Targeted Marketing: Seeding the Network with Positive Externality

TL;DR

In a world of "network effects," your neighbor's choice dictates your own. This paper solves the strategic puzzle of competitive marketing: if two firms have a fixed budget to spread their products, how should they balance product quality against initial free offers (seeding), and more importantly, whom should they target? The answer lies in the mathematical elegance of Eigenvector Centrality and tight adoption bounds.

The "Why": Beyond Simple Contagion

In traditional marketing, we often think of "viral" spread as a simple disease-like contagion. However, products with positive externality (like email, social apps, or online games) are more complex. They require coordination. You don't use Zoom because it's "contagious"; you use it because your colleagues do.

The authors recognize that consumers are rational yet myopic—they choose the product that gives them the highest immediate payoff based on what their immediate neighbors are doing. This creates a Local Coordination Game.

The Technical Core: Solving the "Where" Problem

The biggest contribution of this paper is moving beyond "How many seeds?" to "Which seeds?"

1. The Power of Matrix Exponentials

Prior work provided adoption bounds based on the number of seeds ($S_a,S_b$). This paper uses the actual adjacency matrix $A$ and degree matrix $D$ to derive a tighter bound:

$1-\frac{{\vec{1}}^T\exp ((\frac{D^{-1}A}{r_a})t){\vec{S}}_a}{n}\le E[Adoptions]\le \dots$

This formula captures the "echo" of influence as it ripples out through the specific pathways of the network over time $t$.

2. Optimal Seeding = Eigenvector Centrality

The authors prove that if you want to maximize the lower bound of your product's success, you should target nodes with the highest Eigenvector Centrality.

Intuition: You don't just want nodes with many friends (Degree Centrality); you want nodes whose friends are also influential and well-connected.

Note: The model depicts firms allocating budget between product cost (influencing the risk dominance $r$) and seed selection.

Competition and the Nash Equilibrium

When two firms fight for the same network, they reach a Nash Equilibrium. The researchers explored this in two specific topologies:

• Star Networks: Both firms fight for the central hub. If the hub is taken, they spread to the leaves. The competition here is intense and highly dependent on who captures the "king" of the network.
• k-Regular Networks: In a perfectly symmetric network (where everyone has $k$ neighbors), the location becomes irrelevant. The strategy simplifies back to the scalar problem of "quantity vs. quality."

The Quality-Quantity Tradeoff

A fascinating takeaway is the role of the production cost exponent ($\alpha$):

• If the product's value is highly sensitive to investment ($\alpha <0.5$), firms spend more on quality.
• If quality improvements are "cheap" or reach diminishing returns ($\alpha >0.5$), firms spend more on seeding as many people as possible.

Experimental Insights

The analytical bounds derived are proven to be tighter than the scalar bounds proposed in Fazeli & Jadbabaie (2012). By incorporating the position of initial seeds, firms can more accurately predict their market share.

Visualization Suggestion: A plot showing the gap between the scalar bound and the location-aware bound over time, demonstrating the increased accuracy of the new method.

Critical Insight & Future Outlook

While the paper assumes myopic best response (consumers only look at their immediate neighbors), real-world social influence is often multi-layered and non-myopic.

The Takeaway for Tech Leaders: Don't just spray and pray with free trials. In a competitive market where your product has network effects, your budget is better spent capturing a small cluster of highly "eigen-central" users than a large number of peripheral ones. The "location" of your first 100 users might dictate the fate of your next 1,000,000.