Media Platform Flywheel: The Business of Media - Ritvvij Parrikh Media Platform Flywheel: The Business of Media | Ritvvij Parrikh Humane ClubMade in Humane Club
table of contents

Media Platform Flywheel: The Business of Media

Hi there,

I’m excited to introduce ‘Media Platform Flywheel,’ a series focused on understanding and comparing modern, compute-heavy, algorithmic platforms with traditional, compute-light digital media and news products.

Objectives: This series aims to help forward-thinking media leaders:

  • Identify Core Issues: Understand that many digital media products face a fundamental value proposition problem that can’t be fixed with superficial solutions.
  • Seek Root-Cause Solutions: Each of the flywheels in this document look for ways to address the root causes by developing algorithmic platforms that prioritize growth and journalistic ethics over cost-cutting and clickbait.

Disclaimers: While this series won’t provide exact solutions, it will offer a framework that you still need to apply to your specific situation.

  • Additionally, the series is written as working notes that will be continuously updated and refined on a weekly basis.

Call for Contributors: I don’t have all the expertise needed to complete this series, so I’m looking for contributors to help bring this project to life.

Your feedback and contributions are welcome.


1. Grounding in Reality

1.1 Facts

The series assumes that the following statements are first principles on which everything is built. If you disagree with any of these principles, please reach out and let’s brainstorm.

1.1.1 Category to operate in

Fact 1.1.1.1

Growth mandates expansion beyond news: Firms aiming for growth, not just cost-cutting, need to shift from news to content.

  • No growth potential (in India?): Adjusting The New York Times’ 2023 revenue ($2.1 billion) for purchasing power parity, the largest potential revenue from news in India is about $556 million (₹5000 crore). Once this point is reached, growth is limited. Most news firms know this and have moved into entertainment and other areas.
  • Baseline news vs. content firms: News is a subset of content. Changing the category can elevate a firm to a different league. For example, The New York Times, the most successful news company, had a 2023 revenue of $2.1 billion. In contrast, X.com, an underperforming content firm posted $3.3 billion in 2023. Facebook’s 2023 revenue exceeded $200 billion, while Netflix reported $31 billion.
  • Interest in news is declining.

Fact 1.1.1.2

Yet, bet long on news: AI disruption won’t eliminate the human need for news and entertainment. The key question is whether news companies will continue to provide news.

1.1.2 Know thy competitor

Fact 1.1.2.1

Competition: All media—news, OTT, social, search, ChatGPT, Wikipedia, and blogs—vie for users’ attention and ad dollars, blending entertainment (cat videos, memes, etc.) with accurate information.

1.1.3 Value proposition of platform

Fact 1.1.3.1

The promise of Media: “IF YOU GIVE ME FIVE MINUTES OF YOUR TIME, I WILL GIVE YOU CONTENT THAT IS WORTH YOUR FIVE MINUTES.”

Fact 1.1.3.2

Convenience matters: Users prefer platforms offering the best value for their effort or money. Algorithmic platforms excel here, attracting significant capital and talent.

  • Unbalanced media platforms see low engagement and poor ad monetization. In India, average time spent on Twitter is 20 minutes, Instagram 29 minutes, YouTube 19 minutes, but news sites get only a few minutes.

Fact 1.1.3.3

People are willing to pay extra for premium safe spaces with seamless experience.

  • Consumers: Apple builds walled gardens—no random developer can publish iOS apps. While this can be detrimental to the open web, it ensures Apple devices are secure, privacy-friendly, and safe for various users, including children.
  • Brands: All brands want to advertise in premium spaces.

Fact 1.1.3.4

In subscription, comprehensiveness matters. Amazon offers Amazon Prime with free shipping, discounts, music, and movies. YouTube’s Premium Subscription is ad-free for a small fee. Google One Subscription provides extensive storage across Google services. These comprehensive offerings drive subscriptions. Yet, most news products can only offer premium content.

1.1.4 Value proposition of content on the platform

Fact 1.1.4.1

Selection of content determines platform’s value, affecting both users and expenditure on content.

  • Exclusiveness matters: Platforms lose audiences without exclusive content. For example, Disney + Hotstar in India lost 30% of its subscribers after Jio Cinema acquired exclusive IPL rights.
  • Comprehensiveness matters: Top content platforms need to hold most of the information audiences seek. We frequently visit search engines, social media, and ChatGPT, but only occasionally specialized sites like Coursera.
  • It can get expensive: Content creation and acquisition are costly. Netflix spent about 40% of its 2023 revenue (approximately $13 billion out of $33 billion) on content and plans to increase this by 30% to $17 billion in 2024.

Fact 1.1.4.2

High production quality elevates the content’s premium value. We watched the Spider-Man movie in the 1990s, then again in the 2000s, and once more in the 2010s. Although the story remained broadly the same, we revisited it for the significantly enhanced production quality, especially the graphics.

