How to Optimize O2O Delivery Fulfillment for Maximum Efficiency?

The O2O home‑delivery market has expanded dramatically, especially after the pandemic, turning online purchase into a rigid demand and creating huge commercial opportunities.

Suning Retail Technology Research Institute conducted extensive surveys and deep interviews with senior personnel from leading O2O platforms, covering business models, order flow, product supply chain, store fulfillment, and delivery fulfillment.

This article focuses on the delivery fulfillment stage, a core component of the O2O business where poor execution often leads to unprofitability and high delivery costs.

Large chain supermarkets typically outsource delivery to third parties because building an in‑house team is costly. However, third‑party outsourcing brings issues such as lack of price control, inability to monitor rider performance, and missed opportunities for rider‑driven customer engagement.

Some supermarkets have successfully built their own rider teams, especially single‑store operations, achieving better control and cost savings.

A hybrid model can combine self‑owned riders for peak periods (e.g., 11:00‑13:00, 17:00‑19:00) with third‑party riders for off‑peak times, creating a "self‑owned + outsourced" delivery system.

In‑store rider models keep riders waiting at the store to dispatch immediately once orders are packed, improving efficiency, enhancing brand visibility, and allowing riders to interact with customers, form groups, and drive repeat purchases.

Self‑built teams can also reduce material costs, such as reusing insulated bags, which third‑party services cannot achieve.

When order volume is low, riders become idle; therefore, batch‑ordering (集单) is essential. By aggregating orders—e.g., a store handling 400 orders daily, surging to 2000 during promotions—riders can deliver multiple orders per trip, drastically lowering per‑order delivery cost.

Batching can be organized in time windows (e.g., 30‑minute grids) to ensure routes are sequential and efficient.

Smart batching algorithms drive rider planning by decomposing order fulfillment into POI‑level batching, process breakdown, and plan calculation, then using routing algorithms to generate batch tasks, packaging, navigation, and delivery execution, with machine‑learning models continuously optimizing routes.

The algorithm considers freshness, temperature, path, duration, cost, batch size, weight, and number of trips.

Leading vendors have built comprehensive instant‑delivery systems divided into seven modules, including LBS (location‑based services) and machine‑learning components for ETA (estimated time of arrival) prediction.

LBS uses geographic data, while machine‑learning models evaluate time series, predict delivery times, and apply deep‑learning techniques; the data platform integrates offline big‑data, real‑time feature computation, and ML pipelines.

Multi‑sensor fusion enables real‑time insight into rider behavior, such as detecting prolonged stays in a community and inferring constraints (e.g., riders forced to walk), which informs scheduling, delivery‑range optimization, and pricing decisions.

The pricing system comprises rider subsidies, order‑structure optimization, and rider operations, targeting delivery‑scale growth, profit‑loss balance, and experience efficiency.

Order‑structure optimization includes range expansion, surge‑assistant tools, dynamic pricing, and base pricing, built on GIS, AOI/BLOCK data mining, traditional ML & deep‑learning, and pricing mechanisms.

Multi‑sensor fusion reconstructs the real delivery process, while rider real‑time data supports WiFi geofencing, motion recognition, and gait analysis, enabling precise handling of scenarios like restricted vehicle zones.

Contributors: researchers Luo Haitang, Chi Shuqiang, and Guo Yonghui from Suning Retail Technology Research Institute.