Size the yearly cost of abusive orders at Amazon and design countermeasures

You are a product manager at Amazon. Some people exploit the ordering process and the policies around it to get goods or money they are not entitled to, and this costs Amazon money. The prompt calls this "purchase order abuse" and does not define it further, so deciding which behaviors it covers is part of the task. Answer in two parts. ### Clarifying Questions - Which behaviors count: false claims that a package never arrived or arrived damaged, returning a used, different or empty item, extra accounts created to reuse new-customer offers, bulk buying of limited items for resale, or business orders placed on invoice terms and never paid? - Should the estimate cover the US store or all countries, and should it be expressed in orders, dollars or both? - Does the cost include only the value refunded or replaced, or also shipping, handling and customer-service time? - How much added friction for honest customers is acceptable when reducing abuse? ### Part 1 — Size the problem Estimate how much purchase order abuse happens at Amazon in a year, in number of orders and in dollars. Label your assumptions, and say how you would firm up the estimate with Amazon's own data. ```hint Follow the money back out Abuse can only happen where a policy gives back money or goods. Count the orders that pass through those policies, then estimate what share of them are illegitimate and what each one costs. ``` #### What This Part Should Cover - A working definition that names which abuse types are counted - An equation with round, labeled assumptions and correct arithmetic, followed by a sanity check - The inputs the result is most sensitive to - A way to measure true prevalence, given that detected cases understate it ### Part 2 — Reduce the abuse Propose and prioritize ways to cut this abuse, and explain how you would measure success. ```hint Different abusers, different controls A one-time opportunist, a repeat abuser and an organized group respond to different measures, and the honest majority still expects refunds to be quick and easy. ``` #### What This Part Should Cover - Segments of abusers and the policies each one exploits - Controls matched to risk, with friction aimed at high-risk claims rather than at everyone - Prioritization by loss prevented against cost and customer impact - Success metrics with guardrails that protect honest customers ### What a Strong Answer Covers - A defensible scope for a vague term, set before any numbers - An estimate whose structure maps onto the controls proposed in Part 2 - An explicit balance between reducing losses and keeping returns and refunds easy for honest customers - A plan for abusers adapting once controls go live ### Follow-up Questions - A new rule cuts refund abuse by a third, but complaints from honest customers about slower refunds double. How do you decide whether to keep it? - How would you counter groups that sell "refund services," coaching buyers on how to talk customer support into a refund? - For orders from third-party sellers, who should absorb the cost of an abusive refund, and why?