Fact 1.1.4.3

Short-form is winning on the Internet

While collectible teapots, shoes, and clothes provide gratification through ownership, media content—whether OTT, theme parks, news, or books—requires both time and money for gratification. To expedite the dopamine rush, content should be in short form.

  • This is also the reason why Tiktok grew so fast and why YouTube and Instagram are pushing for Shorts and Reels respectively.
  • In comparison, long-form or immersive content doesn’t offer the same quick gratification to a vast number of people.
Copied

Fact 1.1.4.4

The Most Effective Format Innovation: Axios’s Smart Brevity format proved to be the non-sexy innovation audiences loved, leading to a $525 million exit for its founders and investors—a rarity in the news industry.

  • Over the last decade or two, multiple format innovations have struggled to scale as consistently. Vox introduced explainer cards. Some experimented with structured journalism. The New York Times pioneered immersive journalism, while others ventured into data journalism. Meanwhile, The New Yorker continued to produce long-form content. I’ve personally spent years cracking modular journalism at scale.

1.1.5 Monetization

Fact 1.1.5.1

Bet long on advertising: AI disruption won’t eliminate the human need for ads. The key question is whether digital media products can monetize with ads profitably?

Fact 1.1.5.2

Subscription is a revenue diversification strategy, at best: Many companies are now focusing on subscriptions, but this is situational.

  • Google and Facebook, two ad-driven giants, generate over $200 billion in revenue each. Netflix, the largest subscription service, makes only $33 billion. Netflix announced its own ad network in May 2024.

1.1.6 How to operate the business

Fact 1.1.6.1

Bet short on the current way of doing things: Digital media is a service businesses built on product businesses—search, social, ad networks, and increasingly, Generative AI.

  • This is akin to airlines and phone companies relying on products from Boeing and Huawei.
  • Service businesses manage annual P&L cash, but product companies take most of the profit due to their intellectual property.

Fact 1.1.6.2

Algorithmic platforms: Amazon changes prices 2.5 million times a day, and most U.S. equity trades are algorithmic, anfd of course social, search, ChatGPT, OTT take 100% of their decisions algorithmically.

Fact 1.1.6.3

Companies face revenue loss when they can’t compete against algorithms: Google takes around 50% of the topline from indirect ads.

1.1.7 Working on AI

Fact 1.1.7.1

Media adopting AI is a big change. Switching to AI for a media company is as transformative as a car company transitioning from petrol to electric vehicles (EVs). It’s not just about replacing the engine. An EV is a drastically simpler and more efficient product than a petrol car. The entire product, factory floor, and team have to be retooled, retrained, and upgraded.

Fact 1.1.7.2

AI is a big change for everyone.

Challenges in Adopting AI

  • Building Judgment: AI implementation is costly and complex. Months can be spent developing a model, only to find that an underlying hypothesis was wrong, necessitating a complete restart.
  • Change Management: Critical functions within the organization must adapt, upskill, or retrain to integrate AI. This often requires changes in the scope, mandate, and power of various stakeholders, which is never easy.
  • Expanding Talent Pool: The market has a limited pool of AI/ML talent, and many companies are competing for the same expertise.
  • Data Quality: Do you have clean, unbiased datasets to train your machine learning models on?

Industry Insight: Even giants like Microsoft admit they are not fully prepared to compete against leaders like OpenAI and Google. Check out this CTO admission to Satya Nadella and Bill Gates.

Copied

1.2 Decisions That Impact

So far, I’ve found four decisions that plague digital media products, though there might be more, which we’ll add as we discover them.

  • Use of Divisional Organizational Structure
  • Operating Two-sided Platforms with Business Rules
  • Editorial Control of Distribution
  • Lack of Compute-Heavy Infrastructure Required to Adopt AI Meaningfully
  • ?
  • Have a hypothesis about more decisions? Reach out.

Any attempts at fixing the foundational value proposition problem without addressing the below decisions first will either be ineffective in kickstarting growth, or, if you use AI, will only amplify what’s not working and worsen the situation.

1.2.1 Use of Divisional Organizational Structure

Pre-requisites:

Offline businesses, run by humans, are divided into nearly autonomous units or divisions. Each division is led by someone responsible for their outcomes, often focusing on their specific unit with predefined business rules.

  • In contrast, algorithmic platforms are unified. There is only one Google Search, one YouTube, one Netflix, one Amazon, one Twitter, and one Substack that scale across all user needs and geographies.

Most digital media businesses mimic their offline equivalents by using a divisional organizational structure. They use the house of brands strategy, and within each brand, they create divisions, allowing each division leader to implement specific business rules.

Why It Matters: Using a divisional organizational structure in digital media impacts the ability to deliver what users want, gather data, create a compelling subscription model, and fail to generate network effects.

FAQs

Why Do Business Leaders Prefer Divisional Structures?

Is Divisional Structure Universally Bad?

What Are the Problems with Divisional Structure When There Is Overlap?

