index Introduction
Concepts, Techniques and Tools

Data Exploration Descriptive Analysis

What is Happening?
Set 01 · Advanced Analytical Skills

Data Exploration & Descriptive Analysis

Before you analyse anything — you must first understand what you are looking at.

Descriptive analysis is the foundation of every analytical process. It answers the most fundamental question in business intelligence: What happened? Before drawing conclusions, building models, or making recommendations, every analyst — regardless of technical level — must first explore and describe the data they are working with.

You cannot explain what you have not yet understood. Descriptive analysis is how understanding begins.

This set is designed for all experience levels. Whether you have never opened a statistical tool before or you work with data daily, the techniques in this set form the bedrock of sound analytical practice. They are used at the start of every project — and revisited throughout.

Where this sits in the Analytics Ladder

Level 1 — Describe

The analytics ladder moves from description through diagnosis, explanation, forecasting, classification, segmentation, decision-making, and optimisation. You cannot climb higher rungs reliably without mastering this one. Skipping descriptive analysis leads to misinterpreted models, wrong conclusions, and decisions built on misunderstood data.

What this set covers

Technique 01

Descriptive Statistics

Mean, median, mode, standard deviation, variance, skewness, kurtosis — the numbers that summarise what data looks like.

Technique 02

Frequency & Distributions

How often values appear, how data is shaped, and what that shape tells you about the underlying process.

Technique 03

Cross-Tabulation & Pivot Tables

Summarising relationships between two categorical variables — the fastest way to find patterns in business data.

Technique 04

Data Visualisation

Box plots, heat maps, histograms — turning numbers into pictures that communicate patterns instantly.

Technique 05

KPIs & Dashboards

Consolidating descriptive analysis into monitoring systems that keep decision-makers informed in real time.

Tools Covered

Excel · SPSS · Power BI · Python

Step-by-step guides for every technique across all three tiers — no-code, low-code, and full-code.

A note on technical background

This set assumes no prior statistical knowledge. Every concept is introduced with a plain-language explanation and a real business example before any formula or tool is shown. If you are highly technical, the later tabs provide full process guides with code. Start with the concept — the tool follows naturally.

Set 01 · Technique 01

Descriptive Statistics

The numbers that summarise what your data looks like — before you do anything else.

Descriptive statistics reduce a large set of numbers into a small set of meaningful summaries. Think of them as a health check on your data. They tell you where the centre is, how spread out the values are, and whether the distribution is symmetrical or skewed. Every dataset should be described before it is modelled.

Measures of Central Tendency

These answer: what is the typical value?

MeasureWhat it isWhen to use itBusiness example
Mean The arithmetic average. Sum of all values ÷ count. When data is roughly symmetrical with no extreme outliers. Average monthly sales revenue across 12 months.
Median The middle value when data is sorted. 50% above, 50% below. When data is skewed or contains outliers that distort the mean. Median employee salary — a few very high earners inflate the mean and mislead.
Mode The most frequently occurring value. For categorical data or when you need the most common occurrence. The most common customer complaint category. The most purchased product size.
⚠ The mean vs median trap — one of the most common mistakes in business reporting

A company reports average employee salary of $85,000. Sounds reasonable. But the CEO earns $2M, and 80% of employees earn under $50,000. The median is $47,000 — a completely different picture. Whenever data contains outliers or is heavily skewed, report the median, not the mean. Property prices, incomes, response times, and project costs are classic examples where the median tells the truth and the mean misleads.

Measures of Spread

These answer: how consistent or variable are the values?

MeasureWhat it isBusiness use
Range Maximum − Minimum. Simplest measure of spread. Quick check: "Our delivery times range from 1 to 14 days." But sensitive to a single extreme value.
Standard Deviation (SD) Average distance of each value from the mean. Low SD = values are clustered together. High SD = values are spread widely. Measuring consistency. Low SD in production quality = consistent process. High SD in sales performance = highly variable team.
Variance Standard deviation squared. Less intuitive but used in statistical calculations. Foundation for more advanced analysis. Not typically reported directly to business stakeholders.
Percentiles The value below which X% of observations fall. P25, P50 (= median), P75, P90, P95. SLA management: "90% of support tickets are resolved within 4 hours" = P90 resolution time of 4h.

Skewness & Kurtosis — Shape of the Data

Skewness

Is the distribution symmetrical, or does it lean to one side?

