History

  • In 1998, Netflix launched a website to rent or buy DVD’s.
  • Customer selected movies online and Netflix mailed the selected DVD’s (pay-per-rental).
  • Arrived in 1-5 days (with return envelops).
  • Targeting early adopters of technology (DVDs).

 


Business Model Strategy

(1) promote demand

  • “Unlimited” Subscription service  and Free trials (4 DVD’s) – (1999)
  • Online queue -> Optimization problem of demand (algorithms)

(2) maintain demand

  • Delivery time*: Distribution system (algorithms)
  • Marketing Editorial (digital) content (algorithms)
  • Demand prediction (algorithms)
  • Recomendation system (algorithms)

*First mover advantage: by 2004, Netflix had filed five patents for such items as its approach for estimating user ratings, managing movie queues or user wish lists, and executing rapid distribution across many locations.


Cinematch Algorithm 

Cinematch was launched in February 2002 and was based on a Collaborative Filtering System (CFS). CFS predicted a customer’s taste by comparing the customer’s past preferences to those of people with similar tastes-> Item based.

  1. Were difficult to scale and also required a significant volume of ratings.
  2. Recommendation systems were extremely hard to improve on: problem to accuracy.
By 2006, estimates indicated that 60% of Netflix rentals were recommended by Cinematch.

Problem definition

Churn problem: Not renewing subscriptions

  • Cinematch algorithm was 6 year old at that time.
  • Dedicated team of developers was unable to significant improve its performance.

When you’re banging heads together in an office trying to come up with new ideas, you sometimes run out of ideas.

Options: Increase the performance of their Cinematch algorithm, via (1) more developers, (2) using an outside consulting firm, (3) conducting cutting edge academic research on recommendation algorithms or (4) crowd based-solutions.

 

If the Cinematch algorithm could be improved by 10%, it would generate between $60 million and $89 million in annual revenue over the next four years.

Proposed Solution: Run a prize-based contest (2006) and see what solutions came from the “wisdom of the crowd.”


Legal Implications

How they would determine who owned the IP? -> Property rights

What data sets could they release without compromising the security of their customers’ private information? ->Privacy & Security


 The devil is in the details

  • Third party or in-house platform (control, evaluation, minimize exposure of data)
  • How to anonymize?
  • Precise and explicit contract over IP (own vs license).
    • application of non-winning solutions
  • Use of data/algorithms in future work.
  • Protection of new models and algorithms (discoveries).
  • Annonymity of participants (solvers).
  • Disclousure of progress (blogs and forums).
    • Quantity and diversity
  • Internal NDA’s at Netflix.
  • Clear definition of legal terms of the prize challenge.

Contest Desgin

  • Anyone with a computer could participate
  • 100 million anonymous database from 1998-2005
  • RMSE measured
  • 1 million USD prize for the 60~90 million annual value revenue
  • Participation of individuals or teams.
  • Not current or former Netflix employees and contractors could participate.
  • Participants’ identities would be reveal if they won the prize. (In fact, the winning team collaborated online and met in-person for the first time just before accepting the $1 million prize).
  • Data for research purposes not for commercial.
  • Allow participants to maintain ownership of their own code and required the winning to grant a non-exclusive license. (Before the contest even, ended, NEtflix began licensing some participants code into its system.
  • Creation of a community (leaderboard, blgos, online forums and recommendation research conferences).

_loading

Results

51,051 contestants on 41,305 teams from 186 countries, with 44,014 valid submissions.

In 2007, Korbell achieved 8.43% improvement. (2000 hours combining 107 algorithms). In 2008, BellKor achieved  9.44%.

The contest closed on July 26, 2009. BellKor’s Pragmatic Chaos and The Ensemble both with 10.6%. The ensemble was submitted 20 minutes after BellKor.

Netflix purchased a fully-paid, irrevocable, worldwide, non-exclusive license for every algorithm Bellkor submitted.

Netflix also published papers authored by the winners.

Winner algorithm was not implemented!!!!!!!!!!!!!!!!!!


