Advanced Analytics: Segmentation, Cohort Analysis, Attribution

# Advanced Analytics: Segmentation, Cohort Analysis, Attribution

Basic analytics answers: "How much revenue?" Advanced analytics answers: "Which customers are worth most? When will they churn? Which marketing campaign worked?"

These questions drive business decisions worth millions.

Customer Segmentation

Divide customers into groups; tailor strategy for each.

**Segmentation approaches:**

**RFM (Recency, Frequency, Monetary):**

``` R: Days since last purchase (lower = better) F: Number of purchases (higher = better) M: Total spent (higher = better)

Score = weighted(R, F, M)

Segments:

Champions: High RFM (R ≤ 30, F ≥ 10, M ≥ $1000)

→ Strategy: Upsell, ask for referrals

At Risk: Low RFM (R > 180, F < 3, M < $500)

→ Strategy: Win-back campaigns, discounts

Dormant: Low RFM (R > 365)

→ Strategy: Reactivation or remove from list ```

**Behavioral segmentation:**

```

High-value browsers: Spend time on site, don't buy

→ Strategy: Site remarketing, improve UX

Deal seekers: Only buy on discount

→ Strategy: Exclusive discounts, loyalty program

Bulk buyers: Buy multiple units

→ Strategy: Volume discounts, wholesale program ```

**Psychographic segmentation:**

```

Price-sensitive

Quality-focused

Convenience-seeking

Environmentally conscious

```

Cohort Analysis

Track groups (cohorts) over time to understand lifecycle.

**Cohort = customers acquired in same time period**

``` Retention by acquisition month:

Month 0 Month 1 Month 2 Month 3 Jan 2026 100% 85% 72% 60% Feb 2026 100% 80% 65% Mar 2026 100% 82%

Insight: Jan cohort retains better (60% after 3 months) Why? Investigate product changes, marketing quality ```

**Cohort analysis reveals:**

Churn rate by cohort (which customers leak?)

Lifetime value trajectory (when do they peak?)

Seasonal patterns (summer users vs. winter users?)

Attribution Modeling

Which marketing touchpoint drove the sale?

**Challenge:** Customer typically touches multiple channels before buying.

``` Customer journey: Day 1: Google Ad → Click → Visit site Day 3: Email campaign → Click → Visit site Day 5: Facebook Ad → Click → Buy

Which channel gets credit?

Last-click attribution: Facebook gets 100% credit (last click)

Multi-touch attribution: Google: 30%, Email: 40%, Facebook: 30% (based on actual impact)

Result: Allocate marketing budget to highest-impact channels ```

**Attribution models:**

``` First-touch: Credit first channel (good for awareness campaigns) Last-touch: Credit last channel (easy to measure; often misleading) Linear: Equal credit to all channels Time-decay: More credit to recent channels (customers remember recent) Data-driven: ML model learns actual channel impact ```

Advanced Metrics

**Churn prediction:**

``` Build model: "Will customer churn in next 30 days?"

Features: - Days since last purchase - Decrease in purchase frequency - Support tickets (high tickets = churn risk) - Plan downgrade attempted - Competitor mention in support chat

Model output: - Customer 123: 85% churn risk → Proactive outreach - Customer 456: 10% churn risk → No action needed ```

**Lifetime value (LTV):**

``` Average revenue per customer × Average customer lifespan = $100/month × 36 months = $3600

LTV:CAC ratio = $3600 / $300 = 12 (Should be >3; higher is better)

Action: If LTV:CAC drops below threshold, review marketing spend ```

**Net Revenue Retention (NRR):**

``` (Revenue from existing customers + expansion) / Revenue from start of period

NRR > 100% = Expansion revenue (good) NRR < 100% = Churn exceeds growth (bad)

Example: Start: $1M revenue from 100 customers End: $900K (10% churn) + $200K expansion = $1.1M NRR: $1.1M / $1M = 110% ✓ (healthy) ```

Tools for Advanced Analytics

**SQL + Python:**

Manual but flexible

Good for ad-hoc analysis

**dbt + Python:**

Segment definitions as dbt models

Python for ML/cohort analysis

**Mixpanel, Amplitude:**

Behavioral analytics

Cohort analysis built-in

Good for mobile apps, SaaS

**Mode, Looker:**

SQL + visualization

Cohort templates available

Real-World Advanced Analytics Scenarios

Scenario 1: Churn Prediction

SaaS company: 10% monthly churn.

Without model: Reactive. Customer churns → try to win back (usually fails).

With churn prediction model:

Identify high-risk customers proactively

Reach out before they leave (much easier to retain)

Personalized retention offers

Result: Churn drops from 10% to 6% (40% improvement)

Revenue impact: $500K annually (saved)

Scenario 2: Cohort Analysis Reveals Quality Issue

Acquisition month: January (100 customers) vs. March (100 customers)

Cohort analysis: Jan: 60% retained after 3 months Mar: 75% retained after 3 months

Investigation: What changed between Jan and Mar?

Finding: Product quality improved; new feature released Feb 15.

Action: Backport fix to older versions; quality improves across customer base.

Scenario 3: Attribution Optimization

Company spends $1M/month on marketing across 5 channels.

Naive allocation: Equal spend ($200K each channel)

With attribution modeling: Channel A (Google): Drives 5 conversions per $100K spend (50 total) Channel B (Email): Drives 3 conversions per $100K spend (30 total) Channel C (Facebook): Drives 8 conversions per $100K spend (80 total)

Optimal allocation: Channel A: $250K (high efficiency) Channel B: $100K (low efficiency, but brand awareness valuable) Channel C: $400K (highest ROI) Other: $250K (test/explore)

Result: Same $1M spend, 20% more conversions

The Bottom Line

Basic analytics describes "what happened." Advanced analytics explains "why" and predicts "what will happen next."

These insights drive millions in business impact.

Start with cohorts and segmentation. Build from there.