Before Netflix began making its own content, it noticed that streaming subscribers tended to watch the television series in their catalog back-to-back-to-back. This insight was then reinforced by a study it published in The Wall Street Journal in 2013, which revealed that half the users Netflix studied watched an entire season of a television series (up to 22 episodes) in one week. When it then looked at the viewing patterns of subscribers who were watching ten "currently popular" shows on Netflix and finished a season of at least one of those shows within the space of a month, 25% of viewers in the case of one serialized drama “finished the entire 13-episode season in two days, while it took 48% of them one week to do so.” The pace was similar for a 22-episode sitcom, with 16% of viewers finishing the season in two days, while 47% completed it within seven. (Jurgensen).
Netflix interprets this data to mean that the majority of streamers would prefer to have a whole season of a show available to watch at their own pace. The company then revolutionized the television industry by deciding to release its own original series whole seasons at a time, rather than making viewers wait each week for a new episode to become available. Sarandos believes that viewer-controlled, multi-episodic viewing – aka binge watching – is most appropriate for the new digital, on-demand generation. However, he admits that in the beginning, no U.S. network partner would have ever worked with the company to co-produce a series released in such a manner. In order to change how television is distributed, then, as in the case of the Sandler deal, Netflix had to buy a domestic program that it could control start to finish (Next). That program was House of Cards. [open notes in new window]
Of course, the company’s delivery strategy makes Netflix’s television development deals profoundly more risky because it means they must purchase an entire season of a show before testing its audience appeal. Controlling this risk is why Sarandos admits that “big data is a very important resource to allow us to see how much to invest in a project,” adding that commissioning decisions at the service are “70% science and 30% art” (Martinson). In the case of House of Cards, though, the risk seems to have paid off. The show has already been nominated for thirty-three Emmy Awards and has taken home four, including Best Casting, Directing, Cinematography and Sound Mixing for a drama series. Kevin Spacey and his co-star, Robin Wright, also won the Golden Globe Award for best acting in a television series in 2015 and 2014 respectively. The series’ critical acclaim also put Netflix on the map for original television programming and helped grow subscribers from 27.1 million to 33.4 million in 2013, the year the series premiered. It is highly unlikely that the Adam Sandler deal will earn Netflix similar kinds of accolades, but clearly both House of Cards and the Sandler deal are industry game changers, and Netflix has decided to continue its binge-release strategy for its subsequent programs.
For all of its data acquisition, Netflix has surprisingly refused to release one of its most basic metrics to either its content creators or the public: namely, how many people actually stream Netflix shows or movies. And this refusal is changing the way television content and artists’ values are determined.
On some levels, Netflix’s secrecy makes sense. Because its streaming business offers a subscription-based, ad-free service, it sees little reason to release ratings data to the public. In the United States, viewing numbers are typically collected by the Nielsen Media Company, an independent, third party that tells producers, networks and advertisers how many people watched a particular episode of a program on a given week. The television industry uses this data as a comparative metric to evaluate a program’s success, establish advertising rates and negotiate creative talents’ contracts when up for renewal.
Because Netflix does not earn money from advertisers and because people can watch episodes whenever they want, traditional ratings data means little to the company and thus it sees few reasons to release viewership numbers to the public. But oddly, the secrecy over how many people have watched its original programs extends to the creators of the programs themselves. For example, Beau Willimon, the show runner of House of Cards said,
"To this day, I have no idea how many people have watched [my series] on Netflix. They have never given me any data whatsoever. All they say is, 'Well, we're doing well and we'd like another season.' And that's really all I need to know" (Howard).
Jenji Kohan, the creator of Orange is the New Black, echoed Willimon when she stated that she too had never received viewership numbers, noting she’s only received “cryptic” statements like, “We’re very pleased,” which admittedly she found disconcerting at first (Manly).
While Kohan suggests that she now finds the “darkness” somewhat liberating as she doesn’t have to anxiously track her show’s performance on a weekly basis to see how long it might survive, it’s hard to imagine that the darkness is sheer bliss.
