Deliver a reproducible exploratory evidence pack

Understand the method

An exploratory analysis is complete when another person can understand what you inspected, reproduce the key results and see how strongly the evidence supports your proposed action. A folder of charts and a long list of metrics is not enough. Organise the work around Northstar’s question: does this supplied extract justify scaling the initiative? Your answer should explain the observed change, the conditions that affect it and what evidence is still needed.

Build a small evidence pack

Keep the untouched raw extract, your prepared analytical copy, the quality log and an analysis worksheet or script. Add a scope note defining one-order grain, April–May dates, EUR units and completed-status eligibility. Include a row bridge from 46 records to 44 verified orders and then 40 completed orders. This bridge helps a reviewer trace inclusion decisions before reading a monetary result.

Your monthly summary should contain count, total value, AOV and median. Your channel table should reconcile to the monthly counts and values, show the order shares and disclose sparse groups. A separate delivery table should show measured and eligible counts alongside averages. Keep actual results separate from the equal-mix and outlier-exclusion scenarios. The same label should never refer to two populations.

Choose graphics that answer a question

For distribution, create the five-bin frequency chart from Lesson 4. State the interval boundaries and counts so the chart is reproducible. Keep the €1,000 order represented. For composition, use grouped columns showing April and May AOV for web, store and the overall population, or a small multiple with a common EUR axis. Include group counts in the adjacent table. A second chart of channel order shares can help explain the changing weights.

Use a zero baseline for value bars, clear month labels and a restrained colour pair consistent across charts. Provide the numeric table and a text description so colour is not the only source of meaning. Do not combine order counts and EUR on a single unlabeled scale. A two-month teaching extract does not establish a seasonal pattern or a stable trend; avoid adding a forecast line to make the graphic feel more sophisticated.

Write a decision note

In 120–180 words, state the scope, observed change, major interpretation issue, data-quality treatment, coverage limitation and next evidence-gathering action. Name counts and denominators where they influence the conclusion. Explain that the store baseline has two orders and contains A020 at €1,000. Keep the legitimate order in the baseline and label any exclusion scenario. Do not invent an advertising return, a profit estimate or a causal effect from the supplied fields.

Check reproducibility and boundaries

Reopen your files and repeat the calculations from your prepared inputs. Compare your monthly and channel outputs with the downloadable results after attempting the exercises. If using the optional Python worked analysis, read its comments and run it on a local practice copy; it does not require third-party packages, credentials or a website connection. It validates inputs and prints JSON without modifying the practice files.

The rubric supports self-review rather than an accredited assessment. A concise analysis with explicit uncertainty is stronger than a confident recommendation built on an unsupported metric. Your final action may be to collect comparable periods, verify the channel-mix explanation and obtain cost information before any expansion decision. That is a practical analytical outcome even when the answer is not “scale now.”