From business question to recommendation¶

A useful analysis supports a decision. A chart becomes trustworthy when its population, calculations and limitations are clear.

We follow the same seven steps in every lesson and case.

The seven steps¶

Ask: Who must decide what, and by when?

Keep: the decision and a one-sentence question.

Ask: What does each row represent? What is missing?

Keep: the grain, keys, units, data dictionary and scope.

Ask: What changes are needed, and could someone else repeat them?

Keep: an untouched copy of the original data and a log of every change.

Ask: Which comparison or metric actually answers the question?

Keep: the metric definition, filters and calculation.

Ask: Do totals reconcile? Can I explain any odd values?

Keep: a manual spot check, count and total checks, and edge cases.

Ask: Can the audience see the important comparison at a glance?

Keep: a labelled chart showing the period, units and population.

Ask: What action follows, and what is still uncertain?

Keep: an action linked to evidence, one caveat and a follow-up measure.

Why are orders falling?

A manager asks: "Why are orders falling?"

The data file has one row per order line, so a single order with three products takes up three rows.

Counting rows would count lines, not orders.
First, define what an order is, which period to look at and which orders are eligible.
Then count distinct order IDs in that population.
Compare the result with an order-level summary before reading anything into the trend.

The same care applies elsewhere in the course:

Work through these checks in order:

Then write down why the wrong answer was tempting and what evidence supports the corrected one.

A plausible number is not a checked number

Looking reasonable is not the same as being validated.

A pattern in the data can tell you where to look. It does not prove that a discount, campaign, carrier or HR policy caused a result.

When a decision depends on cause and effect, recommend a follow-up comparison or a test.