How Robust is Your Innovation Network: Comparing the Networks of Netflix and Hulu

Innovation is a social network phenomenon. While hard to measure directly it is possible to infer things about a company’s innovation network by looking at outputs such as publications, patents, products and financial performance. We have found it instructive to use patents

Netflix and Hulu are two companies competing in the rapidly expanding market for streaming media. Netflix began in 1998 as a DVD rental and sales business and started its streaming media service in 2007, the same year Hulu was founded. Streaming media is a hotbed of innovation and this paper examines the Innovation Networks of these two competitors. We identify key inventors that have become the “backbone” of their capacity for innovation and the relative resiliency of their networks to loss of these key people. We also examine the technical domains in which they invent and those where they differ, leading to some interesting questions about possible future directions. Finally, we examine one technical domain that is critical to both companies and identify many other industries that also depend upon this expertise that might be interesting collaborators or sources of “star” employees.

Innovation Networks

Innovation Business Partners (IBP) has developed a model of innovation as a social network phenomenon. In this model individuals possess the knowledge, the organization’s social network[1] enables the opportunity to share that knowledge, challenges provide the motivation to share and trust determines the willingness to share. Together this creates a “collective intelligence” that enables an organization to acquire and apply knowledge and skills at a rate that far exceeds that of the individual.

To visualize and measure innovation networks IBP uses various “markers” of creativity such as publications, technical documents or patents. For this analysis we have chosen the US Patent & Trademark Office database of published patent applications and granted patents. Throughout this analysis we will use the term “invention” to indicate the creative act documented in either a granted patent or published patent application. Finally, in cases where both a published patent application and granted patent exist for the same invention we remove the published application from our data to avoid duplication.

The Netflix & Hulu Innovation Networks

Hulu was founded in 2007 as a streaming media service. Hulu’s invention portfolio begins in 2008 and by the end of 2013 they have an innovation network of sixty-seven (67) inventors that produced sixty-eight (68) inventions. Netflix while founded in 1998 began its streaming service in 2007. Examining the Netflix portfolio there is a distinct shift beginning in April of 2009. Prior to that point the inventions dealt with the DVD rental business, after that point the inventions focus on the streaming media service. To make a more equitable comparison between Netflix and Hulu we will restrict analysis to the Netflix portfolio from April 2009 through 2013. With this restriction the Netflix innovation network is comprised of sixty-eight (68) inventors that produced fifty (50) inventions.

It is impossible to assess the quality or business impact of a collection of patents without detailed technical, legal and financial analysis of each portfolio but from a pure numbers perspective Hulu’s innovation network is more prolific. On the other hand the shift in the Netflix portfolio from DVD to streaming inventions is impressive. Such a dramatic shift might indicate Netflix anticipated the future was in streaming services and rapidly adjusted its focus to achieve that vision. This is characteristic of a culture of innovation that can examine and modify or discard its existing paradigms in favor of new paradigms to remain competitive. To understand how important that is, consider the counter example of Blockbuster that clung to its outdated paradigm to the very end.

In the figures below we present the comparable Netflix and Hulu Innovation Networks based upon published patent applications and granted patents from the U.S. Patent & Trademark Office (USPTO). Each node (sphere) in a network is an inventor, each edge (line) indicates that the inventors have been co-inventors on one or more inventions. A node with a circle attached indicates that inventor has one or more inventions on which he or she is the sole inventor. Finally, the diameter of each node is scaled according to the inventor’s betweenness centrality, an important network characteristic, and node color is an indication of clustering, another important characteristic. In our network model of innovation, betweenness centrality is a measure of how important an inventor is to the generation and or propagation of new knowledge. Clustering is an indication of the strength of the social bonds and trust between a given collections of inventors.

Figure 1: Netflix Innovation Network
Figure 2 Hulu Innovation Network

The first thing to notice is the different topologies of the two networks. The Hulu Innovation Network is completely connected, that is you can start with any of the sixty-seven Hulu inventors and through a series of hops connect to any other Hulu inventor. This is not true for the Netflix innovation network. Netflix is broken into five disjoint components. The largest connected component is comprised of thirty-one (31) inventors which is sixty-two (62%) of the total population. The significance of this is that new knowledge is willingly shared and propagated between people that trust one another, inventing together is one means of establishing such a trust bridge and therefore we would expect new knowledge to propagate faster and further in the Hulu innovation network.

