Leveraging MongoDB's aggregation framework for business intelligence insights.
Instruction: Describe how you would use MongoDB's aggregation framework to generate actionable business intelligence insights from raw data.
Updated
Official answer
Start with a decision and a metric contract: eligible population, time window, timezone, currency, refunds, missing data and denominator. Aggregation can summarize data accurately only after those choices are explicit.
For average order value, use one eligible order per record or first collapse line items to that grain. If three completed orders have net values 40, 60 and 100 in one currency, net revenue is 200 and average order value is 200/3. Unwinding two lines on an order and then counting rows would inflate the denominator. Deduplicate source events and handle refunds under the stated policy.
Build a pipeline with selective match, deliberate joins/unwind, grouping and final projection. Reconcile counts/totals against a small known fixture and a separate control total. Retention requires a defined cohort and observation period; a simple average-purchase-times-frequency proxy is not automatically a discounted lifetime-profit model.
Present trends with as-of time, coverage and relevant segments. Observational changes do not establish causation. For scheduled summaries, use idempotent refreshes, detect stale/failed runs and preserve prior results with a clear stale state.
Inspect query plans and memory limits, then test duplicates, empty cohorts, mixed currencies and late adjustments. Measure both data quality and decision usefulness.