Without knowing viewership numbers, creative personnel have no idea how much to ask for financially when a series is up for renewal. And even if that data were released, the talent and their agents would still be “potentially at the mercy of Netflix who [could] argue that the data says you’re only worth this much’” without the creative side ever knowing if that internal data is accurate since Netflix is the only party monitoring the numbers (Barr). It is unsurprising then that Netflix’s refusal to release its viewing data is forcing Hollywood to adjust to a new way of evaluating a show’s worth.
As Willimon suggests, a successful television program is more than just ratings and shares: “there are a lot of different metrics on which to gauge success now” (Howard). Those metrics may include awards nominations, Twitter chatter, merchandising figures and more. In today’s market, talent agents and managers have to use a new set of data to enter negotiation territory. "In a typical network renegotiation,” said David Fox, who handled renegotiations for Two and a Half Men's Jon Cryer, “you use critical reviews, awards, ratings and ad revenue to support your argument." But with Netflix and other streaming channels, he admits, lawyers are entering unfamiliar territory and the industry needs a new metric: “that's what everyone is looking for right now" (Rose). But when a show lacks buzz or positive reviews, such as the Netflix Original Marco Polo or Hemlock Grove, both of which were surprisingly renewed for a second season, the refusal to release viewership numbers means that reps for these stars lack almost all specific data points on which to argue (Rose), further complicating the renewal process.
As such, Netflix’s refusal to release viewership numbers has prompted new technological developments in the industry. Content creators have now asked third-party monitors, like Nielsen Media Research, to create an opt-in system that independently tracks online platform viewing. According to a high-level Nielsen employee who spoke to me about this issue on background, companies like Netflix and Amazon Instant Video have removed the watermarks that would traditionally allow a third party to track viewing through their software. Nielsen has now developed a system that circumvents those companies’ systems, allowing it to track viewership through audio content. More specifically, when a content creator agrees to share the audio files that accompany their episodes or films, the company inputs those files into a signature audio library. Whenever a “Nielsen family” watches one of that creator’s programs through a TV-connected device, its People Meter matches that audio file to deliver information about when a particular episode is being played by a unique viewer and how long that play lasted.
According to the Neilsen representative I spoke with, Netflix has been reluctant to allow for third-party measurement because keeping its creators in the dark about how many people watch a show works in Netflix’s favor when it comes time to strike a new content deal. The Nielsen service, which content creators buy, helps balance the power dynamic involved in those talks. But the system is not yet perfect. For one, Nielsen cannot monitor streaming on mobile devices (like iPhones) that are not connected to a TV. Likewise, the company cannot measure platform-specific streaming behavior without a company’s cooperation. Therefore, it cannot say how much show X has been streamed on Netflix versus Amazon. However as my source points out, “If the series is exclusive to Netflix, for instance, Orange is the New Black, then that’s a bit of a no brainer.”
Unless Netflix decides to release its ratings data, systems like Nielsen’s will only become more important. Netflix has recently announced its intent to invest over three billion dollars annually in content acquisition, meaning that more and more creators will be negotiating and re-negotiating with the service, all of them wanting to know just how much they and their properties are worth. However, Netflix may have even found a way around this by purchasing multiple seasons of a series upfront to avoid the re-negotiation problem altogether. To cite just one example, the company just announced its decision to pour $150 million into The Queen, a period drama about the reign of Queen Elizabeth, which is being developed as a six-season deal. Like House of Cards, the initial investment in the property is significant and risky. But in the long run, a multiple-season deal may help to reduce the overall cost of the property, which mostly increases in the traditional model when the talent’s contracts are up for renegotiation.
