Derive insights and improve complaint resolutions

You work on customer experience analytics for an e-commerce furniture retailer. You have 3 data tables related to customer complaints and how they were resolved.

Table 1: complaints

Each row is a complaint/contact.

• complaint_id (string, PK)
• customer_id (string)
• order_id (string)
• product_category (string)
• issue_type (string; e.g., damaged item, late delivery, missing parts)
• channel (string; phone/chat/email)
• created_at (timestamp)
• region (string)

Table 2: resolutions

One row per complaint resolution attempt (may be multiple per complaint).

• resolution_id (string, PK)
• complaint_id (string, FK -> complaints)
• resolution_type (string; e.g., refund, replacement, coupon, escalate-to-carrier)
• resolved_at (timestamp)
• sla_met (boolean)
• refund_amount (numeric)

Table 3: service_ratings

Customer satisfaction after the resolution.

• complaint_id (string, FK -> complaints)
• rating (int 1–10; 10 = most satisfied)
• rated_at (timestamp)

Questions

1. Provide 3 insights you would try to extract from these data. (Be specific about cuts/segments and what “insight” means.)
2. Provide 3 actionable recommendations based on those insights.
3. Choose one prioritized recommendation and explain why it should be prioritized (impact vs effort, customer value, cost, risk).
4. Describe how you would measure success of that recommendation:
   • define a primary metric and supporting diagnostic/guardrail metrics,
   • propose an evaluation method (e.g., A/B test, phased rollout, difference-in-differences),
   • discuss key confounders/biases (seasonality, selection into channels, case-mix differences, delayed ratings).
5. What risks are associated with implementing it, and how would you mitigate them?

You may assume standard data quality issues (missing ratings, multiple resolutions per complaint, backfilled timestamps) and should call out how you’d handle them.