Zero skewness: Symmetrical. Mean = Median. Normal distribution.
Positive skew (right-tailed): Most values are low, but a few extreme high values pull the mean up. Common in income data, claim sizes, response times. The mean is higher than the median.
Negative skew (left-tailed): Most values are high, a few extreme lows drag the mean down. Common in test scores where most people pass, customer satisfaction ratings.

Kurtosis

Are the tails heavy or light — how extreme are the outliers?

High kurtosis (leptokurtic): Sharp peak, heavy tails — extreme values are more common than expected. Risk-relevant: financial returns, fraud amounts.
Low kurtosis (platykurtic): Flat distribution, thin tails — values spread evenly, fewer extremes.
Normal (mesokurtic): Kurtosis = 3. The benchmark.

Practice · Descriptive Stats Interpreter

Each scenario shows a business dataset summary. Choose which measure best answers the business question — and why.

Scenario: 1 / 5
Which measure best answers the business question?
Read the scenario and choose the most appropriate measure.
Set 01 · Technique 02

Frequency & Distributions

How your data is shaped tells you as much as what it contains.

A frequency distribution shows how often each value — or range of values — appears in your dataset. When plotted as a histogram, the shape of this distribution reveals patterns that a single average number completely hides. Two datasets with identical means can have completely different distributions — and completely different business implications.

Common Distribution Shapes

ShapeWhat it looks likeWhat it usually meansBusiness example
Normal (Bell Curve) Symmetrical. Most values cluster around the centre. Tails taper evenly on both sides. A natural, stable process. Variation exists but is predictable and bounded. Heights of employees. Daily production output on a stable line. Test scores in a well-designed assessment.
Right-Skewed (Positive) Peak is on the left. Long tail stretches right. Many low values, a few very high ones. A floor exists (e.g., zero) but no ceiling. A minority of extreme cases pulls the mean up. Customer spending (most spend little, a few spend a lot). Insurance claim amounts. Time to resolve complex issues.
Left-Skewed (Negative) Peak is on the right. Long tail stretches left. Most values are high, a few are very low. A ceiling effect exists. Performance clusters near maximum. Customer satisfaction scores when service is excellent. Days to invoice payment when most customers pay on time.
Bimodal Two peaks. The distribution has two humps. Two distinct subgroups exist in your data and are being mixed together. Product sales across two different customer segments. Two geographic markets with different usage patterns being reported together.
Uniform Flat. All values appear with roughly equal frequency. No dominant pattern. Could indicate random process, data quality issue, or true equiprobability. Dice rolls. Randomly assigned customer IDs. Sometimes seen in survey responses suggesting acquiescence bias.
⚠ Why bimodal distributions are a warning sign

A bimodal distribution often means you are mixing two different populations. A histogram of "customer transaction value" with two peaks might mean you have both retail and wholesale customers in the same dataset. Analysing them together produces a mean that describes neither group accurately. The correct action is to separate them before any further analysis. Distribution shape catches this problem early.

Percentiles in Practice

Percentiles divide your data into 100 equal parts. The P90 value, for example, is the value below which 90% of observations fall.

Business Example
Using percentiles for SLA management

A customer service team tracks response times. The mean is 3.2 hours — which sounds acceptable. But the P90 is 11 hours, meaning 10% of customers wait more than 11 hours. The mean hid an unacceptable tail.

The SLA is set at P90 ≤ 4 hours. This means 90% of customers must receive a response within 4 hours. Tracking only the mean would have obscured a significant service failure affecting 1 in 10 customers.

Reading a Histogram

A histogram groups values into bins (ranges) and shows a bar for each bin. The height of each bar shows how many values fall in that range. Key things to read:

Step 1 — Look at the overall shape

Is it symmetrical, skewed left, skewed right, or bimodal?

This immediately tells you whether the mean is a reliable summary and whether you have a mixed population problem.

Step 2 — Identify where the peak is

Where are most values concentrated?

The modal class (tallest bar) shows the most common range of values. Compare it to the mean and median — are they close together or far apart? Large gaps signal skewness or outliers.

Step 3 — Look at the tails

Are there values far from the centre?

Long tails indicate extreme values. Right tails are especially common in business data (revenue, costs, times). These extremes will strongly influence the mean and should be investigated — they may represent your most important customers or your most serious problems.

Set 01 · Technique 03

Cross-Tabulation & Pivot Tables

The fastest way to find patterns when you have two or more categorical variables.

