Key Concepts
- 1Generous return and agent policies attract abuse — wardrobing (wear-and-return), serial returners, empty-box returns, and receipt/label fraud all erode margin
- 2Agent-specific abuse is new: prompt injection to extract discounts, socially engineering refunds, or automating many sessions to farm promo codes
- 3Detection relies on patterns over time: return-rate per customer, value-at-risk, and cross-session behavior — a single return is fine, a habit is a signal
- 4Response should be graduated, not binary: friction (require receipt, restrict methods) before outright bans, to avoid punishing occasional legitimate edge cases
- 5The agent must resist manipulation: refund and override authority should be bounded by policy limits it cannot be talked past, with human escalation above thresholds
- 6Fairness matters: abuse controls that flag too aggressively harm honest customers and their CLV, so the false-positive cost must be weighed like in fraud
- 7Feedback loops: confirmed abuse outcomes tune the thresholds, and the agent's logs provide the evidence trail to justify actions
Why It Matters for PMs
Margin leakage: Return and refund abuse is a quiet, compounding cost. Encoding limits and detection into the agent protects profitability without a heavy-handed policy.
New manipulation surface: Conversational agents can be socially engineered for refunds and discounts. Bounded authority and injection resistance are genuine security requirements.
Fairness/CLV balance: Over-aggressive abuse controls alienate good customers. PMs must weigh false-positive cost, just as with fraud, to protect long-term value.
Test Your Understanding
Question 1 (Multiple Choice)
What distinguishes abuse from a legitimate return in agent detection?
Question 2 (Multiple Choice)
Why should abuse responses be graduated rather than binary bans?
Question 3 (Multiple Choice)
Which is an agent-specific abuse pattern?
Question 4 (Short Answer)
How would you bound an agent's refund/override authority to resist social engineering?
Think about your answer before revealing the explanation.
Question 5 (Short Answer)
How would you balance abuse detection against the false-positive cost to honest customers?
Think about your answer before revealing the explanation.