Netflix prize II

In 2008, a pair of researchers at the University of Texas used the Netflix Prize dataset as a case study for their de-anonymization algorithm. They published a paper revealing that they were able to identify Netflix members based on patterns in the anonymous data, raising serious concerns about privacy. KamberLaw L.L.C. opened a lawsuit against Netflix, and the Federal Trade Commission pressed Netflix on the privacy risks of future contests. In 2010, Netflix reached a settlement with KamberLaw and cancelled its plans for a second contest.


References

Lakhani, Karim R., Wesley M. Cohen, Kynon Ingram, Tushar Kothalkar, Maxim Kuzemchenko, Santosh Malik, Cynthia Meyn, Greta Friar, and Stephanie Healy Pokrywa. “Netflix: Designing the Netflix Prize (A).” Harvard Business School Case 615-015, August 2014.

Lakhani, Karim R., Wesley M. Cohen, Kynon Ingram, Tushar Kothalkar, Maxim Kuzemchenko, Santosh Malik, Cynthia Meyn, Stephanie Healy Pokrywa, and Greta Friar. “Netflix: Designing the Netflix Prize (B).” Harvard Business School Supplement 615-025, September 2014.

Digital Ubiquity: How Connections, Sensors, and Data Are Revolutionizing Business. Marco Iansiti & Karim R. Lakhani. Harvard Business School, November 2014 Issue. https://hbr.org/2014/11/digital-ubiquity-how-connections-sensors-and-data-are-revolutionizing-business.


Other personal thoughts as conclusion

The early adoption (2002) of their digital capabilities and exploitation of data combined with a mindset of intellectual property rights protection, gave Netflix’s a first-mover advantage to innovate and become the leader in the industry. Non-only Netflix was able to use Operational Data but was able to connect all the information that was possible to collect. As the “Digital Ubiquity” article states, is that the new global economy will rely on companies that are able to “rethink their business models, identifying new opportunities for creating and capturing value”. The way to go are the new modes of value creation, based in new data that organizations can accumulate. Netflix did not just used the information (data) from the traditional perspective of analyzing segments of the operations data, but were able to accumulate data from each part the customer experience and connect in a optimum “network” database.

The keystone, from my point of view, is this mindset of  “learning from the masses”. It reminded me about the research from Alex Pentland’s Social Physics concept of social learning: how human behavior is driven by the exchange of ideas –how people cooperate to discover, select and learn strategies and coordinate their actions. Netflix, knew that as company have limited resources. Their source of comparative advantage needed to develop further, therefore, there was a need to change, despite the risks that might encounter. The idea to open to a crowdsourcing mindset would require consideration of many possible outcomes and implications such as: intellectual property, incentives of participation and cooperation, infrastructure investment, integration of solutions, leakage of information to competitors, etc. Everything with the sole objective to optimizing the maximum percentage of possible solution space.

Technological development is a source of amplifying the limits of what humans can understand and develop. Many of the social economist imply that development must be assure by the “right” incentives, specially those seen as a non-zero sum game of cooperation and participation. From my point of view, this kind of challenges gives this kind of “ecosystem” that create the perfect environment to give people the capacity to cooperate to discover, select and learn strategies and coordinate their actions. And the spillovers of this could impact the well-being of society. On the other hand, the question is about how to maximize well-being taking into account ethics and the respect of freedom of information and data protection rights. Netflix, should launch another challenge that can combine all this constraints, with the only reasons that we as a humans can develop further.

j.a. sanchez castro

 

Responder

Introduce tus datos o haz clic en un icono para iniciar sesión:

Logo de WordPress.com

Estás comentando usando tu cuenta de WordPress.com. Cerrar sesión / Cambiar )

Imagen de Twitter

Estás comentando usando tu cuenta de Twitter. Cerrar sesión / Cambiar )

Foto de Facebook

Estás comentando usando tu cuenta de Facebook. Cerrar sesión / Cambiar )

Google+ photo

Estás comentando usando tu cuenta de Google+. Cerrar sesión / Cambiar )

Conectando a %s