A cross-tabulation (also called a contingency table or crosstab) places two categorical variables against each other and counts how many observations fall in each combination. It is one of the most powerful and underused tools in business analysis. Most business questions involve comparing groups — and crosstabs are how you do that quickly and clearly.

Business Example
Customer segment × Churn status

A telecoms company wants to understand which customer segments churn most. They cross-tabulate Segment (Premium / Standard / Basic) against Churn Status (Churned / Retained):

SegmentChurnedRetainedChurn Rate
Premium459554.5%
Standard21079021.0%
Basic38062038.0%

The overall churn rate is 19.2% — a number that completely hides the fact that Basic customers churn at over 8× the rate of Premium customers. This single crosstab redefines where the problem actually lives.

Pivot Tables — Dynamic Crosstabs

A PivotTable is an interactive crosstab built inside Excel, Power BI, or similar tools. Instead of being static, it lets you rotate, filter, and reconfigure the analysis without re-entering data. The four zones of a PivotTable:

Zone 01

Rows

The variable whose categories appear as rows. Usually the primary dimension you are analysing (e.g., product category, region, department).

Zone 02

Columns

The variable whose categories appear as columns. Usually a secondary dimension or time period (e.g., month, quarter, status).

Zone 03

Values

The numbers in the cells — counts, sums, averages, percentages. This is what you are measuring at the intersection of each row and column.

Zone 04

Filters

Restricts the entire table to a subset of data. Useful for focusing on a specific time period, region, or product line.

Always look at percentages, not just counts

A crosstab showing 200 churned customers in Region A and 200 in Region B looks balanced. But if Region A has 400 customers total and Region B has 2,000, Region A's churn rate (50%) is catastrophically worse than Region B's (10%). Always convert counts to percentages or rates before drawing conclusions. In Excel PivotTables: right-click a value → Show Values As → % of Row Total or % of Column Total.

Common Business Applications

Business questionRow variableColumn variableValue
Which product category sells best in each region?RegionProduct CategorySum of Revenue
Are defect rates higher in certain shifts?ShiftDefect TypeCount / % of total
Does customer satisfaction vary by service channel?Channel (phone/chat/email)Satisfaction ratingCount / % of row
Which sales rep performs best by product line?Sales RepProduct LineSum of Sales / Avg deal size
Set 01 · Technique 04

Data Visualisation

The right chart communicates in seconds what a table takes minutes to read.

Choosing the right visualisation is as important as choosing the right statistic. The wrong chart does not just look unappealing — it can actively mislead. Every chart type is designed to answer a specific kind of question. Using the wrong one produces an answer to a question nobody asked.

Chart Selection Guide

Chart typeBest forWhen to avoidBusiness use
Bar / Column Chart Comparing quantities across categories When you have more than ~10 categories — use a table instead Revenue by product line · Headcount by department · Support tickets by type
Line Chart Showing change over time (trends) For categorical comparisons with no time element Monthly sales trend · Weekly website traffic · Daily stock price
Histogram Showing the distribution shape of one continuous variable For categorical data or comparisons between groups Distribution of transaction values · Age distribution of customers · Order processing times
Box Plot Comparing distributions across groups, showing spread and outliers When the audience is not data-literate — misread frequently Salary ranges by department · Delivery times by courier · Product scores by market
Scatter Plot Showing relationship between two continuous variables For categorical data or more than two variables Marketing spend vs revenue · Temperature vs energy consumption
Heat Map Showing magnitude at the intersection of two dimensions using colour When precise values matter more than relative magnitude Call centre volume by hour × day · Sales by region × product · Correlation matrices
Pie / Donut Chart Showing part-to-whole for 2–5 categories only For more than 5 categories or for comparing across multiple pies Market share split · Budget allocation · Survey response split

Reading a Box Plot

Box plots are powerful but misunderstood. Here is exactly what each element means:

The Box (IQR — Interquartile Range)

The box spans from Q1 (25th percentile) to Q3 (75th percentile)

This represents the middle 50% of your data. The wider the box, the more variable the middle portion of your data is. A wide IQR means there is substantial spread even among "typical" values.

The Line Inside the Box

This is the median (Q2 — 50th percentile)

If the median line is centred in the box, the middle 50% is symmetrically distributed. If it is closer to Q1, the data is right-skewed. If closer to Q3, left-skewed. A median line sitting very near one end of the box signals strong skewness.

