Communicating EDA Findings

The most brilliant analysis is worthless if it cannot be communicated. This page covers how to transform raw EDA outputs into compelling, audience-appropriate reports and presentations.

---

Communication Framework

---

Report Structure

The Pyramid Principle

Start with the conclusion, then support with evidence. Do not make the reader wait.

report_structure = """
EDA REPORT: Customer Churn Analysis
=====================================

EXECUTIVE SUMMARY (read this only if short on time)
----------------------------------------------------
- 18% of customers churned last quarter, up from 14%
- Top 3 predictors: contract type, tenure, monthly charges
- Recommended action: target month-to-month customers with
  retention offers before month 3

KEY FINDINGS (5-7 findings with evidence)
----------------------------------------------------
1. Finding: [Statement]
   Evidence: [Chart + number]
   Implication: [So what?]

2. Finding: ...

DATA QUALITY NOTES
----------------------------------------------------
- 3% missing in income (imputed with median)
- 200 duplicate records removed
- Date range: 2023-01-01 to 2024-12-31

METHODOLOGY
----------------------------------------------------
- Tools: Python 3.11, pandas 2.2, scikit-learn 1.4
- Statistical tests: Mann-Whitney U, Chi-squared
- Significance level: alpha = 0.05

APPENDIX
----------------------------------------------------
- Detailed tables
- Additional charts
- Code reference
"""
print(report_structure)

---

Chart Simplification

Before and After

import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

np.random.seed(42)
categories = ['Electronics', 'Clothing', 'Home', 'Books', 'Sports']
values = [42, 38, 55, 28, 47]

# BAD: Cluttered, default styling
fig, axes = plt.subplots(1, 2, figsize=(16, 6))

axes[0].bar(categories, values, color=['red', 'blue', 'green', 'orange', 'purple'],
            edgecolor='black', linewidth=2)
axes[0].set_title('Sales by Category', fontsize=8)
axes[0].grid(True)
axes[0].set_ylabel('Units')
axes[0].legend(['Sales'])  # unnecessary legend
for i, v in enumerate(values):
    axes[0].text(i, v + 0.5, str(v), ha='center', fontsize=7)
axes[0].set_ylim(0, 70)
# Title: "BEFORE — Too many colors, unnecessary legend, small font"
axes[0].set_title('BEFORE: Default/cluttered', fontweight='bold')

# GOOD: Clean, focused, tells a story
sorted_idx = np.argsort(values)[::-1]
sorted_cats = [categories[i] for i in sorted_idx]
sorted_vals = [values[i] for i in sorted_idx]
colors = ['#2563eb' if v == max(values) else '#94a3b8' for v in sorted_vals]

axes[1].barh(sorted_cats, sorted_vals, color=colors, height=0.6)
axes[1].set_title('Home leads category sales', fontsize=14, fontweight='bold', loc='left')
axes[1].spines['top'].set_visible(False)
axes[1].spines['right'].set_visible(False)
axes[1].spines['bottom'].set_visible(False)
axes[1].xaxis.set_visible(False)
axes[1].invert_yaxis()
for i, v in enumerate(sorted_vals):
    axes[1].text(v + 0.5, i, f'{v} units', va='center', fontsize=11,
                  fontweight='bold' if v == max(values) else 'normal')

plt.tight_layout()
plt.show()

Chart Simplification Rules

Rule	Before	After
Remove chartjunk	Grid lines, borders, legends	Only essential elements
Highlight the insight	All bars same color	Hero bar is colored, rest gray
Use descriptive titles	"Revenue by Quarter"	"Q3 revenue dropped 15% after pricing change"
Sort meaningfully	Alphabetical	By value (largest first)
Remove redundancy	Legend + axis labels + data labels	Pick one or two
Choose right orientation	Vertical bars for 10 categories	Horizontal bars for readability

---

Audience-Aware Communication

For Executives

def executive_summary(df, target_col, key_metrics):
    """Generate an executive-friendly summary."""
    summary = []
    summary.append("KEY METRICS")
    summary.append("-" * 40)

    for name, value, change in key_metrics:
        direction = "up" if change > 0 else "down"
        summary.append(f"  {name}: {value} ({direction} {abs(change):.1f}% vs prior period)")

    # Top insight
    summary.append("")
    summary.append("TOP INSIGHT")
    summary.append("-" * 40)

    if target_col in df.columns:
        rate = df[target_col].mean()
        summary.append(f"  {target_col} rate: {rate:.1%}")

    summary.append("")
    summary.append("RECOMMENDED ACTIONS")
    summary.append("-" * 40)
    summary.append("  1. [Specific action based on findings]")
    summary.append("  2. [Second action]")
    summary.append("  3. [Third action]")

    return '\n'.join(summary)

