail site [in 2002], only 22 percent found it valuable” (Kouwe 2003: D2). One problem inherent to collaborative filtering is “cold start”: when making matches, a company needs a large group of consumers who have made a large number of purchases before it can predict future choices. A related problem is the “popularity effect.” As one executive said, “The holy grail [of CRM] is to be able to capture all the customer’s interactions in detail and get smarter about what not to recommend…. We can recommend very well. Knowing when not to bother someone is much harder” (Guernsey 2003: G5). Thus, systems err on the side of false negatives (not offering music you might like) rather than false positives (offering music you might not like). This inherent conservatism results in predictable choices. According to one critic of the Firefly service,
The service would rarely, if ever, break out of the mold of mainstream bands and recommend fringe music you’d never heard of before. And if your tastes strayed across numerous niches—say, you liked country and pop and techno, but weren’t particularly devoted to any one genre, Firefly was equally problematic; the odds of finding a community of users with identically eclectic tastes were slim. (Guernsey 2003: G5)
A similar concern was raised about GroupLens, which has been used by Amazon.com, CDNow.com, and Music Boulevard. A music critic wrote,
Just because I like a particular jazz musician such as Sonny Rollins—and I do—doesn’t mean I want to hear everyone who sounds like him. Sometimes I want to hear music unlike anything I’ve heard before and no recommendation engine will be able to find those elusive musicians for me. I won’t even be able to say what I want, or what I like, unless I hear it. (Jossie n.d., n.p.)
These systems share the weaknesses of contemporary radio market research practices such as “callouts,” wherein listeners evaluate snippets of songs, or “hooks.” Given such a limited sample, listeners respond favorably to familiar music while rejecting works that are unfamiliar. Such systems decrease the likelihood that listeners will be exposed to unfamiliar genres and develop new tastes and interests. Filtering systems, similarly, can know only what the user already likes, and they are further restricted by the arbitrary parameters imposed by coders.
Human-based genre/mood classifications, such as the Music Genome Project, are no less problematic. A comprehensive set of genre definitions and mappings is impossible to establish. Genre classification is an intensely subjective process, and continuously proliferating and evolving categories compound the problem. “Mood” classification is even more subjective. For example,
When MoodLogic’s database was searched for all songs with the mood “aggressive,” the system comes up with relatively mild fare[….] CantaMetrix [in comparison] responded to the same word with snarling tracks by Black Sabbath and System of a Down. But when asked to find songs similar to Eric Clapton’s recording of “Hoochie Coochie Man,” CantaMetrix suggested a sweet song called “God Only Knows” by some female gospel singers called the Martins, and the country classic “I Saw the Light” by Hank Williams. (Clark 2000: B1)
Ultimately, autorecommendation systems attempt the impossible. Tastes are not fixed; they are plastic and highly subjective. Yet these systems are built on the assumption that tastes are objective, mechanical, knowable—that tastes are easily reduced to mathematical formulas. Such systems “[take] the stunning breadth of choices and boils them down to a limited number. In doing so, filtering fails to unearth the incredible diversity of our tastes, the quirkiness inherent to being human” (Smith 2000: 5). Music portals construct these online “fan clubs” exclusively to promote consumption; the Rhapsody system excludes group communication except for exchanges of playlists. Their automated personalization software inverts the dynamics of informal, “bottom-up” websites created by artists or fans by funneling visitors into “digital enclosures” of continual surveillance and analysis.
Andrejevic (2003: 141) notes, “The value of a cybernetic commodity exceeds its production cost [when] the minimum of convenience or customization required to induce a consumer to relinquish personal data or to submit to detailed forms of monitoring” has been achieved. In the absence of tangible commodities such as CDs and DVDs, the support structure itself (in this case, cyberspace) becomes the commodity. It will be controlled not by users but by cultural industries that create value through transactions, or the process of circulation. To compensate for the lack of hard goods, and to rationalize their intrusion into privacy, digital cultural providers tout greater selectivity, personalization, and community as “value-added” features. These measures are increasingly necessary as consumers pick and choose single releases, rather than bundled collections, from a growing volume of material.