The Whiskers

Extend to the furthest non-outlier values (typically 1.5 × IQR from Q1 and Q3)

Long whiskers indicate a wide spread of non-extreme values. Short whiskers mean most values are close to the median. Any value beyond the whiskers is plotted as an individual point — an outlier.

Individual Dots Beyond the Whiskers

These are outliers

Investigate every outlier. They could be data entry errors, legitimate extreme cases (your largest customer, a one-time event), or signs of a process problem. Never automatically remove outliers — first understand why they exist.

Practice · Chart Type Matcher

Each scenario describes a business question. Select the most appropriate chart type and see why.

Question: 1 / 6
Select the best chart type
Read the business question and select the most appropriate chart type.
Set 01 · Technique 05

KPIs & Dashboards

Descriptive analysis made continuous — monitoring rather than discovering.

A Key Performance Indicator (KPI) is a specific, measurable metric tied to a business objective. A dashboard assembles multiple KPIs into a single view. Together, they turn descriptive analysis from a one-time exercise into an ongoing monitoring system. Good KPIs don't just measure activity — they measure progress toward outcomes that matter.

What Makes a Good KPI

Criterion 01

Specific

Precisely defined. "Customer satisfaction" is not a KPI. "Net Promoter Score measured monthly via post-purchase survey" is.

Criterion 02

Measurable

Can be calculated reliably from available data. If you cannot measure it today, it is an aspiration, not a KPI.

Criterion 03

Actionable

When it moves in the wrong direction, the organisation knows what to do. A KPI that triggers no action is decorative.

Criterion 04

Relevant

Directly tied to a strategic or operational objective. Not measured because it is easy to measure, but because it matters.

Criterion 05

Time-Bound

Has a defined reporting frequency and comparison baseline (vs last month, vs target, vs same period last year).

Criterion 06

Has a Target

A number without a target cannot be judged. "Revenue = $2M" means nothing without "Target = $2.2M." Context is everything.

Leading vs Lagging Indicators

TypeWhat it measuresAdvantageLimitationExample
Lagging indicator Results that have already occurred Accurate — you know exactly what happened Too late to change — the outcome is already determined Monthly revenue · Annual employee turnover rate · Quarterly profit
Leading indicator Early signals that predict future outcomes Actionable — you can still intervene Uncertain — the prediction may not hold Number of qualified leads · Employee engagement score · Days sales outstanding · Customer login frequency
Best practice — balance both types

A dashboard showing only lagging indicators tells you what went wrong after it has already gone wrong. A dashboard showing only leading indicators gives you warnings but no confirmation of actual performance. The strongest dashboards combine both — lagging indicators confirm outcomes; leading indicators provide early warning and time to act.

Dashboard Design Principles

Principle 01 — Maximum 7 KPIs per dashboard view

More than 7 metrics creates cognitive overload

Executives and managers can meaningfully track 5–7 metrics per session. If everything is highlighted, nothing is. Separate strategic dashboards from operational dashboards — they serve different audiences and decisions.

Principle 02 — Show context, not just values

Every metric needs a comparison

A number alone is meaningless. Show vs target, vs prior period, vs benchmark. Use traffic-light indicators (red/amber/green) to make status instantly interpretable. The viewer should understand the situation in under 5 seconds.

Principle 03 — Hierarchy of importance

Place the most important metric top-left

Western readers scan left-to-right, top-to-bottom. The most critical KPI belongs in the top-left. Supporting context metrics belong to the right and below. Use size, position, and colour strategically — not decoratively.

Set 01 · Core Principle
Describe before you conclude. Understand before you model. Every number has a shape, a centre, and a context — and all three matter.
Set 01 · Tool Guide

Excel Step-by-Step

No prior Excel experience required. Follow each step exactly as written.

🟢 No-code / Low-code — Start here

Excel is the most accessible tool for descriptive analysis. Every technique in this set can be performed using built-in Excel functions or the free Data Analysis ToolPak add-in. No programming, no installation beyond enabling an add-in.

