Finding Insights, Not Just Numbers

The Insight Gap

In the previous lesson, we explored visualizations that communicate. Now let’s build on that foundation. Here’s what most data analysis produces:

“Sales were $1.2 million in Q3.”

That’s not an insight. That’s a fact. It doesn’t tell you anything you can act on.

An insight would be:

“Q3 sales dropped 15% because we lost three enterprise accounts to a competitor who undercut our pricing. Without action, we’re likely to lose two more accounts next quarter.”

That’s actionable. It explains why, suggests what might happen, and implies what to do.

The Insight Hierarchy

Move up this ladder:

                           PRESCRIPTIVE
                          "We should focus on
                           retention offers"
                               ↑
                           PREDICTIVE
                        "We'll likely lose
                         2 more accounts"
                               ↑
                          DIAGNOSTIC
                      "We lost them to
                       pricing competition"
                               ↑
                         DESCRIPTIVE
                        "Sales dropped 15%"

Descriptive: What happened Diagnostic: Why it happened Predictive: What will likely happen Prescriptive: What we should do

Most analysis stops at descriptive. Push higher.

Techniques for Finding Insights

Technique 1: The “So What?” Chain

For every finding, ask “so what?” repeatedly.

Finding: Conversion rate dropped from 3.2% to 2.8% → So what? Implication: We’re getting 12% fewer customers from the same traffic → So what? Impact: At current traffic, that’s ~40 fewer customers per month → So what? Action: We need to diagnose the drop and fix it—or it’s $X in lost revenue

Technique 2: Look for Segments

Averages hide insights. Break data into segments.

Average view: “Customer satisfaction is 7.2/10” Segmented view:

• New customers: 8.1/10
• Customers 1-2 years: 7.5/10
• Customers 3+ years: 5.8/10

Now you have an insight: Long-term customers are unhappy. Why?

Technique 3: Compare to Expectations

Numbers need context. Compare to:

• Last period (month, quarter, year)
• Targets or goals
• Competitors or industry benchmarks
• Different segments

“Website traffic was 50,000 visitors” vs. “Website traffic was 50,000 visitors—20% below our target and 15% below last month despite increased ad spend”

Technique 4: Find the Exceptions

Look for what doesn’t fit the pattern.

• Which regions are growing while others decline?
• Which customers are highly satisfied when most aren’t?
• Which products have different margin profiles?

Exceptions often reveal opportunities or problems.

Technique 5: Follow the Driver Chain

Metrics don’t exist in isolation. They have drivers.

Revenue dropped → Why? Fewer orders → Why? Lower conversion rate → Why? Higher page abandonment → Why? Page load time increased → Found it.

Follow the chain until you find something actionable.

Quick check: Before moving on, can you recall the key concept we just covered? Try to explain it in your own words before continuing.

AI-Assisted Insight Generation

Use AI to help push toward insights:

Here's my analysis finding:

[Your finding, e.g., "Q3 sales were $1.2M, down 15% from Q2"]

Help me develop this into an insight by:
1. What questions should I ask to understand WHY this happened?
2. What additional data would help diagnose the cause?
3. What are possible explanations to investigate?
4. What might this imply for the future?
5. What actions might this suggest?

AI won’t know your business context, but it can help you think through the angles.

Validating Insights

Not every pattern is meaningful. Validate before reporting:

Is it statistically significant?

Small samples can show random variation. Make sure patterns are real.

AI: "Is a change from 3.2% to 2.8% conversion rate statistically significant given [X] total visitors? Or could this be random variation?"

Is there a plausible cause?

Correlation isn’t causation. Can you explain why this relationship exists?

Does it hold across segments?

A pattern that only exists in one small segment might not generalize.

Is it actionable?

An interesting finding that can’t be acted on isn’t useful for decision-making.

Structuring Insights

Present insights in a consistent structure:

WHAT WE FOUND: [The data finding]
WHY IT MATTERS: [Business impact]
WHAT'S CAUSING IT: [Diagnosis]
WHAT MIGHT HAPPEN: [Prediction if no action]
WHAT WE RECOMMEND: [Suggested action]

Example:

WHAT WE FOUND: Customer churn increased from 5% to 8% in Q3.

WHY IT MATTERS: This represents ~$150K in lost annual revenue and signals deeper satisfaction issues.

WHAT’S CAUSING IT: Exit surveys indicate price sensitivity—3 of 5 churned customers cited cost.

WHAT MIGHT HAPPEN: Without action, churn may continue rising as competitors maintain lower pricing.

WHAT WE RECOMMEND: Test a retention discount for at-risk accounts; evaluate competitive positioning.

Exercise: Transform Description to Insight

Take this descriptive finding and develop it into an insight:

“Website traffic increased 40% month-over-month.”

1. What questions would you ask to understand why?
2. What comparisons would give context?
3. What action might this inform?

Key Takeaways

• Data describes what happened; insights explain why it matters and what to do
• Move up the hierarchy: descriptive → diagnostic → predictive → prescriptive
• Use the “So what?” chain to push toward actionability
• Segment data—averages hide insights
• Compare to expectations—context creates meaning
• Follow driver chains to find root causes
• Validate insights: significance, causation, consistency, actionability
• Structure insights: finding → why it matters → cause → prediction → recommendation

Next: structuring your findings into reports for different audiences.

Up next: In the next lesson, we’ll dive into Reporting for Different Audiences.