How Dynamic Pricing and Smart Surcharges Boost Freight Platform Efficiency During Peak Seasons

Preface

Freight demand has been rising, with sharp spikes during holidays such as National Day and Double 11, creating peak periods that strain processing efficiency and stability, prompting the need for dynamic strategies.

Background

Four supply‑demand scenarios are identified (driver supply vs. user demand). During peaks, supply is far lower than demand, leading to price hikes, long queues, degraded service quality, driver fatigue, dispute escalation, cargo backlog, and brand damage.

Industry Solutions Overview

Key approaches include:

Surge Pricing (Uber): multiplies fare by a factor when demand exceeds supply.

Dynamic Pricing (Lyft): raises price to match supply‑demand curves without unlimited price spikes.

Allocation & Commission Differentiation: offers drivers different order‑allocation probabilities or commission rates.

Platform

Strategy

Advantages

Disadvantages

Huolala Capability

Uber

Surge Pricing

1. Reduces wait time

2. Lowers reject rate in high‑demand areas

3. Increases supply in hot zones

4. Cuts WGC phenomenon

1. Increases passenger dissatisfaction (price discrimination)

2. Drivers cherry‑pick orders

Yes

Lyft

Dynamic Pricing

Limits unlimited price hikes compared to surge pricing

Same as surge pricing

Yes

Didi China

Allocation Differentiation

Maximizes platform profit

Revenue transfer issues

No

Didi Overseas

Commission Differentiation

Yes

Solution

Huolala adopts a suite of dynamic‑price adjustment algorithms that adjust fares based on real‑time supply‑demand data through five steps:

Data Collection: Gather historical order demand, driver count, response rates, etc.

Supply‑Demand Analysis: Analyze current balance and forecast future shortages.

Price Adjustment: Automatically raise or lower fares when supply is insufficient.

Real‑Time Monitoring: Continuously track market changes to trigger price updates.

Balance User Experience & Profit: Increase fares during peaks to attract drivers while avoiding excessive user churn.

Implementing this framework improves service quality, reduces dispute handling difficulty, and maximizes profit.

Dynamic Pricing Lifecycle

Intelligent Surcharge

Smart surcharges address short‑term supply‑demand imbalances (e.g., peaks, holidays, bad weather) by replacing manual price adjustments with data‑driven algorithms.

Background and Analysis

Surcharges act as a powerful tool to balance supply and demand. Traditional manual adjustments are slow and imprecise; algorithmic approaches use data‑driven demand‑supply predictions to set surcharge levels.

Strategy

Three strategies are designed based on data availability:

Supply‑Demand Strategy: Simulates manual adjustments with minimal elasticity reference.

Elasticity Strategy: Partially simulates manual adjustments and references elasticity data.

Optimization Strategy: Fully relies on algorithmic solutions within business constraints.

Elasticity Strategy

Uses causal inference to estimate the effect of surcharge (treatment) on demand, fitting a linear regression to relate surcharge coefficient to order‑conversion rate. The elasticity formula is:

Elasticity = (ΔConversionRate / (ΔSurchargeCoefficient * BaseConversionRate))

Elasticity Type

Surcharge Elasticity

Definition

Unit: surcharge coefficient; metric: relative change in conversion rate

Formula

Pre‑surcharge conversion rate: eo, surcharge coefficient: Δp, post‑surcharge conversion rate: (image)

Characteristic

Smaller elasticity → higher price sensitivity → poorer effect

Example

0% increase → 0.5 conversion rate; 10% increase → 0.4 conversion rate → Elasticity = -2

Higher elasticity (smaller drop) is preferred for surcharge deployment.

Optimization Strategy

Combines causal S‑learner (LightGBM) to predict user order probability and driver acceptance probability, calibrates with isotonic regression, and formulates an integer programming problem per city and vehicle type to maximize paired orders under constraints (average surcharge range, lower response → higher surcharge, non‑negative profit impact). If no optimal solution exists, fallback to elasticity‑derived surcharge.

Integer Programming

Integer programming solves discrete decision variables; mixed‑integer variants handle both integer and continuous variables. The workflow is illustrated below:

Challenges

Challenge 1 – Supply‑Demand Calculation

Spatial heterogeneity, rapid demand fluctuations, real‑time requirements, and data quality issues make accurate supply‑demand computation difficult.

Flexible Grid

Adopts Uber’s H3 hexagonal indexing to create flexible grids that aggregate neighboring cells to balance order density across regions.

Dynamic Adaptive Supply‑Demand

Combines flexible grids with 30‑minute time slices to form sub‑scenes, applies confidence‑interval‑based thresholds to retain reliable metrics, and aggregates sparse sub‑scenes to improve data coverage.

Challenge 2 – Supply‑Demand Forecasting

Long‑term and real‑time forecasts are built using ARIMA, Holt‑Winters, Prophet, Temporal Fusion Transformers (TFT), XGBoost, and LightGBM. Models predict future demand, response rates, and guide surcharge decisions.

Key formulas (ARIMA, Holt‑Winters) are shown as images.

Challenge 3 – High Iteration Frequency

Rapid business environment changes require frequent algorithm updates; the architecture leverages flexible grids, adaptive algorithms, and forecasting models to maintain responsiveness.

Challenge 4 – Operational Constraints

Operations set algorithmic constraints (e.g., surcharge bounds). Algorithms provide fine‑grained strategies within these limits, and simulation tools help evaluate impact on supply‑demand, coverage, and revenue.

Model Evaluation

Models are assessed using MAPE (to prioritize low‑response scenarios) and strategy‑interval hit rate, which measures whether predicted response falls within the correct surcharge band.

In National Day peak experiments, the model‑driven group achieved a 4.5 pp higher correct price‑adjustment rate than the control group.

Conclusion

Dynamic pricing, intelligent surcharges, flexible grids, and advanced forecasting have been applied across scenarios such as pandemics, adverse weather, and holidays, delivering ~10 % revenue gains during peak periods. Future work includes exploring commission management, driver‑price recommendations, and integrating deeper learning models.

References

[1] Wikipedia. Integer programming. https://zh.wikipedia.org/wiki/整数规划

[2] datarootlabs (2021). https://datarootlabs.com/blog/uber-lift-gett-surge-pricing-algorithms

[3] ridester (2023). https://www.ridester.com/surge-pricing/

[4] Harvard Business Review (2015). https://hbr.org/2015/12/everyone-hates-ubers-surge-pricing-heres-how-to-fix-it

[5] Uber H3 blog (2018). https://www.uber.com/en-HK/blog/h3/

[6] Wang, P., Zhang, H., & Zhang, Z. (2023). A ridesharing platform with an oversupply of drivers. SSRN.

[7] Garg, N., & Nazerzadeh, H. (2022). Driver surge pricing. Management Science, 68(5), 3219‑3235.

[8] Hall, J., Kendrick, C., & Nosko, C. (2015). The effects of Uber’s surge pricing. University of Chicago Booth School of Business.

[9] Chang, Y., Winston, C., & Yan, J. (2022). Does Uber benefit travelers by price discrimination? Journal of Law and Economics, 65(S2), S433‑S459.

[10] Künzel, S. R., Sekhon, J. S., Bickel, P. J., & Yu, B. (2019). Meta‑learners for estimating heterogeneous treatment effects. PNAS, 116(10), 4156‑4165.

[11] Wager, S., & Athey, S. Estimation and inference of heterogeneous treatment effects using random forests. JASA.