COD management in quick commerce: handling cash at 10-minute speed without fraud leakage

COD management in quick commerce: handling cash at 10-minute speed without fraud leakage

Cash on delivery looks like a legacy payment method that quick commerce should have outgrown. In practice, COD remains 20-60% of orders across most emerging-market quick-commerce networks — and each COD order introduces a settlement, reconciliation, and fraud-risk layer that a digital-only model does not have. The operators running COD profitably at scale have built a tightly controlled cash ops layer that handles millions of transactions per day without leakage.

One of Asia’s largest quick-commerce arms across 200+ dark stores handles COD at scale through Shipsy’s cash-ops module. The mechanics matter for any quick-commerce operator in a market where digital payment penetration is below 80%.

Why quick-commerce COD is harder than parcel COD

Traditional CEP COD runs on a daily cycle: rider collects cash, deposits at end-of-shift, finance reconciles next day. The cycle works because parcel delivery has enough slack time to support it.

Quick-commerce COD compresses this into near-real-time. A rider completing 40-60 deliveries in a 6-8 hour shift cannot reliably track cash on paper. The hot-order model means riders return to the store every 15-30 minutes, creating many opportunities for cash drop but also many opportunities for reconciliation error.

Three failure modes dominate in poorly managed quick-commerce COD.

First, rider-level leakage. Cash miscounted, lost, or pocketed accumulates in a high-velocity environment. Without instrument-level tracking, the leakage is invisible until end-of-week finance reconciliation finds a gap.

Second, customer fraud. Customers placing COD orders and refusing acceptance on delivery — either never answering the door or claiming non-receipt after the rider has marked delivered. In quick commerce, fraud rates can reach 2-4% of COD orders without controls.

Third, settlement delay to merchant (in marketplace models). In multi-merchant quick-commerce models, COD collected by the platform must be settled to the merchant on a daily or sub-daily cycle. Manual settlement at this cadence is impossible.

What AI-native COD management looks like

The defining architectural choice is that every COD transaction is instrumented in real time — rider-level, customer-level, store-level, and merchant-level — rather than reconciled after the fact.

The mechanisms in detail

Risk scoring at order placement. Before a COD order is accepted, Shipsy runs a risk score against the customer: historical acceptance rate, address validity, order-value vs typical basket, time-of-order pattern. Orders flagged as high-risk either require digital payment, require an additional verification step, or are routed to higher-experienced riders with cash-handling training.

ePOD at delivery. On completion, the rider triggers an ePOD in the driver app. The sequence: geofence validation (rider is within 50 meters of the delivery address), customer OTP (customer provides the code sent to their phone at drop-off), photo capture for no-contact drops, and cash-received confirmation. The ePOD stack compresses disputed-delivery fraud to near-zero — customer claims of non-receipt do not survive the OTP and geofence record.

Per-return cash drop. Every time a rider returns to the store, the driver app prompts a cash count for orders completed in that trip. The store manager validates the count against the app’s expected total. Discrepancies surface within 15-30 minutes, not at end-of-day.

Nexa-driven settlement. At configurable frequencies — typically every 6-12 hours in quick-commerce deployments — Nexa reconciles the store’s aggregated cash against orders delivered, computes merchant payouts (in marketplace models), and books the ledger entries. The mechanism is identical to how Nexa handles freight settlement at scale — a global alco-bev leader operating across 70+ countries used the same underlying agent to autonomously settle $25M+ in disputes, see a detailed case study.

Fraud control — where Clara plays

A large share of COD “fraud” is actually customer confusion, not malicious behavior. Customer forgets they placed the order. Customer’s family member refuses on their behalf. Customer finds the delivery mis-placed and assumes it never arrived.

Clara, the Shipsy CX agent, handles this category at scale. When a rider marks a COD delivery as refused or non-contactable, Clara immediately contacts the customer with the delivery record (geofence, photo, OTP-attempt log) and resolves the dispute in the same conversation. True fraud — systematic refusal across multiple orders — is flagged to the risk-scoring layer and impacts the customer’s future COD eligibility.

This shifts the economic model of COD fraud handling. Traditional networks spend 1-3% of COD order value on fraud investigation and write-off. AI-native networks compress this to 0.2-0.5% through Clara’s real-time resolution.

What this means for quick-commerce operators scaling COD

Three structural shifts separate operators running COD profitably from those bleeding through it.

First, instrument every COD transaction at four levels. Rider, customer, store, merchant — each needs real-time tracking with reconciliation lag measured in minutes, not days.

Second, pre-empt fraud at order placement, not after delivery. Risk scoring at the point of order acceptance is cheaper than investigation after the fact.

Third, automate settlement at sub-daily frequency. Merchant trust in a quick-commerce marketplace depends on settlement speed. Operators settling daily lose marketplace share to those settling hourly.

For context on the broader settlement architecture, see the Nexa settlement automation playbook. For vertical context, visit the quick-commerce industry page or explore Shipsy’s last-mile product.