Toy Order Completion Analysis

A lightweight analytics exercise on a toy orders dataset: compute a simple slice metric (e.g. completion rate for a given city and order type), define the numerator/denominator precisely, handle edge cases, then reason through a short root-cause case study for why the metric might be low. Problem You are given a small table of orders and asked for a simple slice metric, for example: What is the completion rate for a given city and order_type? How would you define the numerator and denominator? What edge cases would you guard against? The emphasis is on clean reasoning rather than memorized pandas syntax. If you do not remember the exact DataFrame APIs, you can treat the dataset as a plain Python structure such as a list of dictionaries or a dict-of-dicts and still solve the question correctly. Clarify the metric definition up front. In many real datasets, you would want to ask whether the denominator includes: all orders in the slice only terminal orders or only orders that reached a certain stage For a toy dataset, the default assumption is usually: numerator = rows with status == "completed" denominator = all rows in the filtered slice unless there are non-terminal states such as pending or in_progress that should be excluded. After the metric calculation, there is a short case-study follow-up: If completion rate is low, what might be causing it? How would you investigate the root cause? What product, operational, or ML changes would you try? A representative toy schema could look like this: rows = [ {"order_id": 1, "city": "SF", "order_type": "grocery", "status": "completed"}, {"order_id": 2, "city": "SF", "order_type": "grocery", "status": "cancelled"}, {"order_id": 3, "city": "SF", "order_type": "ride", "status": "completed"}, {"order_id": 4, "city": "NYC", "order_type": "grocery", "status": "failed"}, ] Recommended Solution The strongest answer starts by defining the metric precisely: Filter to the requested city and order type Let completion rate be completed_orders / total_orders_in_slice Handle the empty-slice case explicitly Call out any assumptions about whether non-terminal rows are included If you have pandas available, the solution is short: def completion_rate_pandas(df, city: str, order_type: str) -> float: mask = (df["city"] == city) & (df["order_type"] == order_type) subset = df.loc[mask, "status"] if subset.empty: return 0.0 return subset.eq("completed").mean() If you can ignore pandas syntax, solve the same logic directly in Python: def completion_rate(rows: list[dict], city: str, order_type: str) -> float: filtered = [ row for row in rows if row["city"] == city and row["order_type"] == order_type ] if not filtered: return 0.0 completed = sum(1 for row in filtered if row["status"] == "completed") return completed / len(filtered) If a dict-of-dicts style solution is wanted, the exact same logic still works: def completion_rate_dict(data: dict[str, dict], city: str, order_type: str) -> float: total = 0 completed = 0 for row in data.values(): if row["city"] != city or row["order_type"] != order_type: continue total += 1 if row["status"] == "completed": completed += 1 return 0.0 if total == 0 else completed / total This runs in O(n) time. The list-based version above uses O(n) extra space because it materializes the filtered rows, while the counting version uses O(1) extra space. For the case-study follow-up, a strong answer is structured instead of speculative. Common reasons for low completion rate include: supply-demand imbalance in a city or time window poor dispatch or ranking decisions inaccurate ETA or pricing that leads to user cancellations courier cancellations, long pickup times, or merchant inventory issues payment failures, fraud rules, or address quality problems A solid investigation plan is: Break the metric down by city, neighborhood, hour, order type, merchant, and cancel reason. Compare low-performing slices against healthy baselines. Inspect funnel drop-off points such as accepted, picked up, and delivered. Check whether model outputs such as ETA, dispatch ranking, or fraud decisions are correlated with failures. Reasonable improvements include better courier-order matching, targeted incentives during supply shortages, ETA model calibration, and clearer operational guardrails for high-failure merchants or areas.