Diagnose and optimize shared workspace marketplace conversion

You are responsible for a two-sided marketplace of shared workspaces. Traffic runs at about 200,000 sessions each day. Booking conversion declined from 3.2% (2025-08-01 to 2025-08-14) to 2.4% (2025-08-15 to 2025-09-01) after a ranking change was released on 2025-08-15; at the same time, downtown partner supply decreased about 10% due to local events, and the paid traffic mix moved from search to social.

a) Construct a metric tree that starts with overall booking conversion and breaks it into component rates (search → view → contact → booking). Which slice-and-dice checks and counter-metrics would you run first to distinguish cause from correlation, including supply elasticity, position bias, latency, partner acceptance, and cancellations?

b) Propose a log-based analysis that attributes the impact to the ranking change rather than supply or traffic-mix confounders, using approaches such as diff-in-diff across unaffected suburbs, CUPED, or synthetic controls.

c) Design an A/B test for a proposed ranking fix: specify the primary metric, guardrails (cancellations, partner rejection rate, search latency p95), and pre-registration of the plan.

d) Power: given a baseline CVR of 2.4%, a target relative lift of 6%, a 50/50 split, a 14-day runtime, 200k sessions per day, $\alpha=0.05$, and $\text{power}=0.8$, are we sufficiently powered? If not, show what knobs you would adjust—MDE, traffic allocation, duration—and justify the risk controls for marketplace interference and SUTVA violations.

e) Outline what you would ship if the experiment results vary by city, device, and traffic channel.

Overview: This question tests a candidate's skills in marketplace analytics, causal inference, experiment design, metric decomposition, and power analysis for conversion optimization.