Diagnose Why Delivered Food Arrives Cold

Scenario

Over the past six weeks, customer support reports a 28% increase in order complaints tagged as "food arrived cold / lukewarm" in a major metropolitan market. Merchant operations and logistics teams want to understand whether this is driven by reporting changes, dispatch and routing inefficiencies, restaurant prep delays, or external environmental factors.

As the product data scientist on the Fulfillment Experience team, you are tasked with leading the diagnostic investigation and proposing a data-backed intervention.

Part 1: Metric Definition & Measurement Reliability

1. Define a primary metric for "cold food incidence" using customer support tickets, refund requests, and post-delivery ratings. What are the known biases of relying solely on support tickets, and how would you construct an adjusted or proxy metric to monitor real customer dissatisfaction?
2. What instrumentation or telemetry across the order lifecycle would you inspect to validate whether actual physical thermal degradation is occurring versus an increase in customer propensity to complain?

Part 2: Funnel Decomposition & Root Cause Analysis

Assume you have detailed timestamp events for each order: order_placed, order_confirmed_by_merchant, dasher_assigned, dasher_arrived_at_store, order_picked_up, and order_delivered.

1. Decompose the end-to-end delivery timeline into distinct operational segments. Which specific segment duration correlates most strongly with food cooling, and why?
2. During the period of interest, the overall cold food complaint rate rose from 1.2% to 1.54% of delivered orders. Explain how you would apply a mix-shift decomposition (e.g., Kitagawa or Oaxaca-Blinder style rate decomposition) to determine how much of this +0.34 percentage point change is attributable to:
   • Changes in operational performance within segments (e.g., longer wait times or transit times).
   • Mix shifts across covariates (e.g., higher share of batched orders, longer delivery radiuses, seasonal ambient temperature drop, or shifting cuisine mix toward items with high thermal loss like French fries).

Part 3: Experimentation & Guardrails

The logistics team proposes an algorithmic tweak to reduce cold food: tightening the maximum allowable travel distance and capping batching (stacked deliveries) for high-risk cuisines.

1. Formulate the hypothesis and define the unit of randomization (e.g., market-day switchback vs. customer-level vs. Dasher-level) for evaluating this intervention. Justify your choice against marketplace interference (SUTVA violations).
2. Specify your primary success metric, key secondary fulfillment metrics, and marketplace guardrails across customers, Dashers, and merchants.