Step 1 — Enable the Data Analysis ToolPak (one-time setup)

Required for: Descriptive Statistics, Histograms, Correlation, Regression

How to activate

1. Click FileOptions
2. Click Add-ins in the left panel
3. At the bottom of the screen, next to "Manage:", select Excel Add-ins → click Go
4. Check the box next to Analysis ToolPak → click OK
5. The Data Analysis button now appears in the Data tab, far right

Step 2 — Key Descriptive Formulas

DESCRIPTIVE STATISTICS FORMULAS
=AVERAGE(A2:A100) — Arithmetic mean of a range
=MEDIAN(A2:A100) — Middle value (50th percentile)
=MODE.SNGL(A2:A100) — Most frequent value (single mode)
=MODE.MULT(A2:A100) — Returns all modes (use as array formula)
=STDEV.S(A2:A100) — Sample standard deviation
=STDEV.P(A2:A100) — Population standard deviation
=VAR.S(A2:A100) — Sample variance
=MIN(A2:A100) — Minimum value
=MAX(A2:A100) — Maximum value
=PERCENTILE.INC(A2:A100,0.9) — 90th percentile (P90)
=PERCENTILE.INC(A2:A100,0.25)— Q1 (25th percentile)
=PERCENTILE.INC(A2:A100,0.75)— Q3 (75th percentile)
=SKEW(A2:A100) — Skewness (0=symmetric, +ve=right skew)
=KURT(A2:A100) — Kurtosis (3=normal, >3=heavy tails)
=CORREL(A2:A100,B2:B100) — Pearson correlation between two columns

Step 3 — Run the Descriptive Statistics Tool

Produces a complete summary table in seconds

Procedure

1. Click Data tab → Data Analysis
2. Select Descriptive Statistics → click OK
3. Input Range: click and drag to select your data column (include the header row)
4. Check Labels in First Row if your column has a header
5. Under Output, select New Worksheet Ply (outputs to a new sheet — recommended)
6. Check Summary statistics
7. Optionally check Kth Largest and Kth Smallest to see extreme values
8. Click OK

Output: Mean, Standard Error, Median, Mode, Standard Deviation, Sample Variance, Kurtosis, Skewness, Range, Minimum, Maximum, Sum, Count. All in one table.

Step 4 — Create a PivotTable

PivotTable — the analyst's most versatile tool

Procedure

1. Click anywhere inside your data table
2. Click Insert tab → PivotTable
3. Accept the default range → choose New WorksheetOK
4. In the PivotTable Fields panel on the right: drag a categorical field (e.g., "Region") to Rows
5. Drag a second categorical field (e.g., "Product Category") to Columns
6. Drag a numeric field (e.g., "Revenue") to Values — it will default to Sum
7. To show % instead of totals: right-click any value cell → Show Values As% of Row Total

Step 5 — Create a Histogram

Two methods — Quick and Precise

Quick Method (Excel 2016+)

1. Select your data column
2. Click InsertInsert Statistic ChartHistogram
3. Right-click the x-axis → Format Axis to adjust bin width

Precise Method (Data Analysis ToolPak)

1. Create a "Bin" column with your desired upper boundaries (e.g., 10, 20, 30, 40, 50)
2. DataData AnalysisHistogram
3. Input Range: your data · Bin Range: your bin column
4. Check Chart OutputOK

Set 01 · Tool Guide

SPSS & Power BI Step-by-Step

Professional-grade analysis with graphical interfaces — no code required.

🟡 Low-code / GUI — Go deeper

SPSS and Power BI are industry-standard tools used by analysts and researchers worldwide. Both offer professional statistical output through point-and-click interfaces. SPSS is preferred for academic and research-grade reporting. Power BI is preferred for business dashboards and executive reporting.

IBM SPSS Statistics — Descriptive Analysis

SPSS · Frequencies

Best for: frequency tables, histograms, bar charts from categorical data

1. AnalyzeDescriptive StatisticsFrequencies
2. Move variables from the left panel to the Variable(s) box
3. Click Statistics — check Mean, Median, Mode, Std Deviation, Variance, Range, Minimum, Maximum, Percentile values (25, 50, 75), Skewness, Kurtosis
4. Click Charts — select Histograms → check Show normal curve on histogram
5. Click ContinueOK
Output: Full frequency table + statistics table + histogram with normal curve overlay. SPSS also provides standard errors for skewness and kurtosis — useful for testing normality.

SPSS · Descriptives

Best for: quick summary statistics across multiple variables simultaneously

1. AnalyzeDescriptive StatisticsDescriptives
2. Move variables to Variable(s) box
3. Click Options — select all statistics needed
4. Click ContinueOK
Output: Clean table with N, Minimum, Maximum, Mean, Std Deviation for all selected variables. Best format for reporting to stakeholders.

