Abstract
We study the optimal usage-based pricing problem in a resource-constrained network with one profit-maximizing service provider and multiple groups of surplus-maximizing users. With the assumption that the service provider knows the utility function of each user (thus complete information), we find that the complete price differentiation scheme can achieve a large revenue gain (e.g., 50%) compared to no price differentiation, when the total network resource is comparably limited and the high willingness to pay users are minorities. However, the complete price differentiation scheme may lead to a high implementational complexity. To trade off the revenue against the implementational complexity, we further study the partial price differentiation scheme, and design a polynomial-time algorithm that can compute the optimal partial differentiation prices. We also consider the incomplete information case where the service provider does not know which group each user belongs to. We show that it is still possible to realize price differentiation under this scenario, and provide the sufficient and necessary condition under which an incentive compatible differentiation scheme can achieve the same revenue as under complete information.
Introduction
In the early 2010s, major telecom providers shifted from flat-fee "unlimited" data plans to usage-based pricing. While economic theory suggests that highly complex, non-linear pricing is the best way to extract revenue, real-world implementation is hindered by "implementational complexity"—customers prefer simple billing, and providers often lack perfect information about user valuations. This research addresses two primary challenges: the complexity-performance tradeoff in designing simple yet profitable pricing tiers, and the information gap when user group identities are unknown to the provider.
System Model and Methodology
The interaction between a service provider (SP) and users is modeled as a two-stage Stackelberg game. In the first stage, the SP sets prices based on available information (complete or statistical). In the second stage, admitted users decide their resource consumption ($s_i$) to maximize their individual surplus (utility minus cost).
Key methodological contributions include:
- Water-filling Algorithms: Adapting classic resource allocation logic to identify "effective" user groups that receive resources versus those excluded.
- Three-Level Decomposition: Solving the Partial Price Differentiation (PP) problem by breaking it into cluster partitioning, inter-cluster resource allocation, and intra-cluster pricing.
- Incentive Compatibility: Designing quantity-based price menus for incomplete information scenarios to ensure users self-select into appropriate groups.
Key Insights: The Threshold Structure
A central finding is the "Threshold Structure" of the optimal solution. The provider should prioritize resources for high-value users and potentially exclude low-value users when network resources are constrained. This ensures that limited resources are allocated to those with the highest willingness to pay.
Other key insights include:
- The Minority Advantage: Price differentiation yields the highest benefits when high-willingness-to-pay users are a small minority. If everyone has high valuations, a single high price is sufficient.
- Consecutive Partitioning: Users with similar valuations should be grouped into the same tier, reducing the search space for optimal tiers from exponential to polynomial complexity.
- Zero Information Rent: Under specific conditions, an SP can achieve the same revenue with a smart quantity-based menu as they would with perfect knowledge of user identities.
Results and Discussion
Numerical simulations validate the theoretical models:
- Revenue Gain: Complete price differentiation can achieve a 50% revenue increase over single-pricing when resources are limited and high-value users are rare.
- Algorithmic Efficiency: The proposed algorithm for partial price differentiation operates in polynomial-time ($O(I^J)$), making it feasible for large-scale networks with few price tiers.
- Diminishing Returns: Moving from one price to just two tiers often captures the majority of potential revenue gains, suggesting a "sweet spot" for pricing complexity.
Conclusion and Future Work
While this study provides a robust framework for usage-based pricing, it assumes a monopoly environment. Future research should consider competitive markets where "price wars" might occur. Additionally, moving beyond linear pricing to more complex non-linear structures and accounting for dynamic network conditions remain open areas for investigation.
Summary updated