Is This Problem Only in Digital Media?

How to Identify if an Organization Has Escaped Divisional Structure?

1.2.2 Operating Two-Sided Platforms with Business Rules

Pre-requisites:

Running digital media with static business rules is like operating Netflix as if it were AMC Theatres (or PVR Cinemas). The world of digital media thrives on dynamic, algorithmic platforms, not rigid, manual processes.

Why It Matters: When competing against algorithmic platforms, using static business rules is like bringing a knife to a gunfight.

The Details: Digital media operations involve dynamic two-sided platforms with publishers on one side and algorithmic platforms (like search engines, social media, and ad networks) on the other. These transactions are continuous, where slower reactions (often from publishers) can mean lost revenue. Key operations include:

  • SEO Optimization: Attracting traffic from search.
  • eCPM Negotiation: Working with Google Ad Manager and other ad networks.

Balancing engagement and revenue is crucial:

  • Content and Ads: Serving the right content and ads is an algorithmic challenge.
  • Monetization: Deciding which monetization options (ads, subscriptions, affiliate marketing) to show to users.
  • Journalistic Responsibility: Balancing journalistically relevant topics with audience preferences.

Digital media platforms need scalability across user segments and geographies, making it impractical to manage business rules for each category.

FAQs

What are Business Rules?

Why do Business Leaders prefer static Business Rules?

Are Business Rules universally bad?

Why Business Rules fall short in Two-Sided Platforms?

1.2.3 Editorial Control of Distribution

Many digital media products continue to ask their editorial teams to own distribution — homepage, push notifications, newsletters, social media, etc.. However, this clashes with modern industry standards

Why It Matters: Manual distribution serves users less of what they want and need. This practice also injects bias into clickstream data, affecting editorial analytics and algorithm development.

FAQ: No, we’re not dropping editorial standards. Instead, editors should collaborate with algorithms to moderate for:

  • Relevance and significance over click potential
  • Trust and ethical considerations
  • A diverse mix of stories, including less popular yet important content

More on this later.

1.2.4 Lack of Compute-Heavy Infrastructure Required to Adopt AI Meaningfully

Aiming to use Artificial Intelligence at scale without building foundational infrastructure is like putting the cart before the horse. It won’t work. Here’s 14-point checklist for evaluating the compute-heavy readiness of your infrastructure.

Cart before the horse.

1.3 How Decisions Impact

For this, let’s see how average digital media product stacks up an algorithmic platform, say X.com.

1.3.1 Low direct traffic, organic app installs, login rates, and ability to charge for subscriptions

Pre-requisites:

Historically, digital media businesses didn’t invest in recommender systems and programmatic ad targeting platforms. This meant they could only sell access to relevant audiences by creating media properties targeting predefined categories.

As a result, they adopted the House of Brands strategy, launching multiple media properties for different segments like business, women and lifestyle, hard news, college students, and expats. Here’s how it looks:

  • Vox Media Network: New York Magazine, Intelligencer, The Cut, Vulture, The Strategist, Grub Street, Curbed, The Verge, Vox, SB Nation, Polygon, Eater, Punch, and more.
  • Gannett: USA Today, The Arizona Republic, Detroit Free Press, and others.
  • The India Today Group: India Today, Aaj Tak, Daily O, Cosmopolitan, Lallantop, SoSorry, OddNaari, among others.
Copied

This resulted in the media house’s value proposition getting fragmented across brands.

Over time, the problem worsened: Search and social platforms took over distribution, promoting specialized brands in their algorithms.

  • Media executives responded by adding more websites and apps.
  • Some broke out of their niches; for example, lifestyle websites began covering hard news stories. The goal was to dominate search and social results with their media properties.

Problem 1: Low Brand Recall. Unlike X.com, digital media products struggle with low direct traffic, organic app installs, and login percentages due to diffused brand propositions and low top-of-mind recall.

  • For shopping, people go to Amazon. For videos, they go to YouTube. For music, they choose Spotify, YouTube, or Apple Music. For the best phone, they buy iPhone or Samsung. These preferences are consistent across different geographies and economic tiers, from New York to Mumbai.
  • However, no single media property dominates in news, making it hard for audiences to know where to go for news.

Problem 2: Can’t Justify Subscription. Comprehensiveness of offering converts online users to pay for subscriptions. Unlike X.com, digital media products struggle to charge for subscriptions because their value proposition is spread across multiple brands.

1.3.2 Low DAU/MAU ratio

Pre-requisite:

When users are idle, they often check X.com multiple times a day. This behavior boosts the daily active users (DAU) to monthly active users (MAU) ratio on X.com. Here’s why:

  • X.com keeps its feeds constantly updated, offering new content to consume.
  • It offer short-form content with options for long-form reading.

In contrast, digital media products struggle to achieve this because:

  • Editors can’t update feeds as frequently, reducing the incentive for regular check-ins.
  • Users need to open one long-form article at a time, making information consumption slower.

