Prescriptive Analytics
Another street fair is coming next month. Nisha knows what happened last time, she knows why, and she has a forecast of about 95 cold coffees for the day. Now comes the question she actually cares about: "So what should I do?" That is the fourth and last type of analytics.
The rule
Prescriptive analytics answers "What should we do?" by comparing options against a goal.
A recommendation needs four ingredients:
- A goal: what counts as better. More profit, fewer lost sales, shorter queues.
- Options: the actions actually on the table, including "do nothing".
- Predicted outcomes: for each option, what is likely to happen, which comes from predictive work.
- Constraints: budget, staff, space, rules.
At large scale this becomes optimisation software that tests thousands of options. At café scale, a small comparison table does the job:
| option | extra cost | expected extra cups sold | extra profit |
|---|---|---|---|
| Do nothing | ₹0 | 0 | ₹0 |
| Extra barista 11–3 | ₹800 | 30 | ₹1,700 |
| Pre-made bottles | ₹300 | 15 | ₹975 |
Each extra cup earns ₹85 after ingredients. The barista option: 30 × 85 = 2,550, minus 800 = 1,700. The bottles: 15 × 85 = 1,275, minus 300 = 975. The barista wins on the goal of profit, if the budget allows ₹800.
The shortcut
Line the four types up as a ladder and climb it: What happened? → Why? → What next? → What should we do? A prescriptive answer is only as strong as the rungs below it.
Descriptive
The last fair Saturday sold 95 cups against a usual 42.
The trap is recommending without the lower rungs. "Hire more staff" with no evidence of why sales spiked or what to expect is a guess wearing a suit.
Drills
Answer fast — trust the rule.
Which question is prescriptive?
Match the question "Why did Monday sales fall?" to its type.
Each extra cup earns ₹85 and a barista costs ₹800. What is the least number of extra cups the barista must bring in to pay for themselves?
Which ingredient does a prescriptive recommendation borrow directly from predictive analytics?
Which is the correct order of the four types, from simplest to most advanced?
Remember
- Prescriptive analytics answers "What should we do?" by comparing options against a goal.
- Shortcut: climb the ladder: what happened, why, what next, what to do.
- Trap: a recommendation with no evidence underneath is a guess.