Maven Analytics Data Playground: Choose a Dataset for a Useful Portfolio
Evaluate a practice dataset, write a defensible project brief and avoid producing another dashboard without a clear analytical question.
Choose the question before the dashboard
Maven Analytics’ Data Playground offers sample datasets for practice. It is a starting point for finding data, not a substitute for defining a project. This guide is an independent learning resource; it is not affiliated with Maven Analytics.
Before downloading a file, write a question narrow enough to answer with the available fields. “Analyse retail” is too broad. “Which stores contributed most to the change in recorded sales between two complete months?” gives you a metric, a comparison and a useful breakdown.
Read the dataset description, source and any accompanying terms. Public availability does not by itself establish permission to redistribute every file. Keep the original source link and record the download date in your project notes. Do not present practice data as confidential work completed for a client.
Use a dataset acceptance checklist
| Check | What to record | Reason to pause |
|---|---|---|
| Grain | One row represents a transaction, item, customer or period | No reliable way to identify what a row means |
| Coverage | Start and end dates; included locations | One comparison month is incomplete |
| Definitions | Revenue, quantity, cancellation and currency rules | The intended outcome is not measured |
| Keys | Identifiers and expected table relationships | Unexplained duplicate lookup keys |
| Provenance | Original source, description and reuse terms | You cannot explain where the data came from |
You do not need every field to be perfect. A documented limitation can become part of a useful analysis. However, missing essential information should change the question. If a file contains sales but no costs, describe sales performance; do not label the result profit or margin.
Inspect a small sample before loading every table into a dashboard. Check date parsing, missing identifiers and the range of numeric values. Keep an unchanged copy of the input so that a cleaning decision can be revisited.
Write a one-page project brief
Here is an original project brief you can adapt after checking your chosen dataset. The figures below are synthetic and are not claimed to come from Maven.
Grain: One order line.
Metric: Quantity × unit price, summed in one currency under a documented returns rule.
Output: A reconciled store comparison and two follow-up questions.
Limit: Sales data alone cannot establish the cause of a change.
| Store | Month A | Month B | Change |
|---|---|---|---|
| North | 10000 | 12000 | +2000 |
| South | 8000 | 6000 | −2000 |
| West | 6000 | 7000 | +1000 |
| Total | 24000 | 25000 | +1000 |
The total rises by 1,000, or about 4.17%. North adds 2,000 while South offsets that increase by 2,000. West supplies the remaining 1,000. Showing only the total would hide the offsetting store movements.
South’s fall is a place to investigate, not evidence that its manager performed poorly. Check missing days, stock availability, store opening hours and changes in product mix. Those explanations require additional evidence.
Build an analysis someone can reproduce
- Save the project brief and the source information beside your analysis.
- Write the cleaning rules, including how you treat returns, duplicate lines and missing prices.
- Calculate the total for each month before adding store details.
- Check that the store changes sum to the overall change.
- Use a simple comparison table or bar chart with currency and period labels.
- Explain one finding, one limitation and the next evidence you would request.
A screenshot is useful, but the reviewer also needs to understand how you got there. Include queries or a notebook where appropriate, a small data dictionary and instructions for rebuilding the result. If redistribution is restricted, link to the original data source and describe the import steps instead of uploading the raw files.
Make the project your own
Choose a specific definition to defend. For example, compare calendar months or equal numbers of trading days, and explain why. Explore whether an apparent decline disappears after excluding incomplete reporting days. Show an example of a failed validation check and how you resolved it.
Avoid copying another learner’s dashboard and changing only its colours. A modest analysis with traceable calculations gives you more to discuss than a large collection of unexplained charts. You can also extend the same dataset with a second question, but keep its metric and evidence separate.
Before publishing, ask a reader to find the question, the denominator and the limitation without opening your code. If those are hard to locate, revise the written case study. Continue with the portfolio storytelling example or the SQL revenue investigation.