SPSS · Explore

Best for: comprehensive normality testing, box plots, stem-and-leaf plots

1. AnalyzeDescriptive StatisticsExplore
2. Move continuous variables to Dependent List
3. Optionally move a grouping variable to Factor List (produces side-by-side box plots by group)
4. Click Plots → check Normality plots with tests
5. Click ContinueOK
Output: Statistics table + 95% confidence intervals for mean + box plots (by group if specified) + Kolmogorov-Smirnov and Shapiro-Wilk normality test results.

SPSS · Crosstabs

Best for: cross-tabulation with chi-square test of independence

1. AnalyzeDescriptive StatisticsCrosstabs
2. Move one variable to Row(s), another to Column(s)
3. Click Cells → check Row percentages (and Column if needed)
4. Click Statistics → check Chi-square to test whether the relationship is statistically significant
5. Click ContinueOK
Interpreting chi-square: If Sig. (p-value) < 0.05 → the two variables are significantly associated. If p ≥ 0.05 → no significant association.

Microsoft Power BI — Dashboard & Visual Analysis

Power BI · Building a Descriptive Dashboard

Step-by-step from data to dashboard

1. Load data: HomeGet Data → select source (Excel, SQL, CSV, etc.) → Load
2. Create a KPI card: Click Card visual in the Visualizations pane → drag a numeric field to Fields. Add conditional formatting via FormatCallout value.
3. Create a bar chart: Click Clustered bar chart → drag category to Y-axis → drag value to X-axis. Sort descending: click the "…" menu on the visual → Sort axis.
4. Add a histogram: Click Histogram visual (may require downloading from AppSource) → drag your continuous field to Values.
5. Add a heat map matrix: Click Matrix visual → drag row dimension to Rows, column dimension to Columns, value to Values → enable Conditional formattingBackground color → select colour scale.
6. Add slicers (filters): Click Slicer visual → drag a dimension field. All visuals on the page will filter when a slicer value is selected.

Set 01 · Tool Guide

Python Step-by-Step

Full programmatic control — for analysts who want to automate, scale, or customise.

🔵 Code-based — Full power

Python is the most widely used language for data analysis. The three libraries you need for this set are pandas (data manipulation), matplotlib (charts), and seaborn (statistical visualisations). All are free and installed with: pip install pandas matplotlib seaborn

Loading & Exploring Data

IMPORT AND LOAD DATA
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

# Load from Excel or CSV
df = pd.read_excel('sales_data.xlsx') ← Excel file
df = pd.read_csv('sales_data.csv') ← CSV file

# First look at the data
df.head(10) ← First 10 rows
df.shape ← (rows, columns)
df.info() ← Column names, data types, non-null counts
df.isnull().sum() ← Count missing values per column

Descriptive Statistics

COMPLETE DESCRIPTIVE SUMMARY
# Full summary — count, mean, std, min, percentiles, max
df.describe()

# Include all column types (not just numeric)
df.describe(include='all')

# Individual statistics
df['revenue'].mean()
df['revenue'].median()
df['revenue'].mode()[0]
df['revenue'].std()
df['revenue'].var()
df['revenue'].skew() ← Skewness
df['revenue'].kurtosis() ← Excess kurtosis (normal = 0)
df['revenue'].quantile([0.25, 0.5, 0.75, 0.90]) ← Percentiles

Visualisations

HISTOGRAM
plt.figure(figsize=(10, 5))
sns.histplot(df['revenue'], bins=30, kde=True) ← kde=True adds smoothed curve
plt.title('Revenue Distribution')
plt.xlabel('Revenue ($)')
plt.ylabel('Frequency')
plt.show()
BOX PLOT — COMPARE GROUPS
plt.figure(figsize=(10, 5))
sns.boxplot(x='region', y='revenue', data=df)
plt.title('Revenue Distribution by Region')
plt.show()
HEAT MAP — CORRELATION MATRIX
corr_matrix = df.select_dtypes(include='number').corr()
plt.figure(figsize=(10, 8))
sns.heatmap(corr_matrix, annot=True, fmt='.2f', cmap='coolwarm')
plt.title('Correlation Matrix')
plt.show()
CROSS-TABULATION (CROSSTAB)
# Count crosstab
ct = pd.crosstab(df['segment'], df['churn_status'])

# Percentage crosstab (% of row)
ct_pct = pd.crosstab(df['segment'], df['churn_status'], normalize='index') * 100
ct_pct.round(1)
Set 01 · Practice

Business Practice Lab

Apply the techniques to realistic business scenarios. Three exercises — one per skill level.

