Customer Health & Retention Analysis — 6,500 Accounts
Healthy All 14 columns have appropriate data types. Numeric metrics (MRR, health scores, engagement) are correctly typed as float64/int64. Categorical dimensions (plan, segment, acquisition channel) are properly typed as objects.
Monitor MRR: 334 accounts (5.14%) exceed the IQR-based outlier threshold, driven by high-value Enterprise contracts. The distribution is highly right-skewed (skewness = 2.52), with a long tail extending to $6,564/month. This reflects natural revenue concentration rather than data quality issues.
Monitor Support tickets: 387 accounts (5.95%) show unusually high ticket volumes, with the distribution heavily right-skewed (skewness = 2.04). Maximum observed value is 39 tickets in 90 days. This minority of high-support accounts likely represents struggling customers and should be prioritized for intervention.
Monitor Unresolved tickets: 415 accounts (6.38%) have elevated unresolved ticket counts. 68.4% of accounts have zero unresolved tickets, but the tail extends to 15 unresolved issues. This extreme right skew (skewness = 3.35) indicates a subset of customers experiencing severe support response gaps.
Healthy Health score: 220 outliers (3.38%) represent accounts with unusually low health scores. The distribution is left-skewed (skewness = -1.14), clustering around 82.7 (median), with a concerning tail of struggling accounts below 60.
The left-skewed distribution (skewness = -1.14) indicates most accounts are healthy (clustering near 83), but a meaningful tail of struggling accounts pulls the mean down. This shape is operationally significant: the business has a baseline healthy customer base, but a visible minority requires intervention. The 3.38% outlier rate below the lower fence represents accounts with severely depressed health scores that require priority intervention.
Nearly symmetric distribution (skewness = -0.20) with wide spread. The 17-point standard deviation means onboarding quality varies dramatically across accounts. Approximately 18% of accounts score below 50, representing a cohort that never achieved strong initial activation. The correlation with health score (r = 0.43) confirms that onboarding deficits compound over time.
Approximately symmetric distribution centered on 15 days per month, indicating moderate engagement as the norm. The 6.8-day standard deviation reveals substantial variation: some accounts log in nearly daily, others barely weekly. Only 0.49% outliers, suggesting the engagement range is natural rather than anomalous.
Highly right-skewed (skewness = 2.04), with median well below mean. Most accounts (60%+) submit 0-3 tickets, but a significant minority generates heavy support load. The 5.95% outlier rate captures accounts submitting 10+ tickets in 90 days, signaling either product friction or complex use cases.
Extreme right skew (skewness = 3.35). 68.4% of accounts have zero unresolved tickets, indicating baseline support resolution is functional. However, the tail matters: 6.38% of accounts have 3+ unresolved tickets, representing sustained support failures. The nonzero values present are: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, and 15.
Slightly right-skewed (skewness = 0.34), with most accounts achieving value in 12-25 days. The 9.7-day standard deviation indicates variability in activation speed. Churned accounts took 24.6 days on average vs. 18.5 for retained accounts (Cohen's d = 0.64, medium effect), confirming that delayed value realization predicts attrition.
Highly right-skewed (skewness = 2.52), with median significantly below mean due to high-value outliers. The distribution reflects a typical SaaS customer base: many small-to-mid accounts, with Enterprise contracts driving the tail to $6,564/month. The 5.14% outlier rate represents natural revenue concentration rather than data issues.
Approximately symmetric (skewness = 0.11), indicating balanced customer tenure. The range spans 1 to 40 months, with only 0.17% outliers. Churned vs. retained accounts show minimal tenure difference (15.2 vs. 15.6 months, Cohen's d = -0.06), suggesting churn is not primarily a maturity or "honeymoon period" phenomenon.
Approximately symmetric ratio field representing account growth or contraction. Mean of 0.08 indicates slight average expansion. Churned accounts show 0.06 vs. 0.08 for retained (Cohen's d = -0.11, small effect), indicating expansion rate has minimal predictive power for 90-day churn.
