The Four Types of Analytics

Four questions, four jobs

Lena's budget decision contains several smaller questions, and they are not all the same kind. A common way to sort them uses four types. The types do not rank analysts from junior to senior. Each answers a different question and needs different evidence.

• Descriptive: what happened? Revenue last quarter, orders per region.
• Diagnostic: why did it happen? Which part of the business moved a number.
• Predictive: what is likely to happen? Next quarter's revenue if nothing changes.
• Prescriptive: what should we do? Which budget split is most likely to pay off.

Descriptive: count before you explain

Everything starts with a correct description. In the first quarter of 2026 Alpenkorb delivered 648 orders worth CHF 37,107.20 after discounts. The same quarter a year earlier had 649 delivered orders worth CHF 36,756.76. Revenue grew by about 1 % with almost exactly the same number of orders. That one comparison already rules out a story Lena might hear in her first week: that the business is growing fast.

SELECT COUNT(*) AS orders,
       ROUND(SUM(net_amount), 2) AS revenue
FROM orders
WHERE status = 'delivered'
  AND order_date >= '2025-01-01'
  AND order_date < '2025-04-01';

COUNT(*) counts rows and SUM adds up a column; AS gives the result a name. WHERE keeps only the rows that meet every condition joined by AND. The date filter uses greater or equal for the first day and strictly less for the day after the period, which works for every month length.

Diagnostic: take the number apart

July 2025 was the weakest month: CHF 7,963.79 of delivered revenue against CHF 12,422.21 in June. Revenue is the number of orders times the average order value, so a diagnostic asks which factor moved. Orders fell from 203 to 147. The average order fell less, from CHF 61.19 to CHF 54.18, and the number of different customers who ordered fell from 157 to 127. Most of the drop came from fewer customers ordering, which fits a holiday month, not from smaller baskets.

Predictive and prescriptive

A prediction does not need a model to be useful. A simple baseline for the second quarter of 2026 is the second quarter of 2025, CHF 37,265.67, adjusted by the 1 % growth seen in the first quarter: about CHF 37,600. Any plan Lena proposes should be judged against that baseline, not against the first quarter, because seasons differ.

Prescriptive analytics is where the decision lives: given what happened, why it happened and what is likely, which option is best? It rests on the three types before it. A budget recommendation built on a wrong description is confidently wrong.

Lena's own questions fall neatly into the four boxes. What did each channel cost last quarter, and how many customers did it bring? That is descriptive. Why did the share of social visits that ended in an order fall after September? Diagnostic. What will the second quarter bring if nothing changes? Predictive. How should the budget be split? Prescriptive. The rest of this course works through them in that order, because each answer is an input to the next.