Combining CRM and virtual communities makes good marketing sense, because these virtual communities emulate the organic solidarity of music fan cultures, and analyzing fan interactions improves the recommendation power of CRM’s data-mining software. Yet these communities, while supposedly communal, require their participants to be individuated and isolated. The personalization of CRM furthers media trends towards narrowcasting that have long driven cable and satellite television and now drive satellite radio. You have to pay for access, and you get something ever narrower; eventually, you get a “channel” heard only by yourself. Such micro-segmentation, however, has social costs. As one observer noted, “The music may be speaking right to me, but it’s alienating being a niche market of one” (Goldberg 2000: n.p.). Critics contend that “iTunes is about music as a commodity; Napster was about music as mutual experience. iTunes is about cheap downloads; Napster was about file sharing—with sharing the key word” (Jenkins 2003: n.p.). Napster enabled users to discover new content by browsing shared folders, but such coincidental collaborative filtering is unlikely with new “authorized” digital music sites like iTunes, and even second- and third-generation network and service sites like KaZaA and Gnutella, which lack Napster’s sense of community. As Napster morphed into Napster II, music fans decried the loss of community: “What I loved about Napster was the ability to connect, often in the wee hours of the morning, with total strangers who shared my tastes and interests and to discover new music, which I would never have heard otherwise” (Jenkins 2003: n.p.).
In real communities, whether in physical space or cyberspace, members share affinities, interests, and needs; this commonality is recognized and mediated by the members themselves. Overseers or overlords of a fan community attract suspicion, and may even be ousted as members, if the fan community does not first disperse. In contrast, online cultural distributors construct the appearance of community, while largely denying members the ability to communicate or otherwise interact directly. Playlists from different services are incompatible, user behavior is inconsistent, and metadata may be missing or incomplete. While celebrity playlists or endorsements may serve a valuable function for marketers, they are constructed to support a public persona. Merchandisers also may abuse playlists as they seek to get rid of excess hard goods inventory (Kouwe 2003: D2).
As online fan communities supersede music scenes linked to places, these communities in turn will develop into “walled gardens” of online shops and light virtual communities moderated and policed by distributors. Internet visionaries see the online universe governed one day by an “adaptive Web” that will conform to the needs of its users, rather than vice versa (Gray in Wellman 2002: 96). Yet present digital delivery systems are profoundly indifferent to those needs. They strengthen norms of consumerism and quiescence; they have permitted the entertainment industry—the consciousness-making sector—to resolve its crises of production and consumption at our expense. CRM subjects us to unprecedented surveillance and manipulation. Participants have so little control of the terms of their membership that the intimate information they are required to surrender may be transferred without notice to anybody. In the process of squeezing value out of participation, these systems atomize fans and subject them to “monitored mobility” (Andrejevic 2003: 142) within the system. Intrusive online agents demanding personal information and pushing advertising and media on us will give us all the harried feeling of airport travelers, funneled through an enclosed media fun-house where we are repeatedly required to identify ourselves, and our privacy and our property exposed to search and seizure.
Online cultural distributors create “audiences” by isolating their users and reaggregating them into a manufactured community of atomized streamers and downloaders. The purpose is to encourage more consumption, faster production cycles by creators, and more disposable culture. You use it once and throw it away; you are thinking about, and paying for, not this song or that artist, but the virtual machine that delivers a reliable stream of pleasure to you alone. Aside from the corporate-controlled “communities” enabled by CRM, online cultural distributors lack common spaces and public forums for sharing tastes and experiences, further alienating fans from their music, from each other, and ultimately from themselves. The fan “scenes” vital to the creation of culture become moribund as their rituals for sharing and social spaces for group identification are supplanted by cultural industries seeking to automate fandom. On both the individual and collective levels, CRM furthers the reification of culture.