During a project presentation you propose building an ETA model, but stakeholders insist the team is “pure analytics.” Make a case for or against an ML model to predict delivery time from rider pickup to customer drop-off (consistent with the delivery definition). 1) Define the target rigorously (handling cancellations, reassignments, multi-stop batches) and list features available only at dispatch time to avoid leakage. 2) Propose an offline evaluation plan (metrics like MAE and P90 error across distance/time-of-day/platform strata), calibration checks, and out-of-time validation to capture seasonality. 3) Design the online rollout: guardrail metrics (cost/order, SLA breaches), shadow vs. interleaved traffic, fallback heuristics on model outage, and fairness across zones. 4) Detail a simple-but-strong baseline (segment-aware median) and the minimal measurable lift needed to justify productionization given engineering cost. 5) If you decide against ML, prescribe an analytics-first alternative (policy changes, pricing/surge rules) and a decision tree that revisits ML only when predefined trigger thresholds are met.