OTT platforms are a content-driven business.
Consumers wait for the next big show to binge-watch and do their bit in telling everyone how amazing the show was. OTT content teams strive hard and pay big money to create high-impact tentpole shows like The Crown or The Family Man and want to maximize every inch of promotional space to make sure consumers watch.
Jayesh is the Founder of Spectra Global Pte Ltd which focuses on converting data into revenue by offering one-demand Chief Data Officer (CDO) and Chief Growth Officer (CGO) services to digital ventures, marketers, and media companies.
Gireesh, the Founding Collaborator of Fractal Dataminds , brings over 20 years of experience in Financial Markets and Technology in a wide-ranging career that spans selling corporate bonds and building data science teams.
Product and tech teams on the other hand are trying to get the best mix of rails right on the first scroll of the app to make content discovery seamless and personalized so that consumers are able to discover value in the platform that goes over and above what they initially came for.
Any OTT platform will have its content highs and lulls, so personalization is the glue for retention and revenue.
So how can OTT platforms strike a balance between content push and data-driven personalization pull?
Enter Economic Graph Models for Personalization!
Imagine your OTT streaming app as a giant jigsaw puzzle.
Each piece represents a user of the app, and each user has unique preferences, like the types of movies or shows they enjoy watching.
By modeling competing incentives in your product offering with an economic graph model and aligning these insights with personalization powered by multi-objective recommender systems, you can connect the right users with the right content at the right time so that you strike an impeccable balance between advertising impressions and subscription or retention nudges.
For example, if a user got drawn to your platform due to a promotion on a recently released action movie, the economic graph model can serve retention nudges at the end of their first watch session. These nudges will prompt them to explore content that was enjoyed by other users who came for the same reason, thus increasing the chances that they will explore other value propositions that the platform can offer them.
Additionally, the economic graph model helps the app discover when is the right time to encourage the exploration of new content v/s the exploitation of familiar content.
It can identify trends and patterns among different groups of users, helping the app understand which shows or movies are popular or gaining traction. This information can then be used to curate content collections and promote new releases to the right target audience.
So, why is this almost back-to-basic type adage “connecting right users with the right content at the right time” difficult to achieve?
1. Technical Team Expertise: Implementing economic graph models requires a skilled team with expertise in data science, machine learning, graph analysis, and probabilistic programming. Finding and retaining talented professionals with the right skill sets can be challenging, particularly given the high demand for these skills in the industry.
2. Data Pipeline Bottlenecks: Building the necessary infrastructure to support economic graph models can be resource-intensive. Data pipelines to handle large volumes of data and perform complex computations efficiently are a key bottleneck.
3. Data Harmonization and Preparation: Economic graph models require high-quality data from various sources to provide meaningful insights. However, data collected from different systems and databases may vary in terms of accuracy, completeness, and consistency. Integrating and harmonizing disparate data sources can be complex and time-consuming, requiring data cleaning and preprocessing efforts.
4. Empowering the Product Team to Drive the Process: Implementing economic graph models may require organizational changes and alignment with content and marketing teams.
5. Continuous Iteration and Improvement: Economic graph models require continuous iteration and improvement to adapt to changing user preferences, evolving trends, and emerging technologies. Establishing a feedback loop, monitoring model performance, and incorporating user feedback is vital to ensure the accuracy and effectiveness of the models over time.
What best practices exist to implement this successfully?
Running a successful economic graph model pilot requires careful planning, execution, and continuous improvement.
1. Clearly define the measurable metrics to evaluate its performance. Specifically, we recommend –
- Focussing on MEVV (Monthly Entertainment Video Views) and DEVV (Daily Entertainment Video Views)
- Estimating the sensitivity of revenue (such as overall growth, subscription vs advertising revenue mix) to MEVV and DEVV
- Estimating the sensitivity of MEVV and DEVV to user and session level behavioral metrics, including, but not limited to CTR, Watch Time, Time to First Engagement and Time to Next Visit helps establish a well-calibrated economic graph with estimated sensitivities between these metrics.
2. Secure stakeholder buy-in and establish cross-functional collaboration. A good starting point is to pre-allocate the personalization rails accessible to the product team.
3. Start with a focused use case that can show results in 3 months. We recommend improving the watch time of the original series
4. Ensure data readiness with a robust infrastructure to support the pilot, including data storage, processing, and analysis capabilities. We recommend the adoption of cloud-native data warehouses like Snowflake and robust MLOps tools like Metaflow/Ray.
5. Implement an agile and iterative process that allows for rapid experimentation, learning, and improvement
6. Create a culture of curiosity and healthy skepticism in the product and tech teams where they are encouraged to take bold decisions whose impact can be measured transparently, rather than focus on defending their past decisions.
These techniques were evangelized by disruptors like Spotify, TikTok, ShareChat, Netflix, Youtube, and LinkedIn and have been instrumental in their ability to gain market share at the expense of incumbent media behemoths.
Some of the incumbents, now forced to innovate at an accelerated pace to compete and retain market share are jumping onto this bandwagon and are reaping the rewards. For example, at a major SEA OTT Platform, these techniques demonstrated that a 25% increase in the reach of personalized rails could grow subscriptions by 8% MoM.
While content serves as the foundation of an OTT platform, personalization sharpens and enhances the experience because individual preferences have diversified faster than content production cycles.
Harnessing the strength of both, platforms can captivate audiences, ignite the fire of engagement, and emerge as true champions in the battle for viewers’ hearts and screens.
Valar Morghulis – all content must serve, but personalized experiences shall prevail.
Fractal data and Spectra Global have struck a partnership to enable OTT players to develop personalization in the way Netflix, and Lyft do. We help product teams balance out complex decisions with economic graph models, build data pipeline efficiency and drive cross-functional alignment. We also train product teams for future success. If you are interested in building the in-house capabilities described in this blog, get in touch with us at [email protected].