There are many possible reasons for the disjoint components in the Netflix innovation network. In our studies of six US Naval Laboratories we have found factors such as reporting structure, geographic separation, different technical domains, management policy, length of employment, etc. all can give rise to isolated components. Regardless of the cause, disjoint components should be of concern to leadership. The “next big thing” might lie at the intersection of those components and isolation will inhibit its development. It should be of even more concern to Netflix because of the higher connectivity of Hulu. Senior leadership should be asking what management policies of Hulu encourage and facilitate the higher degree of collaboration within the inventing population?

Next examine the color coded clusters in each network. Clustering is measured by the degree of triadic closure of a given set of inventors. In social networks clusters emerge if given three people A, B and C where A-B are friends and B-C are friends implies that A-C have the strong potential to be friends, that is friends of our friends are likely to be friends. In terms of innovation networks, triadic closure implies if A-B are co-inventors, and B-C are co-inventors then A-C have a strong potential to be co-inventors. The higher the triadic closure the more likely the sharing of new ideas.

Neither Netflix nor Hulu exhibit total triadic closure and hence the multiple color coded clusters of inventors. An examination of Hulu’s completely connected innovation network reveals there are sub-clusters within it. As with connectivity this may be caused by external factors such as reporting structures or geographic location, but they may also be caused by distinct communities of interest focused on a particular technical domain or set of challenges. Detailed examination of such clusters can lead to opportunities for innovation at the intersection of clusters.

The Netflix Innovation Backbone™

In mammals the backbone provides a structure to support the animal and a conduit through which information in the form of nerve impulses propagates throughout the body. By analogy we define the Innovation Backbone™ as the collection of inventors that support and provide the knowledge conduit of the organizations innovation network.

The Innovation Backbone is comprised of the inventors with high betweenness centrality. Our experience is that this is a very small number of inventors, usually less than ten or twenty percent (<10-20%) of the inventor population. We find that these inventors and those they directly connect to, that is those they have invented with, will span the majority of the network.

In the figures below we have color coded the Netflix and Hulu Innovation Backbones. Members of the backbone and first degree links are highlighted in bright red. Note that the backbone concept only makes sense for the largest connected component in a given innovation network. For Hulu, which is totally connected that largest component is comprised of sixty-seven inventors of which we have highlighted the top ten which equates to fifteen percent of the population. For Netflix the largest connected component has thirty-one inventors and we have highlighted the top five, again fifteen percent of the population.

Figure 3: Netflix Innovation Backbone™

Figure 4: Hulu Innovation Backbone™

In both cases a very small number of inventors effectively span the largest connected component of the innovation network. Within the connected component any other inventor is no more than two hops away from a member of the backbone. Why is this important? Invention is a social phenomenon where trust is paramount. People trust their direct friends (one hop out) but trust drops quickly as you proceed two, three or four hops out. The difference between Netflix and Hulu is that Hulu’s trust bridges span the entire inventing population while the Netflix backbone only reaches sixty-two percent.

But why be concerned with the backbone? In our work with the United States Navy we have interviewed nearly eighty backbone members from five laboratories. We found them to be of two types, one the deep technical expert, and two the internal entrepreneur who understood the important problems to solve, obtained funding from management and assembled the team of deep experts to invent a solution. Further, these are the people who understand how innovation works in the organization’s culture and are trusted influencers of the larger population (trust bridges). They are an invaluable resource for leadership in the design and propagation of policies and mechanisms to increase the organization’s capacity for innovation.

Innovation Network Resilience

One interpretation of network resilience is how well a network “holds together” as key nodes are removed, for example members of the backbone. A high resilience innovation network would maintain its connectivity even after a key node is removed, a low resilience innovation network will splinter into smaller disjoint components losing key trust bridges critical to the generation and propagation of new knowledge.

Reexamining the Netflix and Hulu innovation networks we see obvious differences in the distribution of node size. Remember that the nodes have been scaled by the inventor’s betweenness centrality which is a measure of how frequently that inventor falls on the shortest path between any two other randomly chosen inventors. We interpret betweenness centrality in our model as a measure of how likely that inventor is to be involved in either the generation and/or propagation of new knowledge with those other nodes. In the Netflix innovation network there is one inventor in the center of the largest component that stands out above all others. In contrast Hulu’s innovation network has many high betweenness centrality inventors. This would indicate that there are multiple nodes through which you can connect two randomly chosen inventors on relatively short paths. Removing the highest betweenness centrality inventor in the Netflix network would in theory have significantly more impact than removing one of the many high betweenness centrality inventors at Hulu. In this sense the Hulu innovation network should be more resilient to poaching of inventors by the competition.

We decided to test this idea by removing the three (3) highest betweenness centrality inventors from both networks; this is depicted in the figures below.

