Descriptive, Diagnostic, Predictive, and Prescriptive Analytics — With Examples

What You Will Learn

• What each of the four analytics types actually produces
• The typical tools used for each type
• A worked business scenario that runs through all four
• How to identify which type a piece of work belongs to

Why This Topic Matters

Lesson 07 introduced the four types. This lesson makes them concrete. Without concrete examples, the words "descriptive" and "predictive" feel abstract and slip out of memory. With them, you will recognize the type of any analysis you encounter within seconds.

The Scenario

You work for an online grocery delivery app. Last week the support team was flooded with complaints about late deliveries. Your manager asks: "What's going on with deliveries?" You will answer this question four different ways, one for each analytics type.

1. Descriptive Analytics — What Happened?

Output: A clear, factual summary of the situation.

What you do

1. Pull the last 4 weeks of delivery data from the database.
2. Compute the average delivery time per week.
3. Compute the percentage of deliveries that were late (> 45 minutes).
4. Break it down by city and by time of day.

Example finding

"Average delivery time rose from 32 minutes in W1 to 47 minutes in W4. The share of late deliveries nearly quadrupled, from 8% to 31%."

Typical tools

• SQL aggregations (SUM, AVG, COUNT) — see lesson 20
• Excel pivot tables — see lesson 15
• pandas groupby — see lesson 26
• Bar and line charts — see lesson 33

2. Diagnostic Analytics — Why Did It Happen?

Output: An explanation, supported by comparisons across groups.

What you do

1. Compare late vs on-time deliveries. What is different about the late ones?
2. Break down by city, store, time of day, order size, rider supply.
3. Look for a variable that separates late from on-time deliveries cleanly.

Example finding

"The spike is almost entirely in Mumbai. Delhi and Bengaluru barely changed. Drilling into Mumbai: 70% of late deliveries came from orders picked at the Andheri store between 6 PM and 9 PM."

Typical tools

• SQL GROUP BY + comparisons — see lesson 21
• Excel filters and pivot tables
• Segmentation analysis — see lesson 37

3. Predictive Analytics — What Will Happen Next?

Output: A forecast, with a stated confidence range.

What you do

1. Use the historical pattern to project next week's average delivery time.
2. State the assumptions: "if Mumbai's Andheri store keeps the same order volume and rider count".
3. Give a range, not a single number — predictions are always uncertain.

Example finding

"If nothing changes, next week's average delivery time in Mumbai is likely to be 48–55 minutes, with the late-share rising to 33–37%. The Andheri store alone accounts for ~60% of this projected shortfall."

Typical tools

• Time-series forecasting (moving averages, exponential smoothing)
• Regression models
• Machine learning (beyond this course)

4. Prescriptive Analytics — What Should We Do?

Output: A recommendation, with expected impact and trade-offs.

What you do

1. Use the diagnosis to propose a specific action.
2. Estimate the impact of the action on the metric.
3. List the cost / trade-off.

Example recommendation

"Add 3 riders to the Andheri store for the 6–9 PM window for the next 2 weeks. Based on current order volume (180 orders/peak hour) and current rider supply (12 riders), 3 additional riders should bring average delivery time back to ~36 minutes and the late-share down to ~12%. Cost: ₹45,000/week. Expected retention benefit: ~₹2,00,000/week in retained orders that would otherwise be refunded."

Typical tools

• Scenario modeling in Excel or Python
• A/B testing to validate the recommendation
• Optimization algorithms (beyond this course)

The Four Types in One Picture

How to Identify the Type of Any Analysis

Read or listen to the conclusion of the analysis. Then match:

• Ends with a number or trend → Descriptive
• Ends with "because..." → Diagnostic
• Ends with "will be..." or "likely to..." → Predictive
• Ends with "we should..." → Prescriptive

This single test correctly classifies 90% of real analyses you will encounter.

Common Mistakes

1. Mistaking descriptive for diagnostic. "Sales are down" is descriptive. "Sales are down because the Mumbai store was closed" is diagnostic. Many reports stop at the first.
2. Making predictions without diagnosing first. A forecast that ignores the underlying driver is just curve-fitting.
3. Prescribing without estimating impact. "We should add more riders" is a slogan, not a prescription. A prescription includes the expected outcome.
4. Skipping the confidence range on predictions. Single-number forecasts always look more confident than they actually are.

Practical Exercise (5 minutes)

Read any data-related article or business report. Identify the type of analytics in it by looking at the conclusion. Was the type the author intended, or did they claim predictive but only deliver descriptive?

Mini Challenge

Take any number from your own life — your monthly expenses, your daily screen time, your weekly step count. Run it through all four analytics types in 4 short sentences. Notice how each type adds value but also adds assumptions.

Key Takeaways