Growth plan dominates at 37.34% (2,427 accounts), followed by Scale (25.14%), Starter (23.42%), and Enterprise (14.11%). The distribution shows moderate concentration with entropy = 1.92. No plan type is dangerously underrepresented. However, churn rate analysis reveals that Starter plan carries the highest risk at 14.72%, while Enterprise shows the lowest at 9.16%.
Ecommerce leads at 33.28% (2,163 accounts), followed by B2B SaaS (24.45%), Marketplace (15.42%), Agency (15.20%), and Creator Tools (11.66%). Entropy = 2.22 indicates reasonable diversity. Creator Tools, despite being the smallest segment, shows the highest churn rate at 15.04%, indicating product-market fit challenges or service gaps for this vertical.
Nearly balanced across five channels with entropy = 2.32 (very high diversity). Organic leads narrowly at 21.23%, followed by Outbound (20.55%), Marketplace (19.78%), Partner (19.48%), and Paid Search (18.95%). This balance is operationally healthy, avoiding over-reliance on any single channel. However, Paid Search shows the highest churn rate at 13.8%, raising CAC efficiency concerns.
Unique identifier with 6,500 distinct values (100% cardinality). No duplication, confirming one row per account. This field is excluded from analysis per user instructions.
Accounts with high total ticket volume accumulate unresolved issues at a near-linear rate. This is not merely a volume effect — it indicates a capacity constraint in support operations. As ticket load increases, resolution speed degrades, creating a backlog that correlates with churn. The operational implication: scaling support capacity or triaging high-volume accounts should reduce both unresolved tickets and downstream churn.
Strong onboarding scores translate to higher health scores months later. This 0.43 correlation is mechanically significant because onboarding is a one-time intervention that produces lasting effects. Accounts scoring below 50 on onboarding rarely recover to above-average health. The business should treat onboarding score as a leading indicator and trigger proactive outreach for low-scoring cohorts within the first 30 days.
Both total support tickets and unresolved tickets correlate negatively with health score at nearly identical magnitudes. This dual correlation reveals two distinct pathways: high ticket volume signals product friction or complex use cases, while unresolved tickets signal support system failure. Both pathways damage customer health. The equal magnitudes suggest both dimensions require attention.
Health score shows the strongest direct correlation with 90-day churn among all metrics. This confirms it is functioning as intended: a composite signal aggregating multiple risk factors. However, the correlation magnitude (-0.38) is only moderate, indicating health score does not fully capture all churn drivers. Feature engineering to incorporate support experience more heavily could strengthen predictive power.
Accounts with poor onboarding churn at higher rates even months later. The -0.28 correlation is weaker than health score but still meaningful. Combined with the 0.43 correlation between onboarding and health, the causal chain is clear: poor onboarding → persistently low health → eventual churn. Intervening at the onboarding stage offers the earliest prevention opportunity.
Accounts that took longer to reach initial value show persistently lower health scores. This correlation suggests activation speed is not just a vanishing metric — it sets the baseline trajectory. Churned accounts took 24.6 days vs. 18.5 for retained (6.1-day gap, Cohen's d = 0.64). Reducing time-to-value below 15 days should become a retention priority.
Both ticket volume and unresolved tickets show positive correlations with churn, indicating support struggles predict attrition. The similar magnitudes (0.27 and 0.26) suggest both dimensions matter roughly equally. This is actionable: improving resolution speed (reducing unresolved count) may be as effective as reducing total ticket volume.
Higher monthly login days correlate with better health, but the effect size for churn prediction is small (Cohen's d = -0.40). Churned accounts averaged 12.2 login days vs. 14.9 for retained. While engagement matters, it is a weaker signal than onboarding, support, or health score. Login frequency is a symptom, not a root cause.
Starter plan customers churn at 14.72%, 2.6 percentage points above the company average of 12.12% and 5.56pp above the best-performing Enterprise plan (9.16%). In relative terms, Starter churn is 60.7% higher than Enterprise. This 224-account churn cohort from Starter suggests pricing or packaging misalignment: customers on the entry tier may not be receiving sufficient value to justify renewal, or self-serve onboarding may be inadequate for this segment.
