Research Question

Can a merchant reduce wasted COD shipping cost by scoring risky orders before dispatch?

This research direction is derived from the Wasilio COD operations platform. The Systems case study explains the product workflow; this research note will focus on the analytical layer behind it.

Why This Matters

In COD commerce, the merchant pays operational cost before the customer pays. A bad order can consume confirmation time, courier capacity, delivery fees, and inventory availability.

The goal is not to block every uncertain order. The goal is to help the merchant choose the right next action:

• ship normally;
• call again;
• require prepayment;
• hold for manual review;
• reject clear abuse.

Candidate Signals

The first research version should stay explainable. Useful signals include:

| Signal | Reason | | --- | --- | | Failed confirmation attempts | Customer intent is weak or unreachable | | Callback history | Order may need delay instead of dispatch | | Repeated refusal by phone or address | Possible fake intent or repeat abuse | | Address mismatch or incomplete location | Higher delivery failure probability | | Courier failure reason | Separates buyer risk from logistics risk | | Merchant-specific failure rate | Risk should adapt to each merchant context |

Planned Model Shape

The first model should be deterministic and inspectable:

order features
  -> weighted risk signals
  -> normalized score
  -> risk level
  -> recommended action
  -> human-readable explanation

Later versions can compare this baseline against probabilistic or machine-learning approaches, but only after clean outcome data exists.

What This Will Demonstrate

This research will show the bridge between product workflows and applied intelligence:

• turning failed delivery outcomes into features;
• designing explainable risk scores;
• choosing thresholds based on business cost;
• avoiding black-box fraud decisions;
• connecting Wasilio operations data to merchant decision support.