# Usage
metrics = [
    ("Churn Rate", "18.2%", 4.2),
    ("Avg Revenue/Customer", "$142", -3.1),
    ("Customer Satisfaction", "4.2/5", 0.1),
]
# print(executive_summary(df, 'churned', metrics))

For Technical Audience

def technical_report_section(finding_name, description, statistical_test,
                              effect_size, visualization_fn, caveats):
    """Generate a technical EDA finding section."""
    report = f"""
### {finding_name}

**Description**: {description}

**Statistical Evidence**:
- Test: {statistical_test['name']}
- Statistic: {statistical_test['statistic']:.4f}
- p-value: {statistical_test['p_value']:.4f}
- Effect size: {effect_size['name']} = {effect_size['value']:.4f} ({effect_size['interpretation']})

**Caveats**:
"""
    for caveat in caveats:
        report += f"- {caveat}\n"

    return report

# Example
finding = technical_report_section(
    "Churn rate differs by contract type",
    "Month-to-month customers churn at 3x the rate of annual contract customers",
    {'name': 'Chi-squared', 'statistic': 245.8, 'p_value': 0.0001},
    {'name': "Cramer's V", 'value': 0.42, 'interpretation': 'medium-large'},
    None,
    ['Self-selection bias: customers who intend to stay choose annual contracts',
     'Analysis period covers only 12 months'],
)
print(finding)

---

Data Storytelling

The SCR Framework (Situation-Complication-Resolution)

story_template = """
SITUATION:
  Our customer base grew 25% last year to 50,000 active customers.
  Revenue per customer averaged $142/month.

COMPLICATION:
  However, churn increased from 14% to 18% — we are losing customers
  faster than we are acquiring them. At this rate, net growth turns
  negative in Q3.

RESOLUTION:
  EDA reveals that 72% of churn comes from month-to-month contracts
  in their first 3 months. A targeted onboarding program for this
  segment could reduce churn by an estimated 4 percentage points,
  saving $2.1M annually.
"""

Annotation-Driven Charts

# Charts should tell stories through annotations
np.random.seed(42)
months = pd.date_range('2023-01-01', periods=24, freq='M')
churn_rates = np.array([14, 13.5, 14.2, 14.8, 15, 14.5, 15.2, 15.8,
                         16, 16.5, 17, 17.2, 17.5, 18, 18.2, 17.8,
                         17.5, 17, 16.5, 16, 15.5, 15, 14.8, 14.5])

fig, ax = plt.subplots(figsize=(14, 6))
ax.plot(months, churn_rates, 'o-', color='#dc2626', linewidth=2, markersize=5)
ax.fill_between(months, churn_rates, alpha=0.1, color='#dc2626')

# Annotate key events
ax.annotate('Price increase\nimplemented',
            xy=(months[4], churn_rates[4]),
            xytext=(months[6], churn_rates[4] + 2),
            arrowprops=dict(arrowstyle='->', color='black'),
            fontsize=10, ha='center',
            bbox=dict(boxstyle='round,pad=0.3', facecolor='yellow', alpha=0.8))

ax.annotate('Peak churn:\n18.2%',
            xy=(months[14], churn_rates[14]),
            xytext=(months[16], churn_rates[14] + 1.5),
            arrowprops=dict(arrowstyle='->', color='black'),
            fontsize=10, ha='center', fontweight='bold',
            bbox=dict(boxstyle='round,pad=0.3', facecolor='#fecaca', alpha=0.8))

ax.annotate('Retention program\nlaunched',
            xy=(months[15], churn_rates[15]),
            xytext=(months[13], churn_rates[15] - 2.5),
            arrowprops=dict(arrowstyle='->', color='green'),
            fontsize=10, ha='center',
            bbox=dict(boxstyle='round,pad=0.3', facecolor='#bbf7d0', alpha=0.8))

# Clean styling
ax.set_title('Monthly Churn Rate: Price increase drove churn up; retention program brought it back down',
             fontsize=12, fontweight='bold', loc='left')
ax.set_ylabel('Churn Rate (%)')
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.grid(axis='y', alpha=0.3)
ax.set_ylim(12, 22)

plt.tight_layout()
plt.show()

---

Common Communication Mistakes

Mistake	Problem	Fix
Showing all 50 charts	Information overload	Pick 5-7 that tell the story
Technical jargon	"p-value of 0.003" means nothing to executives	"Statistically significant" or "very likely real"
No "so what"	Charts without interpretation	Every chart needs a takeaway sentence
Correlation as causation	"Higher income causes lower churn"	"Higher income is associated with lower churn"
Ignoring uncertainty	"Churn rate is 18.2%"	"Churn rate is 18.2% (95% CI: 17.1%-19.3%)"
Missing context	"Revenue is $5M"	"Revenue is $5M, up 12% YoY, but below $5.5M target"