Practice Scenario
Retail chain — 12-month performance review

You are an analyst at a retail chain with 8 regional stores. Leadership has asked for a descriptive summary of sales performance for the past 12 months. The following data is available: monthly revenue per store, number of transactions, average transaction value, customer satisfaction score (1–10), and staff headcount.

Work through the three exercises below using the tools available to you.

Practice · Distribution Reader

Each scenario describes a distribution. Identify the shape and explain what it means for the business.

Scenario: 1 / 4
What does this distribution shape tell you?
Read the distribution description and select what it most likely indicates.
Common mistakes in descriptive analysis — checklist

✗ Reporting only the mean when data is skewed — always check the mean vs median gap first.
✗ Drawing conclusions from a single statistic — mean, median, and standard deviation together tell a richer story than any one alone.
✗ Ignoring outliers — they are often the most important data points.
✗ Confusing correlation in a crosstab with causation — a pattern in a crosstab is a candidate for further investigation, not a proved relationship.
✗ Creating dashboards without targets — a metric without a benchmark cannot be evaluated.

Set 01 · Assessment

Knowledge Check

Five questions covering the full set. Select the best answer.

Q1 — A company reports that the average employee salary is $85,000, but 80% of employees earn under $50,000. Which measure should be used to represent the typical salary — and why?

A
Mean — it uses all the data values and is the most accurate
B
Median — the distribution is right-skewed by high earners, and the median represents the typical employee more accurately
C
Mode — the most common salary is the most useful measure
D
Standard deviation — it shows how spread out the salaries are
When data is skewed by extreme values (a few very high salaries), the mean is pulled toward those extremes and no longer represents what is typical. The median — the middle value — is unaffected by extreme values and should always be used for skewed distributions such as salaries, house prices, income, and claim amounts.

Q2 — A histogram of customer transaction values shows two distinct peaks — one around $20 and another around $200. What does this most likely indicate?

A
The data contains errors and should be cleaned before analysis
B
There are two distinct customer groups (e.g., retail and wholesale) being analysed together — they should be segmented before further analysis
C
The data is normally distributed
D
The mean and median are equal
A bimodal distribution (two peaks) almost always indicates two distinct sub-populations within the data. Analysing them together produces a mean that describes neither group accurately. The correct action is to identify the grouping variable (e.g., customer type, channel, geography) and segment the analysis before proceeding.

Q3 — A cross-tabulation shows 150 churned customers in Region A and 150 churned customers in Region B. A manager concludes "churn is equal across regions." What is wrong with this conclusion?

A
Nothing — equal counts means equal churn
B
The conclusion ignores base rates — if Region A has 300 customers (50% churn) and Region B has 3,000 (5% churn), churn is dramatically worse in Region A
C
The sample size is too small to draw any conclusion
D
Cross-tabulations cannot be used to compare regions
Raw counts in a crosstab are meaningless without context. Always convert to percentages or rates. Equal counts across groups of very different sizes indicate a massive difference in rates. In this example, Region A's 50% churn rate vs Region B's 5% rate tells a completely different story than the identical absolute counts suggest.

Q4 — Which Excel function correctly returns the 90th percentile of values in cells A2 to A100?

A
=PERCENTILE(A2:A100, 90)
B
=PERCENTILE.INC(A2:A100, 0.9)
C
=PERCENTILE.INC(A2:A100, 90)
D
=P90(A2:A100)
PERCENTILE.INC is the correct modern function. The key detail: Excel's percentile functions require the percentile to be expressed as a decimal between 0 and 1. So the 90th percentile = 0.9, the 25th percentile = 0.25, the 75th percentile = 0.75. Writing 90 instead of 0.9 returns an error. The older =PERCENTILE() function also works but is less explicit.

Q5 — A dashboard shows monthly revenue of $1.8M. A manager says "I can't tell if this is good or bad." What is the most important thing missing from this KPI display?

A
The standard deviation of revenue
B
A histogram showing revenue distribution
C
A target and/or comparison to prior period — without context, a number cannot be evaluated
D
The number of transactions that generated the revenue
A KPI without a benchmark is uninterpretable. $1.8M could be a triumph (target was $1.5M) or a disaster (target was $2.5M, and last month was $2.1M). Every KPI displayed on a dashboard must have context: vs target, vs prior period, vs benchmark, or vs forecast. This is one of the most common dashboard design failures in practice.