The primary MCQ skill, drilled
◈ 8 cardsGiven one sentence describing a task, name its CRISP-DM phase and task — using the purpose test, with the adjacent phase excluded.
The purpose test
Every "which phase?" question resolves the same way: ask what is this step for? — not which tool it uses, not which sheet it is done in. Six purposes, six phases:
- about the decision and what would change it → business understanding
- about looking at the data without changing it → data understanding
- about changing the table → data preparation
- about producing the summary or line that answers the question → modelling
- about whether the result answers the question → evaluation
- about handing it over and keeping it current → deployment
The same tool serves different phases. A histogram drawn to see the shape is understanding; the same histogram placed in the final report is deployment. A COUNTIF that counts blanks is verify-quality; a COUNTIF that counts orders over $500 to answer the question is the model.
Worked example — eight tasks from the Maple & Birch project
| Task | Purpose | Phase → task |
|---|---|---|
| Agree with the CFO that a $100 gap would change the store decision | the decision | business understanding → objectives/criterion |
| Write a data dictionary for the exported orders file | look and record | data understanding → describe |
Count blank amount cells and note there are none | look | data understanding → verify quality |
Merge the customer sheet into the orders sheet by cust_id | change the table | data preparation → integrate |
| Add a column of amount per unit | change the table | data preparation → construct |
| Build a pivot of average amount by province | answer the question | modelling → build |
| Note that the AB average rests on two orders and cannot support the decision | does it answer? | evaluation → evaluate results |
| Schedule the dashboard to refresh monthly and name its owner | hand-over upkeep | deployment → monitoring and maintenance |
Notice the two adjacent-phase traps in the table. Describe (dictionary) and construct (new column) are both done in the same sheet minutes apart, but one records what is there and the other changes it. Build (the pivot) and evaluate (the note about two orders) look at the same pivot, but one produces the result and the other judges it against the decision.
Tag six on your own
(1) Convert the order_date column from text to real dates — preparation → format. (2) Draw a scatter of units against amount to see whether they move together — understanding → explore. (3) Decide the memo will go to the owner as one page with one chart — deployment → plan deployment. (4) Fit a line predicting amount from units — modelling → build. (5) Record that the export was pulled from the online store's admin panel on 3 April — understanding → collect. (6) Conclude that the fitted line would not change the pricing decision — evaluation.
The five-step scenario, cold
This is the final's signature shape. An analyst (a) confirms with the owner that a 20 % month-over-month drop in Alberta orders would pause the store plan; (b) adds a month column from the order dates; (c) builds a pivot of order counts by month for Alberta; (d) observes the count fell from one to one — no drop, and too few orders to say; (e) emails the owner a one-line update. Phase each step before checking: (a) business understanding, (b) preparation → construct, (c) modelling → build, (d) evaluation, (e) deployment. If you placed (b) in understanding, ask the purpose question again: the table changed.
CRISP-DM: this lesson is the phase-tagging every later lesson ends with.