Reduce airport cancellations under causal constraints
Company: PayPal
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: easy
Interview Round: Onsite
You are a Data Scientist on an **airport rides** team for a ride-hailing marketplace.
Airport rides differ from city rides:
- Drivers often enter an **airport queue** (FIFO/priority rules), so cancellations can impose large costs on drivers (they may lose their queue position, have to leave the holding lot, or re-enter).
- Riders at airports are a special segment (navigation to pickup zones, unfamiliarity, one-time users, business vs personal trips).
- There may be **network effects / interference** (one driver’s assignment affects others in the queue).
- Standard experimentation is hard:
- **Geo tests** are difficult (only a few airports; spillover).
- **Switchback** may not be feasible or could induce operational risk.
- **Diff-in-diff** can be biased by time-varying confounding.
- **Synthetic control** may be hard to operationalize.
## Problem
Cancellations are high for airport pickups, hurting both marketplace efficiency and user experience.
## Questions
1) Propose a measurement framework for **driver experience** and **rider experience** at airports.
- Define **primary metric(s)**, **diagnostic metrics**, and **guardrail metrics**.
- Discuss how you would handle segmentation (e.g., business vs personal, first-time vs repeat, pickup zone complexity).
2) Generate plausible hypotheses for why cancellations happen at airports.
- Include both **data-driven hypotheses** and at least one **behavioral/psychology** hypothesis.
3) Propose an evaluation / causal strategy to test interventions to reduce cancellations, given:
- interference/network effects,
- queue dynamics,
- time-varying demand/supply and seasonality,
- limited number of airports.
4) Outline what data you would need, the main confounders/biases you’d worry about, and how you’d communicate tradeoffs to stakeholders.
Quick Answer: This question evaluates competency in causal inference, experimentation design, metric definition, and marketplace analytics, testing a candidate's ability to reason about interference, queue dynamics, behavioral drivers, and measurement under operational constraints in the Analytics & Experimentation domain and is commonly asked to assess judgment in designing robust measurement frameworks when standard randomized experiments are limited. It probes both conceptual understanding of causal and behavioral drivers and practical application in experimental strategy, data needs, bias identification, and trade-off communication rather than low-level implementation details.