In the fast-paced world of mobility platforms, the efficiency of driver-passenger matching is a critical factor that determines both user satisfaction and business profitability. DiDi, one of the industry giants, has developed an innovative solution called EXHOLD to optimize hold time in real time during the assignment process. This two-stage framework not only improves trip completion rates but also reduces cancellations and maximizes driver income. In this article, we delve into this technology, its technical implications, and how similar models can be applied in other business domains thanks to custom software development and advanced artificial intelligence services.
The core problem in large-scale ride-hailing systems is that every matching decision has immediate and future consequences. If a driver accepts a ride too early, they may miss a better opportunity; if they wait too long, the passenger might cancel. Traditional hold control methods use heuristic thresholds based on multiple predictive models, but these approaches are brittle under non-stationary traffic and difficult to optimize for multi-objective experience signals. EXHOLD proposes a clear separation: first, evaluate each driver-order pair by assigning it to an interpretable experience tier; second, calculate a hold time that respects predefined service guardrails.
In Stage I, a decision model is trained that groups pairs into discrete categories (e.g., 'excellent', 'good', 'fair') by optimizing a unified objective that aggregates satisfaction signals across the matching funnel. This allows the system to understand which combinations are most likely to yield a positive experience. In Stage II, a constrained optimization problem over empirical quantiles is solved to determine a monotonic hold time. In other words, it guarantees that promising matches are not unnecessarily delayed while maximizing overall experience improvement. This explicit design imposes guardrails that prevent the system from holding back pairs with high success probability, thus reducing passenger dissatisfaction risks.
Randomized A/B experiments in DiDi's production system in Brazil showed consistent gains: increased trip completions, higher driver income, and a significant reduction in passenger cancellations. Ablation analyses confirmed that both stages are essential and that the policy makes calibrated decisions under spatiotemporal heterogeneity. EXHOLD is currently deployed and serving real traffic in Brazil, proving its robustness.
Beyond ride-hailing, the conceptual architecture of EXHOLD is applicable to any system with an assignment queue where delayed decisions and multiple objectives exist. For example, in last-mile logistics, where a delivery driver may wait to batch several orders; in on-demand service platforms; or even in customer support systems where tickets are prioritized by expected value. To implement such solutions, companies need advanced mathematics, artificial intelligence models, and scalable infrastructure.
This is where Q2BSTUDIO adds value. As a software development company, we offer capabilities to build hold control systems from scratch, leveraging our experience in custom applications that integrate with cloud platforms like AWS or Azure. For instance, we can design a real-time decision model using machine learning and AI agents that constantly monitor system status. Additionally, cybersecurity is paramount: any driver and passenger data must be protected, and our pentesting and audit services ensure regulatory compliance. We also enhance decision-making with Business Intelligence dashboards (Power BI) that visualize key metrics like hold times, cancellation rates, and revenues.
Consider a courier company that wants to reduce delivery times and increase customer satisfaction. It could implement a hold control system that evaluates each order based on urgency, location, and driver capacity, similar to EXHOLD's Stage I. Then, with quantile-based optimization, it would determine how long to wait before assigning the order, ensuring priority shipments are not delayed. This requires predictive models, cloud infrastructure to handle demand spikes, and real-time monitoring. Q2BSTUDIO can assist throughout the cycle, from consulting to deployment.
Another example: a freelance marketplace. When a client requests a graphic designer, the platform could hold the request for a few seconds to find the best candidate rather than assigning the first available. A pair evaluation model (client-freelancer) based on project history, ratings, and availability, combined with an optimized hold policy, would improve acceptance rates and reduce cancellations. Again, the key is custom development integrating AI, cloud, and BI.
Technically, the cloud plays a critical role. Hold control systems require real-time data processing, large-scale storage of historical data, and the ability to scale under variable load. AWS and Azure offer services such as AWS Lambda, Azure Functions, NoSQL databases, and streaming solutions. Q2BSTUDIO has proven experience in cloud migrations and optimizations, ensuring controlled costs and high availability. Furthermore, implementing AI agents (e.g., based on reinforcement learning) can automate the calibration of hold thresholds without manual intervention, continuously adapting to market changes.
Cybersecurity is not an afterthought but a cornerstone. Any platform handling sensitive user data, such as real-time locations or payment information, must implement protective measures. Our cybersecurity services include vulnerability assessments, penetration testing, and compliance with regulations like GDPR or LGPD. This allows companies to deploy hold control systems with confidence that their data is secure.
Finally, data-driven decision-making is enhanced with BI. Using Power BI, we can build dashboards showing the performance of the hold control system: distribution of experience tiers, average hold times, impact on cancellations, and driver earnings. This enables managers to fine-tune parameters and quickly detect anomalies. The combination of all these technologies—custom software, AI, cloud, cybersecurity, and BI—is what makes it possible to bring concepts like EXHOLD into practice across any industry.
In conclusion, EXHOLD represents a significant advance in real-time hold management, demonstrating that a structured two-stage approach can overcome the limitations of heuristic methods. For companies looking to implement similar systems, collaborating with a technology partner like Q2BSTUDIO is key. Our expertise in custom software development, cloud integration with AWS/Azure, AI models, intelligent agents, cybersecurity, and Power BI visualization offers a comprehensive solution. Whether optimizing matching in a mobility platform or improving supply chain logistics, we are ready to turn data into efficient and secure decisions.