Why It Matters: This metric is crucial because print advertising rates are higher due to print newspapers assuming a 100% daily to monthly user ratio.

1.3.3 Low Avg. Time on Site

Pre-requisites:

X.com will always have a better selection of content compared to digital media products. But even if the content on X.com and a digital media product were similar, where would you consume it?

Finding What You Want: X.com is personalized, making it quicker and easier to discover the content you want.

  • Many digital media products rely on editors for distribution. They cater to a broad audience, often making it harder for users to find specific stories.
  • Print newspapers personalized content by launching city editions. Ironically, today’s digital news homepages are less personalized than those city edition front pages.

User Experience: The user experience on digital news platforms isn’t as intuitive as on X.com.

  • Human editors struggle to maintain a balanced mix of stories. Hence, product teams add static widgets as placeholders for various divisions — news, opinions, subscription, sponsored content, etc.
  • This method might have worked in the print era, but it falls short online. The internet’s dynamic nature requires feeds to either create or react to trends effectively.

News homepages are cluttered with widgets for various sections, leading to higher bounce rates and accidental clicks.

This difference places the cognitive load and choice fatigue on users when navigating news sites.

In contrast, algorithmic platforms—social media, search engines, and OTT services—present users with a simple, sorted list of content to choose from.

Meanwhile, on algorithmic platforms, this burden is shifted to the recommender algorithms, making it easier for users to find what they want.

GIF Credits: Ralph Ammer

Copied

1.3.4 Overall Business Struggles To Reach its Potential

Pre-requisites:

“Low margin businesses (eg many media biz) often create an internal culture of ‘survival of the fittest’, where different divisions compete, viciously, to eke out another 0.5% profit. Makes digital transformation even harder…”

Alex Watson, Ex Head of Product at BBC.

Historically, most digital media businesses originate from offline media and retain offline best practices. Thus, they establish individual divisions within each media property for each revenue model they operate.

Why It Matters: Media is a real estate business. Every pixel on a website or app should be optimized for engagement or revenue. News products struggle because their interfaces are divided into multiple widgets from different divisions. Division leaders negotiate for space, size, and placement, which are implemented through business rules.

  • Can’t optimize for revenue: Division leaders compete for top-of-funnel visibility, a cost center, rather than optimizing conversion rates, a revenue center.
  • Can’t optimize for engagement: Websites and apps reflect internal negotiations rather than being user-first, unlike the simple lists or feeds of scalable online platforms.
Old meme about impact of divisional organizational structure in Microsoft

1.3.5 Poor Selection of Content

Pre-requisites:

X.com caters to a wide range of interests, unlike digital media products.

Diversity of worldviews: Consider Vikramaditya: socially liberal, economically libertarian, and conservative about health due to family history. His interests are diverse and a black swan event like Covid-19 can change him. X.com can serve Vikramaditya’s changing needs because there will be others like him. In contrast, a digital media product publishes only that content that mirrors the judgment and values of the editor.

Comprehensiveness of Coverage: Vikramaditya lives in Mumbai but has clients in Israel and wants to stay abreast with news from Israel. X.com can offer information from Israel, which many Indian news platforms lack due to limited reporter presence.

Breaking News Speed: Breaking news often appears first on X.com, posted by reporters, governments, and experts. Digital news products pick up these stories later.

Insight Depth: Experts on X.com provide deep insights into niche topics. For example, detailed maps explaining Bengaluru’s water scarcity are available on Raj’s X.com account, unlike in traditional news outlets.

Access to Celebrities and Experts: X.com allows direct interaction with experts and celebrities. For instance, discovering and engaging with Nassim Nicholas Taleb on X.com was a unique experience for me that no news product could offer.

Short and Long-Form Content: X.com provides both short-form (tweets) and long-form content (threads), while news products lean towards long articles. Short-form content is more effective online.

While collectible teapots, shoes, and clothes provide gratification through ownership, media content—whether OTT, theme parks, news, or books—requires both time and money for gratification. To expedite the dopamine rush, content should be in short form.

  • This is also the reason why Tiktok grew so fast and why YouTube and Instagram are pushing for Shorts and Reels respectively.
  • In comparison, long-form or immersive content doesn’t offer the same quick gratification to a vast number of people.
Copied

1.3.6 Struggle to extract ROI from investment in content

Pre-requisites:

Editorial control of distribution reduces the digital media product’s ability to leverage long-tail content.

Why This Happens: Editors can only feature stories they know about. What’s top of mind tends to be current news. Thus, most homepage content is a small fraction of total content produced, often less than 24 hours old.

  • Google’s SEO algorithm prioritizes these recent stories, recommending them more.
  • Google’s SEO expects ‘new’ content from ‘news’ websites, de-emphasizing older stories.

