Analytics · automated report

analytics-pipeline

A deliberately messy fictional sales CSV becomes a repeatable KPI report—with every cleaning action and anomaly surfaced for review.

Public · synthetic demo

The problem

A polished chart can still be confidently wrong.

Duplicates, invalid quantities, missing values, and fat-fingered revenue silently distort dashboards. The useful deliverable is a report that shows both the numbers and the reasons a human should question them.

Input

Injected messy rows from the fictional dataset

order_id, order_date, region, product, quantity, unit_price, revenue
10262, 2026-04-05, West, Gadget, 9, 42.00, 25000.00
10141, 2026-04-13, South, Widget, 7, 18.50, 393.50

Raw rows also include exact duplicates, missing regions,
non-positive quantities, mixed dates, and missing revenue.

The money shot

Actual committed report output

Total revenue$120,691.08

Clean orders397

Average order$304.01

Unique customers113

Revenue by region chart from the committed synthetic analytics report — West contributes 43.5% of recorded revenue in the synthetic output.

Monthly revenue chart from the committed synthetic analytics report — April's 340.2% month-over-month jump is flagged, not silently celebrated.

Check	Actual result	Action
Revenue outlier	Order 10262: $25,000 · z-score 19.29	Review
Revenue ≠ qty × price	7 rows	Review
\|MoM change\| ≥ 25%	3 months	Review
Exact duplicates	7 rows	Dropped + logged
Missing revenue	8 rows	Recomputed + logged

How it's verified

Cleaning and anomaly checks ship with the report.

The pipeline records duplicate removal, invalid-row handling, region normalization, and revenue reconstruction. Tests verify KPI consistency and region shares; deterministic checks identify z-score outliers, arithmetic mismatches, and large monthly swings.

Honest limitations

All data is fictional; no client data is included.
The cleaning rules and metrics are readable starting points that must be tuned to a real client's schema and business definitions.
A flagged outlier is not automatically an error. It remains in the reported total and requires human review.
The charts describe this fixed synthetic dataset; they are not a claim about future business performance.