Figure 5: Netflix Network Minus Top Three Betweenness Centrality Inventors

Figure 6: Hulu Network Minus Top Three Betweenness Centrality Inventors

Even after removing the three highest betweenness centrality inventors Hulu’s network remains completely connected, albeit tenuously. Contrast this with the Netflix network which quickly splinters from five into eight disjoint components. So why is this important? Competitive companies frequently poach people from one another, our network analysis reveals that the effects of such poaching could be far more detrimental to Netflix than Hulu.

Renaissance People and Technology Foci

On one hand, visualizing the co-inventor networks tells us who collaborates with whom, but it does not tell us in which domains they are inventing. To visualize this, we have graphed what is considered to be an affiliation network. In this network, the first type of node (black diamond) is the US PTO primary classification of the invention. The primary classification is the mechanism by which patent office examiners categorize inventions according to their technical features. The primary class is meant to capture the essence of the invention. The second type of node (red and green spheres) are the inventors who have invented one or more times in those primary classes. In this case, the red spheres are Netflix inventors, and the green spheres are Hulu inventors.

Like the other network graphs, the nodes in the affiliation graph are sized based on their betweenness centrality. It should be noted that many of the nodes representing the primary classes will naturally have a higher betweenness centrality than the inventors because there are more inventors than primary technology classes, and each inventor must be affiliated with one or more primary classes. Thus primary classes in this bimodal graph will more frequently fall on the shortest path between any two inventors within the network.

Figure 7: Netflix and Hulu Inventor/Primary Class Affiliation Graph

Examining the graph, one can see that some primary technology classes have very high betweenness due to the number of inventors affiliated with them. We would argue that the number of inventors contributing to these technical domains indicates these are the core technologies supporting the Netflix and Hulu business models.

Primary technical class 709: Electrical computers and digital processing systems: multicomputer data transferring, is the largest class with a combined sixty-five (65) inventors from both companies contributing towards solutions in that domain. The second largest, primary class 725: Interactive video distribution systems, has a combined forty-two (42) inventors from both companies. Later we will see that it is possible to identify other industries that are active in these technical domains that might make interesting collaborators or sources of star employees.

But the affiliation network reveals another difference about the Netflix and Hulu networks. Examining the whole network we see that there are inventors that tend to cluster in the center of the graph. These are inventors that have contributed to multiple domains, “Renaissance” people of a sort. It is visually noticeable that as you approach the center of the graph it is dominated by Hulu (green) inventors. This could indicate that Hulu has a policy of hiring people with multidisciplinary backgrounds or perhaps a policy that fosters the development multidisciplinary teams that invent in multiple domains. We cannot answer this question without access to either company’s management policies but the affiliation graph highlights yet another difference in the topologies of the two networks.

Figure 8: Distribution of Inventors Inventing in “N” Primary Classes

Above we have plotted the number of inventors having invented in “N” primary classes. Within Netflix, Neil Hunt is an example of a renaissance person with inventions in eight (8) of their thirteen (13) primary technical classes. Within Hulu, Hua Zheng is an example of a renaissance person with inventions in seven (7) of their ten (10) primary technical classes. The distribution shows that for most inventors, with the exception of Neil Hunt for Netflix, Hulu always has more inventors inventing in N or more classes for N greater than or equal to two (≥2). Again, what Hulu management policies might account for this?

Both Neil Hunt and Hua Zheng have very high betweenness centrality in their respective networks. Could it be their personal social characteristics, position within the organization or multidisciplinary skill sets that causes them to be both centers and renaissance people? It would take greater access to them and their networks to answer this interesting question.

Finally, the affiliation network also shows us that the majority of inventors are domain specific with eighty (80) inventors inventing in only one primary technical class. These may be specialists or simply newer employees who have not had the opportunity yet to invent in other domains.

Innovation Genotype™

When thinking about innovation there is another valuable biological metaphor to consider. A genotype is the genetic constitution of an individual organism. It is the collection of chromosomes containing the specific sequences of genes that express the proteins responsible for the characteristics of the animal. By analogy the collection of inventors and the relationships between them result in the inventions that become the products and services that characterize the organization.

To visualize the Innovation Genotype™ we plot an organization’s inventions against their USPTO primary classification. As noted above the primary classification is the mechanism by which patent office examiners categorize inventions according to their technical features.

Figure 9: Netflix and Hulu Innovation Genotype™

In the figure above we have plotted for comparison the Netflix (red) and Hulu (green) genotypes. The first thing to note is that not surprisingly both companies invent in classes relevant to streaming media. Second, both invent in many of the same classes. Third, there are some interesting differences.