Why this matters:

  • Short shelf life stories reach their maximum PV potential quickly, while long-tail stories can take months to peak. In comparison, long-tail stories often deliver 500-700% higher pageviews.
  • Over time, the entire system—editorial analytics, clickstream history, editorial workflow, brand, ML algorithms, etc.—optimizes for fast-moving, short shelf life content, trapping us in a vicious cycle. This reduces the supply of content, negatively impacting user experience.

Sometimes, the bias is so strong that even long-tail stories behave like short shelf life stories and cap out quickly.

1.3.7 Reduced preparedness to rollout AI

Pre-requisites:

Because of fragmented value proposition, digital media products struggle to collect first-party data. In comparison, X.com collects better user information for several reasons:

  • Users must log in.
  • They provide their name, bio, location, and date of birth.
  • Social platforms require phone number verification.
  • Platforms like X offer low-priced subscription services.
  • Users actively connect with and engage with others.

Even if digital media products could collect data, use of divisional organizational structure hinders execution:

  • The main issue is the need for extensive collaboration among different parties, leading to accidental complexity. As systems expand, this complexity only grows, slowing down execution.
  • As a result, digital businesses with divisional structures frequently find themselves bundling, unbundling, and re-bundling teams and technologies to achieve only minimal efficiency gains.
  • Why It Matters: This structure makes it tough for companies to optimize costs, execute quickly, and compete with algorithmic platforms. It also complicates cross-selling across divisions, which could reduce customer acquisition costs (CAC).

Even if execution of data collection was flawless, editorial control of distribution introduces bias into clickstream data, complicating the rollout of user-first algorithms. Below are two types of biases:

  • Position Bias: This bias occurs when users choose an item that requires more effort to reach, like opting for a cup of chai ten steps away over a cup of coffee right in front of them. In digital terms, it’s crucial to measure how far users scroll before clicking on an article to accurately de-bias the position.
  • Promotion Bias: Featuring a story on the homepage or promoting it via social media or push notifications introduces promotion bias into clickstream patterns. This must be removed for models to reflect actual user needs.

1.3.8 Because of AI, X.com is able to empower users to influence the narrative

In most news products, users can write letters to the editor or leave comments. In comparison, platforms like X.com empower users to set the narrative more directly:

  • Nerds can participate in Community Notes to fact-check information. The community can upvote or rank these fact-checks. In news products, such engagement is buried in comments that often go unnoticed.
  • On X.com, users can make hashtags go viral for causes they care about, getting their hashtag to appear in the trending section.

2. Guiding Policy

2.1 Invest In Fundamentals

Media has always been a celebrity-driven industry, often emphasizing ethics, taste, and style. While important, Media Platform Flywheel focus solely on the business of media.

In a fast-changing world, the Media Platform Flywheel helps identify what will remain true 10 years from now.

“I very frequently get the question: ‘What’s going to change in the next 10 years?’ And that is a very interesting question; it’s a very common one. I almost never get the question: ‘What’s not going to change in the next 10 years?’ And I submit to you that that second question is actually the more important of the two — because you can build a business strategy around the things that are stable in time. … In our retail business, we know that customers want low prices, and I know that’s going to be true 10 years from now. They want fast delivery; they want vast selection. It’s impossible to imagine a future 10 years from now where a customer comes up and says, ‘Jeff I love Amazon; I just wish the prices were a little higher,’ or ‘I love Amazon; I just wish you’d deliver a little more slowly.’ Impossible. And so the effort we put into those things, spinning those things up, we know the energy we put into it today will still be paying off dividends for our customers 10 years from now. When you have something that you know is true, even over the long term, you can afford to put a lot of energy into it.”

Jeff Bezos on E-commerce Flywheel

2.2 Adopt Cross-Functional

2.2.1 Be more accepting of Failures

Newsrooms are obsessed with accuracy, striving for perfection and minimizing errors. However, both AI and reporting are inexact sciences where mistakes can happen. When they do, newsrooms must issue corrections.

  • Reporting in newsrooms is about documenting the best version of reality they know, which can be inaccurate.
  • AI is a probabilistic science, and mistakes can happen.

2.2.2 Have to invest in Learning and Development

Newsrooms will have to invest in growing AI literacy within the newsroom. This will require them to invest in L&D budgets and foster a safe environment where innovation can thrive.

Here’s a proven method to facilitate cultural shift

2.2.3 Be more accepting of Cross-Functional Solutions

In today’s rapidly evolving world, relying solely on your primary skill set won’t cut it anymore. Programmers can’t solve everything with code, and designers can’t fix everything with design.

The future demands cross-functional teamwork, where diverse skills come together. Improved collaboration is now essential for solving complex problems.