Big data and the creative process
The final element I will explore has to do with the way big data may influence and disrupt the creative processes that govern entertainment media. To be sure, Netflix’s decision to release whole television seasons at once has encouraged complicated, serial storytelling, which, historically traditional networks have not favored. As Kevin Spacey stated at the 2013 Edinburgh Television Festival, Netflix’s distribution model has encouraged a serial, storytelling that takes “a long time to tell” and possesses “sophisticated, multi-layered” story arcs with “complex characters” who “reveal themselves over time” in relationships “that take space to play out” (Plunkett and Deans). Of course, such narrative complexity did not start with Netflix; ABC’s Lost and AMC’s Breaking Bad are just two examples of serial narratives that aired before House of Cards. But Spacey is right to suggest that Netflix has certainly been one the biggest proponents of serial storytelling because its viewers’ downtime in between episodes is dramatically shorter, meaning that artists can create more complicated storylines without having to worry that its consumers will have forgotten what happened in the last episode. And both viewers and critics alike have positively embraced this creative trend.
In the case of television, Netflix has also freed its artists from the constraints traditional networks face of interrupting programming at specific intervals for ad-breaks. At the very least, network TV has to fit ads neatly within a one hour or half-hour programming time slot. Because Netflix is ad-free and exclusively streamed, its story lengths can vary significantly from episode to episode. For example, the run times of Bloodline, a series developed for Netflix by Sony Pictures Television, range from forty-nine to sixty-five minutes,while the episodes of season one of Orange is the New Black range between fifty and sixty minutes, with the finale of season two going for a full ninety-two. The goal of such loose timing, says Sarandos, is to allow his artists to
“take as much time as [they] need to tell the story well. You couldn't really do that on linear television because you have a grid, commercial breaks and the like” (Jeffries).
While Netflix may create a freer television culture attractive to creators who don’t need to fear their show won’t proceed past the pilot stage, the company’s high levels of financial investment that it pours into its series also begs the question as to how much big data will influence the creation of that programming. How much will computer algorithms begin to dictate what and how artists must create? As Jessica Leber asks,
“Will screenplays some day be written to meet the whims of data-driven media streaming companies? Will an algorithm direct writers to produce content to appeal to niche audience profiles on Netflix?”
And what about the writing that occurs at the episode level? How will big data affect the direction a show takes? Or as Siraj Datoo asks,
“If it knows that a large percentage of viewers ended an episode early when a specific character appeared often, might the company suggest the character is removed from the second season?”
There is not enough information publically available to answer each of these questions, but Netflix executives have stated that computer algorithms indeed help dictate which shows the company develops. In fact, Sarandos has repeatedly stated that Netflix mostly uses big data to figure out how large the market might be for a particular project but not to reverse engineer what that project should look like (“Next”). This sentiment is echoed by Cindy Holland, Head of Original Content at Netflix, who states that the company’s data allows her to
“identify subscriber populations that gravitate around genre areas … [and] allows us to project a threshold audience size to see if it makes for a viable project for us" (Sweney).
Todd Yellin, Vice President of Product Innovation, also notes that the company shares all of its data with its Los Angeles programming team, “to see how it compares to the shows they are thinking about [developing]” and to see if any of those shows might fit the gaps in Netflix’s catalog and user demand (Sweney). Indeed, this process likely guided Netflix’s decision to create a fourth season of Arrested Development after Fox had canceled the show in 2006. Although it’s speculation, many Netflix subscribers who started to watch the first three seasons on the service likely had high and rather rapid completion rates for the series, suggesting high viewer demand and engagement with the text. Thus to offer fans more of the series they loved, Netflix likely commissioned an additional season to meet that demand.
Sarandos also indicates that while the company uses big data to decide how much to invest in a particular series, what series to make, and which stars should be included in the show’s cast, Netflix does not get involved much in the day-to-day writing process. “One of the things I really believe in is picking really great storytellers who have a history of delivering on great storytelling and letting them tell their stories,” adding that Netflix executives only guide their creators “with a very gentle hand” (“Keynote”). When pushed at the 2015 Cannes Film Festival as to what a “gentle hand” or “light touch” means in Hollywood, Sarandos admitted that his team will often serve as a tie breaker if there is a disagreement between creative members about what works best for a show. And they will offer guidance on issues such as how much to move a character up in the story or how quickly to resolve a story element (“NEXT”). Cindy Holland, however, has stated that one of the biggest misconceptions about Netflix is “the perception that my team isn't creatively involved in [an original] series at all.” She sums up the relation like this:
“On the one hand, we have a very light touch … our culture is about freedom and responsibility, and we believe that people do their best work when you give them both. But on the other hand, I'd like to acknowledge the amazing hard work my team does. So, that's always a bittersweet comment that we hear.”
