Diagnostic Analytics: How to Find Out Why Something Happened
Diagnostic analytics is the process of examining data to understand why something occurred. While descriptive analytics tells you "revenue dropped 15% last month," diagnostic analytics tells you "revenue dropped because enterprise deal closures stalled due to budget freezes in the financial services sector, which accounts for 40% of pipeline."
Where Diagnostic Analytics Fits
| Analytics Type | Question | Output |
|---|---|---|
| Descriptive | What happened? | Reports, dashboards, KPIs |
| Diagnostic | Why did it happen? | Root cause analysis, explanations |
| Predictive | What will happen? | Forecasts, probabilities |
| Prescriptive | What should we do? | Recommendations, optimization |
Most organizations spend 80% of analytics effort on descriptive (dashboards, reports) and skip diagnostic entirely. The result: teams know things are bad but do not know why, making it impossible to fix.
Diagnostic Analytics Techniques
1. Drill-Down Analysis
Start at the aggregate level and decompose into finer granularity until the cause is visible.
Example: Revenue is down 15% month-over-month.
Level 1: Revenue by region
- Northeast: flat
- Southeast: flat
- West: down 35%
- Midwest: up 5%
Level 2: West region by product
- Product A: flat
- Product B: down 60%
- Product C: up 10%
Level 3: Product B in West by sales rep
- Rep 1: flat
- Rep 2: $0 (on leave, not reassigned)
- Rep 3: down 40%
Level 4: Rep 3's pipeline
- 5 deals pushed to next quarter due to client budget freezes
Root cause found: Two compounding factors: a rep on leave whose accounts were not covered, plus budget freezes affecting one rep's pipeline.
2. Correlation Analysis
Identify variables that move together to narrow the potential causes.
Example: Support ticket volume increased 40% this month. What correlates?
| Potential Cause | Correlation with Ticket Volume | Conclusion |
|---|---|---|
| New user signups | r = 0.3 (weak) | Not the driver |
| Bug fix deployment | r = 0.1 (none) | Not related |
| Feature launch date | r = 0.85 (strong) | Likely related |
| Documentation changes | r = 0.2 (weak) | Not the driver |
The feature launch strongly correlates with ticket increases. Drill deeper: which feature? Which ticket categories?
3. Segmentation Analysis
Compare the behavior of different segments to isolate where the change occurred.
Example: Overall conversion rate dropped from 4.2% to 3.5%.
| Segment | Before | After | Change |
|---|---|---|---|
| Direct traffic | 5.1% | 5.0% | -0.1% (stable) |
| Organic search | 4.0% | 3.8% | -0.2% (slight) |
| Paid search | 3.8% | 1.9% | -1.9% (crashed) |
| Social | 2.5% | 2.4% | -0.1% (stable) |
| 6.2% | 6.0% | -0.2% (stable) |
Diagnosis: The problem is isolated to paid search. Investigate: ad targeting changes? Landing page modification? Competitor bidding? Quality score drop?
4. Cohort Comparison
Compare groups that should behave similarly but do not.
Example: Customer retention is declining. But is it all customers or specific cohorts?
| Sign-up Month | 30-day Retention | 90-day Retention |
|---|---|---|
| January | 78% | 62% |
| February | 76% | 60% |
| March | 72% | 55% |
| April | 65% | 48% |
Retention is declining for recent cohorts specifically. What changed in March/April?
- New pricing introduced in March
- Onboarding flow redesigned in April
- Marketing shifted to lower-intent channels in March
5. Root Cause Analysis (5 Whys)
A structured approach to dig past symptoms to root causes:
Problem: Monthly churn increased from 3% to 5%.
- Why? Because 47 accounts cancelled (vs. usual 28).
- Why did they cancel? Exit surveys show "not getting value" (60%) and "too expensive" (25%).
- Why are they not getting value? Usage data shows these accounts never completed onboarding.
- Why did they not complete onboarding? Onboarding requires data connection, and our new connector had a bug that affected 40% of new accounts.
- Why was the bug not caught? QA tested with existing accounts, not new signups with fresh connections.
Root cause: A testing gap in QA for the new connector, not a product-market fit issue or pricing problem.
6. Contribution Analysis
Quantify how much each factor contributed to the overall change.
Example: Revenue grew $500K this quarter. What drove it?
| Factor | Contribution | Percentage |
|---|---|---|
| New customer acquisition | +$280K | 56% |
| Existing customer expansion | +$150K | 30% |
| Price increase | +$120K | 24% |
| Churn (offset) | -$50K | -10% |
| Net | +$500K | 100% |
Now you know: growth is primarily acquisition-driven, with healthy expansion. Price increases helped but churn partially offsets them.
Building a Diagnostic Analytics Capability
Make Drill-Down Easy
The biggest barrier to diagnostic analytics is friction. If drilling from "revenue is down" to "why" requires a new Jira ticket to the data team, diagnosis happens too slowly.
Solutions:
- Interactive dashboards with drill-down capability
- Self-service tools where business users can filter and slice
- AI-powered analytics (platforms like Skopx let users ask "why did revenue drop?" in natural language and get automated root cause analysis)
Pre-Build Common Diagnostic Paths
For metrics that fluctuate regularly, pre-build the diagnostic drill-downs:
- Revenue down? Auto-show by region, product, customer segment, sales rep
- Churn up? Auto-show by cohort, plan, usage level, feature adoption
- Conversion down? Auto-show by channel, device, landing page, user type
- Costs up? Auto-show by category, vendor, department, project
Establish Anomaly Detection
Do not wait for someone to notice a problem. Automated anomaly detection flags when metrics deviate from expected ranges and provides initial segmentation:
"Alert: Conversion rate dropped 25% in the last 24 hours. Primary driver: mobile traffic from paid social (conversion dropped 80%). Desktop and organic are normal."
Document Root Causes
When you find a root cause, document it. Over time, this creates a knowledge base:
- "Revenue dips in January: seasonal budget resets (not a real problem)"
- "Spike in support tickets after deploys: usually new feature confusion (resolve with in-app guidance)"
- "Churn spikes correlate with renewal anniversary of pricing change"
This institutional memory prevents re-investigating the same causes repeatedly.
Common Mistakes
- Stopping at the first plausible explanation. The first thing that correlates is not always the cause. Look for multiple confirming signals.
- Confusing correlation with causation. Two metrics moving together does not mean one causes the other.
- Anchoring on recent events. Teams naturally blame the most recent change. But sometimes the cause predates the symptom (delayed effects).
- Not quantifying the contribution. "It was the pricing change" is incomplete without "the pricing change accounts for 60% of the churn increase."
- Diagnosing noise. Small fluctuations in metrics are often random variation, not signals requiring diagnosis. Establish what constitutes a meaningful deviation before investigating.
Summary
Diagnostic analytics bridges the gap between knowing what happened and knowing what to do about it. Start with drill-down analysis to narrow the scope, use correlation and segmentation to identify suspects, confirm with cohort comparison and contribution analysis, and document findings for future reference. The goal is speed: the faster you can diagnose a problem, the faster you can fix it.
Saad Selim
The Skopx engineering and product team