Want to lead cross-functional teams without deep expertise in their fields? Here’s a strategy: learn the primary “thinking techniques” of different professions. These techniques shape how professionals solve problems and conceptualize ideas. For example:

  • Predictive & Mathematical Thinking: Economists, Data Scientists, Investment Bankers
  • Attention to Detail: Accountants, Quality Assurance Testers, Lawyers
  • Creative Problem-Solving: Engineers, Management Consultants, Product Managers
  • Influencing & Negotiation: Salespeople, Politicians, Diplomats
  • Curating & Organizing Information: Editors, Information Designers, Information Architects
  • Skepticism & Inquiry: User Researchers, Reporters

Training yourself in these techniques helps you understand and empathize with your team’s perspectives. Remember, these are broad generalizations, and individuals may vary. Fostering collaboration among diverse perspectives also brings immense value.

Copied

2.2.4 Be more accepting of Articulating Details To Others

To outsiders, work from another function might seem like random trial and error. However, this perception misses the nuanced decision-making and expertise involved. To facilitate cross-functional collaboration, it’s crucial to clearly and logically articulate the underlying system. This is harder than it seems.

2.2.5 Be willing to be in the details

The devil is always in the details. What seem like minor cracks from a 10,000-foot view turn out to be insurmountable canyons when it comes to implementation. Hence, it is critical for functional leaders to put in enough cognitive effort to understand and appreciate the details of the other functions they collaborate with.

Charles Eames

You cannot delegate understanding.

Brian Chesky’s Interview with Lenny (Paraphrased): Leaders need to dive into the details. This means fully understanding what your team is working on, including processes, challenges, and progress. This comprehensive insight enables you to make informed decisions, offer meaningful guidance, and trust your team to execute. Without this understanding, you can’t accurately assess their performance.

Jason Fried of Basecamp shares a similar philosophy. At Basecamp, they personally perform tasks of a new role before hiring someone. This approach helps grasp the job’s details and nuances, ensuring they understand what success looks like.

Copied

2.3 Platform Business

There are two clips from the movie ‘The Founder.’ In the movie, Ray Kroc is in a position similar to that of digital news companies in relation to algorithmic platforms like search, social, and ad networks.

Then Harry makes him realize that despite providing the system, training, and operational know-how, Ray Kroc is only earning a 1.4% commission on a 15 cent hamburger. This is very similar to how news firms provide the nuance and operational know-how but get pennies on the dollar from indirect programmatic advertising.

Harry makes Ray realize that he needs to own the assets, the platform, and the quality standard, and finally kickstart a flywheel.​⬤

In the case of digital media, the platform (algorithmic distribution) provides direct access to audiences and advertisers. The asset is the first-party data, network effects, and the brand. The quality standard is your Trust and Safety standards, and all of this together kickstarts a flywheel.

You too will need to start operating as a platform. Hence:

  • House of Brands Media Groups: On your website post-login and on your apps, consolidate user attention down to one brand and distribute all content of every brand in the group.
  • Smaller Media Houses: Build coalitions to share content.

2.4 Run Ops with Algorithms

Algorithms offer a self-learning, self-optimizing system to address dynamic platform challenges:

Why It Matters: Investing in self-learning, self-optimizing algorithmic systems not only boosts revenue but also makes it cost prohibitive for your competitors to replicate your success.

2.4.1 What About Edge Cases?

Pre-requisites:

Algorithms, while powerful, can produce unexpected outcomes due to their complexity and the opaque nature of some machine learning models.

  • Upside: If humans along with business rules can make 0.1 million decisions a day, algorithms can make 100 million decisions.
  • Downside: Even the best algorithms will occasionally make errors (say 0.1%), which on a base of 100 million will be 0.1 million bad decisions. Ensuring these errors don’t result in significant reputation loss requires human oversight. That’s where content moderation comes in!

Human intervention is necessary to:

  • Identify and Correct Errors: Analyze and stop bad decisions until automated systems can be adjusted.
  • Train Models: Tag training data to improve future algorithmic decisions.

On a lighter note:


3. Coherent Actions

3.1 Media Platform Flywheel

Here’s a causal loop diagram at a 30,000ft view:

Media platform flywheel

How to read the diagram:

Media firms serve audiences with high-quality content and relevant ads:

  • High-Quality Content: Worth the audience’s time.
  • Relevant Ads and Deals: Beneficial to the audience.

In return, audiences offer:

  • Attention: Measured by daily to monthly user ratio and time spent on the site, monetized through ads.
  • Effort: Interactions like ad clicks and surveys, monetized through performance-based advertising.
  • Money: For valuable subscriptions or purchases.

This increases the platform’s value proposition — appeal and utility.

  • Marketing exclusive content, audiences, brands, or creators improves value proposition.

Over time, platforms expand their network of users, creators, and brands through:

  • Retention: Keeping current users happy.
  • Acquisition: Attracting new users with a strong value proposition.

Feedback Loops

  • Creators and brands supply more content and ads.
  • Platforms gather data to refine algorithms, enhancing user experience over time.

Conclusion: Media business thrives on a give-and-take relationship where quality content and relevant ads drive engagement, fueling growth and monetization.


