Problem
Paying for deliveries using cash after the delivery is made is a popular mode of payment employed by customers transacting online for the first time or those that prefer to have more control, especially in emerging economies like India. While the cash (or pay)-on-delivery (COD or POD) option helps e-commerce platforms, for example in our food delivery platform, tap into new customers, it also opens up substantial risk in the form of fraud and abuse. A common risk mitigation strategy is to impose a limit on the order value (MPL - maximum purchase limit) that can be paid using COD. MPL is typically blunt (a single limit for a city or zip code) and set by business teams using heuristics and primarily from a risk-management-backward view.

Implication
Blunt MPLs are a one-size-fits-all approach which means we leave money on the table for customer groups where the limits are too strict and lose money on groups where they are lax. We need to balance the risk management and the customer preference angles simultaneously and dynamically.

Solution
We try to frame this as a constraint optimisation problem and then try to find solutions to this using analytical models as well as an uplift modeling-based approach.

Outline
In this talk, I wish to present the following:

1. A brief understanding of the COD payment method and MPL and their implications in Indian e-commerce.
2. An understanding of the system that determines the COD eligibility of our customers
3. Mathematical Formulation of the MPL determination problem
4. An overview of the analytical and ML model-based solutions we built at Swiggy
5. A view of the real-time inference framework
6. Some experiment results and future scope