The genotype analysis of these fierce competitors provides a rich opportunity to compare and contrast that which might give them an edge over the other. For example, Hulu has several inventions in Class 463 “Amusement devices: games”. Examining Hulu’s inventions in class 463 we find they are focused on configuring “…advertisements in a streaming video segment from a serving node is based on a result of an interactive game process executing on a client device. A configuration of advertisements in the streaming video is determined based on the game result.” Hulu already offers streaming of video game trailers, could these inventions be signaling Hulu’s intent to offer streaming games with the cost offset by targeted advertisements? Does Hulu envision creating a community of gamers that can purchase interactive games at reduced cost that only work when Hulu streams advertisements to their devices thus producing an important revenue stream?

A second interesting difference is in class 380 “Image analysis”. Hulu is interested in using facial recognition to identify an actor in a streaming video and provide the viewer with a user-selectable-link to obtain further information about the actor. While not mentioned in their published patent application one could imagine linking the user to a fan site or Facebook group creating communities and producing a more interactive experience that might increase customer loyalty and the barrier to switching from Hulu to another streaming service like Netflix.

There might be even more to learn from these inventions. Both are focused on producing a more customizable interactive experience. This could be very different from the standard video viewing experience and attract a much younger audience that grew up with gaming and demand a higher level of interactivity?

Competitors, Collaborators, Acquisitions & Star Employees

Both Netflix and Hulu have multiple inventions in Class 725 “Interactive video distribution systems”. The USPTO defines Class 725 as: “… interactive video distribution processes, systems, and elements thereof, which are characterized by point-to-multipoint system configurations, and which are used for the unidirectional distribution or delivery of motion video data resulting from interactions between systems operators (access or service providers) or users (subscribers) and systems elements.” Inventions in Class 725 are directly related to the core businesses of Hulu and Netflix.

Below in Table 1 we list the assignees receiving granted patents or published patent applications in the last fifteen months in USPTO Primary Class 725. We chose a rather short recent time frame to help identify companies currently active in this field. These assignees could represent competition, collaborators, potential acquisitions or sources of future star employees.

Alternatively, from an industrial intelligence perspective, examination of specific assignees innovation networks, backbone and invention history could produce valuable insights. For example, several of the authors of this paper have used a creative process[2] that leverages patents as the feedstock for brainstorming sessions that rapidly produce valuable innovations. This process enables a company to gain from the research investments of others.

Table 1: Inventions in Primary Class 725, Jan 2013 – Mar 2014

The Larger Culture

From a strictly business perspective one would have to admit that Netflix has been far more successful than Hulu in terms of the number of customers and financial results. This is interesting because if our team had blindly examined the two Innovation Networks without knowing whom they belonged to we would have expected the Hulu network to have been Netflix and vice versa. So how could this be?

Innovation Networks as revealed by patents are only a small part of the larger culture that creates the collective intelligence of an organization. Netflix is led by a hard charging entrepreneur Reed Hastings who has shown an ability to sense change and respond appropriately moving from DVD rentals, to streaming media and now original content. The patent network can only capture the technical side of such acumen as in the shift in 2009 from DVD related inventions to streaming media inventions. Original content is not patented nor are things like the Netflix syntax and lexicon for classifying content and viewers’ preferences into tens of thousands of micro-genre. To capture these types of collective intelligence one would have to examine the networks of documents, email and meetings that document the relevant dialogues. While this is possible it is beyond the scope of this study which is limited to publicly available data.

In contrast, while Jason Kilar, the recently deposed CEO of Hulu, has created an outstanding innovation network it is unfortunately buried in the larger culture of the network media giants that own Hulu and are not about to let Hulu gore their sacred cash cow.

Does this render the invention network analysis useless? Not at all. In fact Netflix would benefit from understanding the management policies that produced the highly connected and resilient network of Hulu. How could they do this? By hiring away two or three of the key backbone inventors. Combining the Hulu network topology with the overarching Netflix culture could be extremely powerful. Conversely the network media giants would benefit from understanding the powerful innovation network Kilar has created and unleash its disruptive powers before they fall victim to an outsider as the music industry did. Unfortunately history would predict that the media giants’ cognitive dissonance will set them up for disruption.



For more information please contact Innovation Business Partners at:

19 May 2014

[1] While the “organization chart” may define formal reporting relationships in an entity it is the underlying social network that determines the primary flow of knowledge. There may be, but need not be considerable overlap between the two network structures.

[2]Bridging Small Worlds to Accelerate Innovation”, by Markovits, Gary et al, Defense AT&L: September-October 2006.

Netflix Hulu Graphic by Mathew Keys via Flicker and cropped.

Posted in Innovation Backbone, Innovation Network Analysis.

Leave a Reply