3.2 Editorial Flywheel

Pre-requisites:

To thrive, these content platforms invest in improving the quantity and quality of content. This ensures a wide selection for users, enhancing the platform’s appeal.

We propose an editorial flywheel to ensure a better selection of content that’s worthy of audience attention, at a lower cost.

All content platforms are marketplaces. Content creators are like shops in this marketplace, each offering their content as a product.

As a marketplace, the content platform must provide facilities to content creators. It includes:

  1. Incentive Profitably: Benefits designed to draw creators to the platform, profitably. Below are some of the ways that algorithmic platforms incentivize users:
    • Grants and Contracts: Google News Initiative and Meta Journalism Project fund news firms to produce content and fact-check misinformation.
    • Financial Rewards: Platforms like TikTok and Medium pay creators based on views and engagement.
    • Revenue Sharing: YouTube, X, IGTV ads, and Twitch share ad revenue with creators.
    • Legal Support: Substack’s Defender program provides legal aid to writers.
    • Subscription and Tips: YouTube, Substack, and Spotify offer tools for creators to charge for subscriptions or receive tips.
    • Content Purchases: Platforms like Spotify and OTT services buy exclusive content, e.g., Spotify’s $100 million deal with Joe Rogan.
    • Structured Content: Google and Facebook encourage websites to format content for easy integration into Google News and Facebook Instant Pages.
  2. Automation and CMS: All content platforms must provide a CMS to creators that attempts to automate as much as possible.
    • Google Photos algorithmically shares memories from the past as interactive photo stories.
  3. Production Tools: All content platforms provide creators with tools that will simplify the content creation process.
    • Instagram famously launched filters for photos and later added tools like Boomerang. Everyone now provides emojis and stickers. Apple offers Memoji.
    • This space will completely blow up with the advent of Generative AI.
  4. User Insights: Data to help creators understand what users are looking for.
    • Google provides Google Keyword Search and Google Trends to help creators study what is trending.
  5. News Gathering and Pre-Content: All content platforms provide some pre-content (raw materials for the final content) to help creators with supporting material.
    • Instagram provides tonnes of background music options as pre-content that users can add to their reels.
    • News companies provide subscriptions from wire companies, stock photos and database companies as pre-content to their writers.

Multiple teams of creators ensure quantity

  • Most creators in a news firm are on staff or paid contributors who reflect the judgment of the Editor.
  • An algorithmic platform doesn’t have on-staff creators. Instead, it focuses on creating an editorial flywheel that is attractive enough for creators to publish on its platform.

As a marketplace, the content platform must ensure trust and safety for audiences.

All creator content must pass through a Trust and Safety system before distribution.

  • When you upload a video to YouTube, it makes you wait until it can complete a plagiarism and a Not Safe For Work (NSFW) evaluation.
  • In news companies, editors verify stories from reporters and ensure consistency to their style guidelines.

Let’s deep dive into some of these building blocks:

3.2.1 Incentive Profitably

“Show me the incentive and I will show you the outcome.”

Charlie Munger
3.2.1.1 Financial Incentive

This is

PENDING (write about cost modeling, how to decide incentives, what to incentivize, and how to payout)

3.2.1.2 Quality Incentive

asdadasd

In the real world, the shelf life of products—like milk, cheese, grains, and soda—is determined by how long they remain edible. Similarly, the shelf life of a story should be based on how long its information stays valid or truthful.

Why It Matters: Today, news firms are trapped in a cycle of short shelf-life content. Writers are often judged by the number of stories they publish, not the long-term impact of those stories. Instead, consider this:

  • Short Shelf-Life Writer: Produces ten stories a day, each relevant for one day. This totals 200 potential story days per month.
  • Long Shelf-Life Writer: Produces one story a week, each relevant for 180 days. This totals 720 potential story days per month.

The Way Forward: Measure a writer’s productivity by the potential shelf life of their stories. Then it is the product or algorithms team’s responsibility to realize this potential shelf life into engagement by recirculating valuable long-tail stories.

  • This is akin to how, in a print newspaper, editors decided the front page, but the distribution team ensured circulation to all households.

3.2.2 Automation and CMS

3.2.3.1 Micro-decisions that Editors take

Today, content management systems (CMS) are ‘Systems of Record,’ meaning they store the final story after it has been planned, researched, reported, written, edited, produced, and packaged. With Generative AI Agents, CMS will also become ‘Systems of Actions,’ meaning they will ask editorial teams to store the steps they took to reach the final output. Over time, the agents will be able to perform similar tasks faster and better.

  • Venture Capitalist Tom Tunguz came up with the idea of ‘Systems of Record’ and ‘Systems of Actions’.

Why it matters: Without the transition towards ‘Systems of Actions,’ AI models won’t collect sufficient data to improve their tasks. “Better” could include accuracy, editorial judgment, bias, taste in production—everything that makes your brand authentically yours. Eventually, the newsroom with the best training data will outcompete competitors.

