Types of Data Analytics
Analytics is conventionally divided into four types, ordered by increasing difficulty and increasing business value.
The Four Types
| Type | Question | Techniques | Example | Value | Difficulty |
|---|---|---|---|---|---|
| Descriptive | What happened? | Aggregation, summary statistics, dashboards, reports | "Sales fell 12% in Q2" | Low | Low |
| Diagnostic | Why did it happen? | Drill-down, correlation, root-cause analysis, data mining | "The fall came entirely from the Delhi region after a competitor's launch" | Medium | Medium |
| Predictive | What will happen? | Regression, classification, time series, ML | "Q3 sales will fall another 6% if nothing changes" | High | High |
| Prescriptive | What should we do? | Optimisation, simulation, decision analysis, reinforcement learning | "Cut price 5% in Delhi and shift ad budget — projected +9% recovery" | Highest | Highest |
Some textbooks add a fifth:
- Cognitive analytics — self-learning systems that use AI/NLP to reason over unstructured data and improve with feedback (e.g. an assistant that reads support tickets and recommends actions).
Worked Illustration — One Dataset, Four Questions
import pandas as pd
df = pd.DataFrame({
"month": ["Jan", "Feb", "Mar", "Apr", "May", "Jun"],
"sales": [520, 505, 498, 460, 441, 430],
"ad_spend": [50, 48, 47, 30, 28, 25],
})
# 1. DESCRIPTIVE — what happened?
print("Total:", df["sales"].sum(), "| Mean:", round(df["sales"].mean(), 1))
print("Change Jan->Jun:", round((df["sales"].iloc[-1] / df["sales"].iloc[0] - 1) * 100, 1), "%")
# Total: 2854 | Mean: 475.7
# Change Jan->Jun: -17.3 %
# 2. DIAGNOSTIC — why did it happen?
print("Correlation sales vs ad_spend:", round(df["sales"].corr(df["ad_spend"]), 3))
# Correlation sales vs ad_spend: 0.966 -> the drop tracks the ad-spend cut
# 3. PREDICTIVE — what will happen next?
from sklearn.linear_model import LinearRegression
import numpy as np
X = df[["ad_spend"]]
y = df["sales"]
model = LinearRegression().fit(X, y)
print("Predicted sales at ad_spend=25:", round(model.predict([[25]])[0], 1))
# Predicted sales at ad_spend=25: 432.9
# 4. PRESCRIPTIVE — what should we do?
target = 520
needed = (target - model.intercept_) / model.coef_[0]
print(f"Ad spend needed to reach {target}: {needed:.1f} lakh")
# Ad spend needed to reach 520: 50.4 lakhNotice the escalation: descriptive reports the decline, diagnostic explains it, predictive projects it forward, and prescriptive recommends the lever to pull.
Descriptive vs Predictive Analytics (Exam Comparison)
| Basis | Descriptive Analytics | Predictive Analytics |
|---|---|---|
| Time orientation | Past | Future |
| Question | What happened? | What is likely to happen? |
| Output | Summaries, KPIs, dashboards | Probabilities, forecasts, classifications |
| Techniques | Mean, median, frequency, clustering, association rules | Regression, classification, time series, neural networks |
| Certainty | Factual | Probabilistic (comes with error/confidence) |
| Unit 3 mapping | Clustering, Apriori | Naïve Bayes, KNN, linear regression |
Note the Unit 3 link: in this syllabus, descriptive analytics corresponds to unsupervised learning (clustering, association rules — describing structure in data), while predictive analytics corresponds to supervised learning (classification, regression — predicting a labelled outcome).
Analytics Maturity
Most organisations sit at levels 1–2 and try to jump to level 4 — which fails, because predictive models are only as good as the descriptive foundations (clean data, defined metrics, trusted dashboards) beneath them.