The issue is further complicated by a story Ted Sarandos recently told at an industry conference. In it, he explains that in the very first minute of House of Cards, Kevin Spacey’s character does two important things: he kills a dog and he talks directly to the camera while he does it. Early watchers of the episode reacted negatively to this scene, as evidenced by the fact that several stopped viewing the program in the first few minutes. One of two things were happening, he noted: either viewers were turned off by the unusual use of second-person; or viewers were not expecting something so startling in the first minute and thus may have decided to stop and then return to the show, when, for example, their kids were no longer in the room. “My job was to figure out which one it was, and then to ignore them both.” Any regular television network, he stated, would have cut that first scene based on the viewing data Netflix received, but he argues, Netflix kept the scene as it was to maintain the artist’s freedom and integrity (“Keynote”).
The fact that Sarandos admits that his job is to figure out why people tuned out of House of Cards suggests that this data is important and will indeed affect which direction the show takes in the future. If, for instance, it turns out that viewers were turned off by the use of second person, it would stand to reason that the second and third seasons would employ less of the device (as they actually did.) For now, however, there is not enough information that is publically available to determine how much the algorithms are influencing the writing of a series, especially in subsequent seasons, or the ways in which algorithms will prove to be vastly different in comparison to other, more traditional types of audience focus groups.
Finally, critics have wondered what the big data approach will mean for the diversity of television shows.
“If businesses perfect the job of giving us exactly what we want, when and how will be exposed to things that are new and different, the movies and TV shows we would never imagine we might like unless given the chance? Can the auteur survive in an age when computer algorithms are the ultimate focus group?” (Leonard).
As more and more media companies develop online delivery platforms, the collection and reliance on big data will only increase, and these questions will only grow more prominent. Yet, as Sarandos argues, and it’s a comment I agree with, the use of data algorithms has so far generated content that has added to rather than detracted from the diversity of the television landscape. For instance, Orange is the New Black is a prison-based dramedy with a predominantly female cast and is one of the most ethnically and physically diverse shows on television. Likewise, Narcos is set in Colombia, is directed by the Brazilian José Padilha and stars another Brazilian, Wagner Moura, in the lead role. As Netflix develops more stories to succeed in the global market, as well as in specific territories such as Latin America, viewers will have greater exposure to internationalized texts through their Netflix subscriptions.
According to Neil Hunt, Chief Product Officer at Netflix, his company spends roughly $150 million and employs roughly 300 people just to improve its video recommendation system each year, but the investment in the utilization of big data is what drives so much of the company’s success (Roetggers). Its use of big data to increase its shipping speeds, make personalized video recommendations and effectively stock its catalog helped the company succeed in the online movie rental business, ultimately playing a pivotal role in the demise of its major competitors. As the company moved into the digital streaming age, Netflix’s data scientists have continued to leverage that data in order to make smarter content acquisition and distribution choices, but in the process, the company has also disrupted several aspects of the film and television industry. More specifically, its business models have caused a shift in how media texts are valued, how contracts are negotiated, how texts are purchased, and how those texts are consumed and distributed in ways that by-pass or challenge the status quo.
As PM Napoli writes, the challenge for media scholars now is to further explore how media content production and its acquisition is increasingly reliant on sophisticated algorithms. The challenge will be for media scholars to examine not only what sorts of algorithms are being produced and how they are affecting the decision making processes of media companies, but also to understand “the assumptions, priorities, and inputs that underlie their construction.”