  • Another way to look at this is: Do today’s editorial analytics tools help editorial teams become better? They tell you what worked, but not why it worked or how it can be repeated. This is because today’s analytics tools are backward-looking, reporting on lagging parameters—pageviews, shares, full reads, etc.—instead of leading parameters—story choice, promotion choice, etc.

New Job Role: AI Training Specialists

Example: 15-31 micro-decisions that editors take when producing stories. Ideally, these should be stored in a way that AI can use it for training.

3.2.3.2 Low value Breaking News

In some topics with short shelf lives, users can get the information they need from brief summaries on social media or through Generative AI searches. They might not feel the need to consume more in-depth articles or videos, which require more effort.

However, comprehensive coverage is still essential for your news product. For these topics, using algorithms to nearly automate content creation is the best approach. Here’s how:

  • Break News Faster: Instead of people, have Generative AI monitor Live TV, social media, wires, and Google News, and rewrite breaking news for your website. This speeds up content creation and ensures timely updates.
  • Curated Opinions: If your teams constantly review expert opinions on Twitter and YouTube and then rewrite them, you can also nearly automate these processes using AI.

By automating these tasks, you ensure your platform remains comprehensive and up-to-date without overburdening your editorial team.

3.2.3.3 Convert News into Knowledge

People communicate to change the recipient’s information, action, ability, or belief. Currently, news primarily serves the audience’s information needs. However, most articles have a short shelf life, and there’s considerable repetition across articles on the same topic (articles are written in time continuum).

This approach made sense in the newspaper era because each day’s newspaper was discarded by the end of the day. But online stories remain accessible, so providing a coherent digest of everything that happened on a topic without redundancy becomes challenging.

Moving Up the Value Chain: News firms can enhance their offerings by transforming short shelf-life articles into Wikipedia-style digests, which aid knowledge assimilation. These digests organize information in a spatial format (space continuum, also called a map) and, in some cases, a space-time continuum similar to Wardley diagrams.

This method reduces redundancy and provides audiences with a comprehensive, cohesive understanding of topics. It’s a step towards transforming news from a collection of fleeting stories into a structured knowledge base.

3.2.3 Production Tools

New Job Role: ‘Full-Stack’ Reporters

3.2.5 News Gathering and Pre-Content

For this, news firms use following approaches:

  • News Gathering: Investing in more reporters, investigations, and access to datasets.
  • Attracting Contributors: As reach and prestige grow, more contributors join.
  • Syndication: Purchasing pre-prepared content.
  • User-Generated Content: Launching blogs for campus ambassadors and young professionals.
  • Rewriting Content: Digital teams rewrite content from Live TV, wires, and social media for search.

3.2.6 Multiple teams of creators ensure Quantity

PENDING (all material in this section is work-in-progress)

But today’s media landscape demands a broader approach to engage diverse audiences. Establishing multiple editorial teams under a unified brand, each bringing unique perspectives from the shared news gathering resources, can be highly effective.

Current Strategies in News Companies:

This model not only improves representation but also enriches content across various domains. It works best with recommender systems to manage the news homepage (instead of human editors), ensuring balanced representation of editorial voices.

Shifting Roles and Responsibilities: The Editor-in-Chief’s role shifts from directing editorial content to moderating content that could impact the publication’s reputation. Editorial product teams support this process by using insights from recommender systems to:

  • Identify Topics Worth Commissioning: Use AI to predict demand on content marketplaces and determine where to invest resources.
  • Estimate Story Lifespans: Allocate resources efficiently by predicting how long a story will remain relevant, ensuring costs are justified by returns.

This strategy creates a dynamic, adaptive editorial environment that aligns with contemporary media consumption needs.

3.2.7 Trust and Safety

Editorial teams are optimized for trust and safety, with editors bringing judgment and verifying content. Beyond this, additional investments are necessary:

  • Content Moderation: As distribution is increasingly managed via algorithms, the role of editorial teams shifts from selecting or curating stories to rejecting or stopping certain inappropriate stories from being distributed.

New Job Role: Content Moderators

  • Algorithms: Invest in algorithms to detect toxicity, NSFW content, plagiarism, and remove short shelf-life, stale content.
  • Crowd-Sourced Verification as Audience Engagement: Platforms need to be trustworthy and safe, but according to whose standards? News products could adopt X’s Community Notes for verification and audience engagement.
    • Audiences add Context: This feature allows audiences to add context to potentially misleading content. These notes, collaboratively written and rated for helpfulness by other users, aim to provide additional information or corrections, promoting transparency and accuracy.
    • Fostering Community Engagement: By allowing readers to participate in the verification process, platforms can create a sense of community. This involvement makes readers feel more connected to the news process.
    • Diverse Viewpoints: Encouraging the inclusion of various perspectives helps counteract bias and presents a more comprehensive picture of events. This leads to a more informed and discerning readership.

FAQs

What is Trust?

What is Safety?