Overwhelmed Merchants - Case Study: Detect and Mitigate Overloaded Restaurants

Question

## Round 1 - Data Science Case Study

## Problem

Some DoorDash restaurants are accepting more orders than they can fulfill on time, leading to late deliveries and cancellations. Design a system to detect overwhelmed merchants and intervene.

**Data Available:**
- `orders(order_id, restaurant_id, placed_at, accepted_at, ready_at, delivered_at, status)`
- `restaurants(restaurant_id, avg_prep_time_min, max_concurrent_orders)`

**Part A - Detection Metric:**
Define a real-time "overwhelm score" per restaurant. Consider:
- Current active order count vs. `max_concurrent_orders`
- Running average actual prep time vs. `avg_prep_time_min` in last 30 minutes
- Cancellation rate in last hour

**Part B - Intervention Decision:**
At what threshold do you temporarily pause new orders to a restaurant? How do you avoid false positives that hurt restaurant revenue?

**Part C - SQL:**
Find restaurants where actual prep time exceeded estimated prep time by more than 50% in the last 2 hours, with at least 5 orders.

```sql
SELECT restaurant_id,
       AVG(EXTRACT(EPOCH FROM (ready_at - placed_at))/60) AS avg_actual_prep,
       COUNT(*) AS order_count
FROM orders
WHERE placed_at >= NOW() - INTERVAL '2 hours'
  AND ready_at IS NOT NULL
GROUP BY restaurant_id
HAVING COUNT(*) >= 5;
-- Join with restaurants to compare against avg_prep_time_min
```

## Follow-ups

1. How do you distinguish a permanently slow restaurant from one having a one-off bad hour?
2. What notification do you send to the restaurant and what action can they take?
3. How would you A/B test whether pausing orders actually improves customer satisfaction?
4. What customer-facing message do you show when a restaurant is temporarily paused?

Overwhelmed Merchants - Case Study: Detect and Mitigate Overloaded Restaurants

Question Details

Full Details

Topics

More from DoorDash