Online Business Models, Digital Ads, and User Welfare

Daron Acemoğlu, Daniel P. Huttenlocher, Asuman Ozdaglar, James Siderius

TL;DR

This paper provides a game-theoretic framework to explain why social media platforms often favor "free" ad-based models over subscriptions, even when they decrease overall user welfare. By modeling a mix of rational ("sophisticated") and over-optimistic ("naive") users, the authors show that digital ads allow platforms to manipulate demand, leading to higher prices and lower utility for everyone—a market failure that can potentially be corrected by digital ad taxes.

What Problem Does This Paper Address?

The digital economy is dominated by "free" platforms funded by targeted advertising. While targeted ads can be more informative than traditional broadcast ads, they also provide a powerful tool for consumer manipulation.

The authors address a fundamental gap in economic theory: the lack of a framework that accounts for the dual role of digital ads as both informative (helping users find products they like) and manipulative (tricking users into overvaluing products). Specifically, they investigate how platforms choose their business models (ads vs. subscriptions) and how these choices impact "user welfare"—the actual benefit users get after accounting for the time lost to ads and the money spent on products.

Methods

The authors develop a microeconomic model featuring three types of agents:

  1. Firms: Producers who want to sell products and are willing to pay to advertise.
  2. Platforms: Intermediaries (like Facebook or YouTube) that provide entertainment and host ads.
  3. Users: Divided into Sophisticated (who understand how ads are targeted) and Naive (who underestimate the likelihood of "false positives"—i.e., they think a "good" signal from an ad is more reliable than it actually is).

Key Technical Components: * Information Structure: Ads are modeled as a Poisson process where the "ad load" (frequency) determines the probability of a user seeing a signal about a product. * Utility Function: A linear-quadratic utility model is used to represent user preferences for entertainment vs. product consumption. * Equilibrium Concept: The paper uses the (Perfect) Berk-Nash equilibrium, a concept designed for games where players may have misspecified models of the world (in this case, the naive users). * Mechanism Design: The authors analyze "separating equilibria" where platforms offer different plans (e.g., a "Premium" ad-free plan and a "Free" ad-heavy plan) to sort users by their level of sophistication.

Datasets

This is a theoretical economics paper; it does not use empirical datasets for training or evaluation. The results are derived through mathematical proofs and comparative statics based on the model's primitives.

Key Insights and Ideas

  • The Manipulation Profit: Platforms find ad-based models more profitable specifically because they can extract surplus from naive users. Because naive users think ads are highly informative, they are more "tolerant" of high ad loads and more likely to over-purchase products.
  • Negative Spillovers on the Rational: Even if you are a "sophisticated" user who isn't fooled by ads, you still lose. Why? Because the inflated demand from naive users allows firms to raise the market price of products, which the sophisticated users must also pay.
  • Competition "Softens" the Market: Counter-intuitively, competition between firms or platforms doesn't necessarily help users. Digital ads create "endogenous horizontal differentiation"—they make identical products look different to different users based on the ads they see. This gives firms more market power, allowing them to charge higher markups than they could in a world without ads.
  • The Paradox of Informativeness: As ads become more informative (better targeting), the platform is actually more likely to adopt an ad-based model that targets naive users, which can lead to a discontinuous jump downward in total user welfare.

Key Results

  • Welfare Loss: In many scenarios, user welfare is lower in the ad-based equilibrium than it would be in a world with no platform at all.
  • Market Segmentation: When platforms can offer a menu of plans, they typically "segment" the market. Sophisticated users are moved to subscription plans, while naive users are targeted with high-intensity ad loads.
  • Policy Intervention: The authors prove that a flat tax on digital ad revenue can improve welfare. By making ads less profitable for the platform, the tax encourages a shift back toward subscription-based models, reducing the "manipulation" effect and lowering product prices.
  • Second-Best Implementation: While a "First-Best" (perfect) outcome is impossible because the government can't see who is naive, a "Second-Best" outcome can be achieved through a combination of nonlinear ad taxes and product subsidies.

Limitations and Future Work

  • Static Content: The model treats the "entertainment content" as a fixed value. In reality, the drive for ad revenue often influences the type of content produced (e.g., clickbait or addictive algorithms), which the authors suggest as a vital area for future research.
  • Social Networks: The model does not account for network effects (joining a platform because your friends are there), which could further complicate the platform's pricing power.
  • Data Externalities: The paper simplifies how user data is collected. Future work could explore the "privacy-welfare" trade-off more deeply, looking at how the sale of personal data feeds back into the manipulation loop.
  • Psychological Factors: The authors acknowledge that "naivety" is a simplified proxy for complex behaviors like digital addiction and mental health impacts, which warrant their own dedicated models.

Summary updated

BibTeX
@misc{acemoglu2024online,
  author = {Daron Acemoğlu and Daniel P. Huttenlocher and Asuman Ozdaglar and James Siderius},
  title = {{Online Business Models, Digital Ads, and User Welfare}},
  year = {2024},
  doi = {10.2139/ssrn.4978696},
  howpublished = {SSRN},
}

Loading related content...

Figure