---

EDA Presentation Template

def generate_eda_presentation_outline(dataset_name, n_findings=5):
    """Generate a presentation outline for EDA findings."""
    outline = f"""
EDA PRESENTATION: {dataset_name}
{'=' * 60}

SLIDE 1: TITLE
  - Dataset name, date range, analyst
  - One-sentence summary of the most important finding

SLIDE 2: CONTEXT
  - Why this analysis was done
  - What question we are answering
  - Data source and time period

SLIDE 3: DATA OVERVIEW
  - Size: rows x columns
  - Key variables
  - Data quality summary (missing %, duplicates)

SLIDES 4-{3+n_findings}: FINDINGS (one per slide)
  - Insight title (not "Chart 1" but "Churn peaks in month 2")
  - ONE chart that supports the insight
  - Key number highlighted
  - Business implication

SLIDE {4+n_findings}: LIMITATIONS & CAVEATS
  - What the data cannot tell us
  - Missing data impact
  - Known biases

SLIDE {5+n_findings}: RECOMMENDATIONS
  - 3-5 specific, actionable recommendations
  - Each tied to a finding
  - Estimated impact if possible

SLIDE {6+n_findings}: NEXT STEPS
  - Additional analysis needed
  - Data collection suggestions
  - Timeline
"""
    return outline

print(generate_eda_presentation_outline("Customer Churn"))

---

Export-Ready Visualization Settings

# Publication-quality figure settings
def setup_publication_style():
    """Configure matplotlib for clean, professional charts."""
    plt.rcParams.update({
        # Figure
        'figure.figsize': (10, 6),
        'figure.dpi': 100,
        'savefig.dpi': 300,
        'savefig.bbox': 'tight',
        'savefig.transparent': False,
        'savefig.facecolor': 'white',

        # Font
        'font.family': 'sans-serif',
        'font.sans-serif': ['Arial', 'Helvetica', 'DejaVu Sans'],
        'font.size': 12,
        'axes.titlesize': 14,
        'axes.titleweight': 'bold',
        'axes.labelsize': 12,

        # Axes
        'axes.spines.top': False,
        'axes.spines.right': False,
        'axes.grid': True,
        'grid.alpha': 0.3,

        # Legend
        'legend.fontsize': 10,
        'legend.framealpha': 0.9,

        # Lines
        'lines.linewidth': 2,
        'lines.markersize': 6,
    })

setup_publication_style()

# Save in multiple formats
def save_figure(fig, name, formats=['png', 'svg']):
    """Save figure in multiple formats for different uses."""
    for fmt in formats:
        path = f'reports/figures/{name}.{fmt}'
        fig.savefig(path, format=fmt, bbox_inches='tight', facecolor='white')
        print(f"Saved: {path}")

---

Writing Insight Statements

# Template for converting analysis into insight statements

insight_templates = {
    'comparison': "{group_a} has {metric} that is {difference} {direction} than {group_b} ({value_a} vs {value_b})",
    'trend': "{metric} has {direction} by {amount} over {period}, from {start} to {end}",
    'distribution': "{percent}% of {entity} have {metric} {condition} ({threshold})",
    'correlation': "{variable_a} and {variable_b} are {strength} {direction}ly correlated (r={correlation})",
    'anomaly': "{entity} shows an unusual {pattern} in {metric}, suggesting {hypothesis}",
}

# Examples:
examples = [
    insight_templates['comparison'].format(
        group_a='Month-to-month customers', metric='churn rate',
        difference='3x', direction='higher', group_b='annual contract customers',
        value_a='42%', value_b='14%'
    ),
    insight_templates['trend'].format(
        metric='Average revenue per user', direction='declined',
        amount='8%', period='the past 6 months',
        start='$156', end='$143'
    ),
    insight_templates['distribution'].format(
        percent=72, entity='churned customers', metric='tenure',
        condition='less than', threshold='6 months'
    ),
]

print("Example Insight Statements:")
for i, ex in enumerate(examples, 1):
    print(f"  {i}. {ex}")

---

Key Takeaways

• Start with the conclusion (Pyramid Principle) — do not make the reader wait through methodology to get the answer
• Every chart needs a "so what" — the title should state the insight, not describe the chart type
• Simplify ruthlessly: fewer colors, fewer grid lines, fewer charts; highlight only what matters
• Know your audience: executives want actions, data teams want methodology, business users want filters
• Annotate charts with context (events, benchmarks, targets) that the reader needs to interpret them
• Use the SCR framework (Situation-Complication-Resolution) for narrative structure
• Never present correlation as causation — use language like "associated with" not "causes"
• Always include caveats and limitations — they build credibility, not doubt