Business and data analysis: applied workshop

Analysis questions and metrics

Use analysis questions and metrics to make decisions inspectable.

Define the metric, denominator, period and dimensions of a business question.

Teaching example: Retention needs a defined cohort and return window.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Data grain and cleaning

Use data grain and cleaning to make decisions inspectable.

Define one row and identify missing, duplicate and unusual records.

Teaching example: Orders and order lines have different row grain.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

SQL filtering and aggregation

Use sql filtering and aggregation to make decisions inspectable.

Practise SELECT, WHERE and GROUP BY and check results by hand.

Teaching example: COUNT(*) counts rows, not necessarily unique users.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Joins and row multiplication

Use joins and row multiplication to make decisions inspectable.

Join customers and orders and reconcile row counts and totals.

Teaching example: A one-to-many join can repeat customer rows.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Charts and interpretation

Use charts and interpretation to make decisions inspectable.

Choose a comparison chart and write two observations.

Teaching example: A truncated axis can exaggerate a difference.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Experiments and inference

Use experiments and inference to make decisions inspectable.

Distinguish observational comparisons from randomised experiments.

Teaching example: Higher spending by purchasers does not establish campaign causation.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Recommendations and validation

Use recommendations and validation to make decisions inspectable.

Document metric definitions, findings, limitations and next steps.

Teaching example: Recommendations include monitoring metrics and validation conditions.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Worked example · fictional

From a problem to inspectable work

Synthetic channels: A has 100 users and ten conversions; B has 900 and 180; aligned windows and definitions, no overlapping users.

1. (10+180)/(100+900)=19%; the unweighted mean of 10% and 20% is inappropriate.
2. Further judgment: Revenue grows 50% after a user-table join. What check?
3. Inspect join cardinality, duplicate keys, deduplication and aggregation unit

Deliver business metrics and data analysis report: the calculation reference is 19; the recommendation must respect Inspect join cardinality, duplicate keys, deduplication and aggregation unit. Separate numbers, inference and unknowns.

Failure case and repair

In Joins and row multiplication, copy tool output without definitions, then retain the conclusion when asked: Revenue grows 50% after a user-table join. What check?

Even a result of 19 does not explain when Inspect join cardinality, duplicate keys, deduplication and aggregation unit is supported.

(10+180)/(100+900)=19%; the unweighted mean of 10% and 20% is inappropriate. Then test assumptions in Experiments and inference.