Balancing Profit and Traveller Acceptance in Ride-Pooling Personalised Fares

Michal Bujak, Rafal KucharskiEur. J. Oper. Res 2024 · DOI · OpenReview

TL;DR

This paper introduces a novel personalized pricing strategy for ride-pooling services that balances operator profit with traveler acceptance. By accounting for individual traveler preferences and ride characteristics, the system can significantly improve efficiency, reduce mileage, and increase profitability compared to traditional flat-discount models.

What Problem Does This Paper Address?

Ride-pooling services offer a promising solution for reducing urban congestion and environmental impact by increasing vehicle occupancy. However, a key challenge is to make these services attractive to both travelers and profit-driven operators. Travelers often face discomfort from longer routes and detours, which is typically compensated with a sharing discount. Existing approaches often use a one-size-fits-all flat discount or a discount proportional to inconvenience. This paper addresses the limitations of these approaches by proposing a personalized pricing strategy that considers the heterogeneity of traveler preferences (value of time, willingness to share) and optimizes fares at an individual level to maximize both operator profit and system-wide efficiency.

Methods

The authors propose a probabilistic framework for personalized pricing in ride-pooling. The core of their method involves:

  • Shareability Graph Construction: Building a graph representing all feasible ride combinations based on trip requests and estimated behavioral traits of the population.
  • Acceptance Probability Modeling: Developing functions that predict a traveler's likelihood of accepting a shared ride based on personalized discounts, their value of time, and penalty for sharing. This accounts for the fact that a shared ride is only realized if all co-travelers accept.
  • Discount Space Discretization: To make the complex optimization problem computationally feasible, the continuous space of potential discounts is discretized.
  • Expected Profitability Maximization: At the ride level, they calculate the expected profitability by considering all possible acceptance/rejection scenarios and their probabilities. This metric is then optimized by finding the best personalized discount vector for each ride.
  • System-Level Optimization (Matching): Finally, an integer linear programming approach is used to select a subset of rides (an "offer") that maximizes the overall system's average expected profitability.

A key novelty is the probabilistic approach that accounts for individual traveler decisions and the fact that a single rejection can derail a shared ride.

Datasets

The study uses publicly available data from the NYC Taxi and Limousine Commission for trip requests in Manhattan. Behavioral preferences of travelers are based on existing literature, specifically drawing from studies that categorize users into different classes (LC1-LC4) with varying value of time and penalty for sharing.

Key Insights and Ideas

  • Heterogeneity is Key: Treating travelers as a homogeneous group with a single discount is suboptimal. Personalized pricing that accounts for individual preferences unlocks significant gains.
  • Balancing Act: The personalized pricing strategy effectively balances the operator's need for profit with the traveler's desire for an attractive and convenient service. It identifies rides that are efficient for the system and incentivizes them, while discouraging less efficient ones.
  • "Shareability Topology" Matters: The method reveals interesting properties of how rides can be combined. It identifies rides that are highly compatible and can be offered with lower discounts, and conversely, identifies travelers or ride combinations that are less compatible and are steered towards private rides with higher fares.
  • Local Optimality Implies Global Optimality: The authors prove that optimizing pricing at the individual ride level leads to an optimal system-wide solution, simplifying the overall optimization process.

Key Results

  • Profitability Boost: The personalized pricing strategy leads to a significant improvement in the operator's profit per mile, with an over 20% increase compared to flat discount baselines.
  • Mileage Reduction: The system achieves a reduction in total mileage by 4.5%, indicating improved environmental efficiency and reduced congestion.
  • Offer Structure: The personalized approach favors more complex shared rides (higher degrees in the shareability graph) compared to flat discount strategies, suggesting better utilization of vehicle capacity.
  • Acceptance Probability: While most potential shared rides in the shareability graph have low attractiveness, the personalized strategy effectively selects rides that are either highly efficient or attractive to pooling enthusiasts, leading to better overall realization rates.

Limitations and Future Work

  • Data Requirements: The model relies on knowing the distribution of behavioral traits within the population. While this can be estimated from studies or historical data, obtaining precise individual preferences remains a challenge.
  • Computational Complexity: While discretization makes the problem tractable, the computational cost can still be significant for very large-scale systems.
  • Future Work: The authors suggest incorporating dynamic learning of traveler preferences over time, allowing the operator to adapt pricing strategies based on observed behavior. Further research could also explore incorporating other factors like vehicle readiness costs and driver commissions into the objective function.

Summary updated

Abstract

In a ride-pooling system, travellers experience discomfort associated with a detour and a longer travel time, which is compensated with a sharing discount. Most studies assume travellers receive either a flat discount or, in rare cases, a proportional to the inconvenience. We show the system benefits from individually tailored fares. We argue that fares that optimise an expected profit of an operator also improve system-wide performance if they include travellers' acceptance. Our pricing method is set in a heterogeneous population, where travellers have varying levels of value-of-time and willingness-to-share, unknown to the operator. A high fare discourages clients from the service, while a low fare reduces the profit margin. Notably, a shared ride is only realised if accepted by all co-travellers (decision is driven by the latent behavioural factors). Our method reveals intriguing properties of the shareability topology. Not only identifies rides efficient for the system and supports them with reduced fares (to increase their realisation probability), but also identifies travellers unattractive for the system (e.g. due to incompatibility with other travellers) and effectively shifts them to private rides via high fares. Unlike in previous methods, such approach naturally balances the travellers satisfaction and the profit maximisation. With an experiment set in NYC, we show that this leads to significant improvements over the flat discount baseline: the mileage (proxy for environmental externalities) is reduced by 4.5% and the operator generates more profit per mile (over 20% improvement). We argue that ride pooling systems with fares that maximise profitability are more sustainable and efficient if they include travellers' satisfaction. Keywords: ride-pooling, personalised pricing, individual discounts
BibTeX
@article{bujak2024balancing,
  author = {Michal Bujak and Rafal Kucharski},
  title = {{Balancing Profit and Traveller Acceptance in Ride-Pooling Personalised Fares}},
  journal = {Eur. J. Oper. Res.},
  year = {2024},
  doi = {10.1016/j.ejor.2025.12.025},
}
Figure