Diagnose rising cold-food complaints and choose metrics
Company: DoorDash
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: Medium
Interview Round: Technical Screen
## Case: Customers complain food arrives cold
You are a **Data Scientist (Analytics-focused)** at a food delivery marketplace (e.g., DoorDash). Over the last few weeks, customer support tickets and app feedback indicate a **material increase in “food arrived cold” complaints**. Leadership asks you to diagnose likely causes and propose data-driven fixes.
### Business context (assume typical delivery flow)
An order goes through:
1. **Order placed** (customer selects items, pays)
2. **Restaurant accepts & prepares**
3. **Dasher assigned & travels to pickup**
4. **Pickup happens** (food is handed to dasher)
5. **Travel to drop-off**
6. **Delivered**
Assume you have access to event-level order logs (timestamps for each step), restaurant and dasher attributes, geography, batching/stacking flags, and customer feedback (ratings, complaint reason codes, refunds/credits).
### Tasks
1. **Hypothesize potential drivers** of colder food (generate multiple plausible root causes across restaurant operations, dispatch/assignment, dasher behavior, batching/stacking, distance/ETA accuracy, and seasonality/weather).
2. For each hypothesis, specify:
- **What data/fields you would analyze** (and what joins or slices you’d do conceptually)
- **What comparisons** you’d run (time series, cohorting, matched comparisons, geo/restaurant segments, etc.)
- **Key confounders** to control for (e.g., order distance, cuisine type, weather, promos changing demand mix)
3. Propose **metrics** to measure:
- **Customer feedback / experience** (include multiple options and tradeoffs)
- **Revenue / business impact** (include multiple options and tradeoffs)
- **Guardrails** (to avoid improving “warmth” at the expense of other outcomes)
Clearly identify **one primary metric** and a small set of **diagnostic/guardrail metrics**.
4. Recommend **at least two interventions** (product/ops/policy) you would test, and outline an **experiment or quasi-experiment** plan:
- Unit of randomization (order/restaurant/dasher/geo)
- Success criteria and expected side effects
- How you would handle interference/network effects (marketplace supply/demand) and delayed outcomes (refunds, repeat rate)
Deliver your answer as a structured investigation plan plus an experimentation plan.
Quick Answer: This question evaluates a data scientist's diagnostic analytics skills, including hypothesis generation, causal inference, metric engineering, and A/B/quasi-experimental design in a food-delivery marketplace context.