Growth plan churn rate is 12.4%, only 0.28pp above the company average, indicating this mid-tier plan performs at baseline. With 301 churned accounts, Growth represents the largest absolute churn volume due to its 37.34% share of the customer base. However, its near-average rate suggests no structural issue with the plan itself.
Scale plan churns at 11.0%, 1.12pp below the company average of 12.12%. With 179 churned accounts from a base of 1,634 (25.14% of customers), Scale sits between Growth and Enterprise, indicating that mid-to-upper tier customers retain more reliably than entry and mid tiers, though not at Enterprise levels.
Enterprise plan shows the lowest churn at 9.16%, 2.96pp below the company average. Only 84 Enterprise accounts churned, representing strong retention in the high-value segment. This likely reflects higher switching costs, more hands-on customer success support, and better product-market fit for complex use cases.
Creator Tools segment has the highest churn risk at 15.04%, 2.92pp above the company average and 4.03pp above the best-performing B2B SaaS segment (11.01%). With 114 churned accounts from a base of 758, Creator Tools shows both rate and count concentration of risk. This suggests the product may not fully address Creator Tools workflows, or that this vertical has higher inherent churn due to business volatility.
Agency segment churns at 12.45%, only 0.33pp above average, indicating baseline performance. With 123 churned accounts, Agency represents a meaningful volume, but no unusual rate-based risk.
Marketplace segment churns at 11.98%, 0.14pp below average. This near-average performance from 1,002 accounts (120 churned) indicates the product serves this vertical adequately.
Ecommerce, the largest segment at 33.28%, churns at 11.84%, 0.28pp below average. Despite contributing 256 churned accounts (the highest absolute count), Ecommerce shows no rate-based risk. The volume is proportional to its size.
B2B SaaS shows the lowest churn at 11.01%, 1.11pp below average and serving as the baseline for comparison. With 175 churned accounts from 1,589 total, B2B SaaS demonstrates the strongest retention among all verticals.
Paid Search customers churn at 13.8%, 1.68pp above the company average and 3.47pp above the best-performing channel (Outbound at 10.33%). This represents a 33.6% relative increase vs. Outbound. With 170 churned accounts from 1,232 total, Paid Search shows both rate and volume risk. This pattern raises CAC payback concerns: if Paid Search acquisition costs are comparable to other channels, the higher churn rate severely damages unit economics.
Organic customers churn at 12.46%, 0.34pp above average and 2.13pp above Outbound. With 172 churned accounts, Organic contributes meaningful churn volume but shows only modest rate elevation.
Partner channel churns at 12.24%, 0.12pp above average, indicating near-baseline performance. The 155 churned accounts from 1,266 total suggest no structural channel issue.
Marketplace channel churns at 11.9%, 0.22pp below average, performing slightly better than baseline with 153 churned accounts from 1,286 total.
Outbound channel shows the lowest churn at 10.33%, 1.79pp below average. With 138 churned accounts from 1,336 total, Outbound demonstrates the strongest retention among all acquisition sources. This likely reflects higher-intent buyers and better pre-qualification during the sales process.
The cross-tabulation of acquisition channel and vertical segment reveals interaction effects that single-dimension analysis misses. Paid Search × Creator Tools shows the highest churn rate at 20.0%, nearly double the company average and 7.88pp above baseline. This 135-account segment (27 churned) represents the single highest-risk profile in the dataset. The combination suggests either poor keyword targeting, mismatched landing page messaging, or fundamental product gaps for paid-acquired Creator Tools customers.
Organic × Agency churns at 17.86%, 5.74pp above average, indicating that organically-acquired agencies struggle more than those from other channels. With 35 churned accounts from 196 total, this segment warrants investigation into why organic discovery does not translate to retention for this vertical.
Marketplace × Creator Tools churns at 16.79%, 4.67pp above average, showing that Creator Tools struggles persist across multiple acquisition channels, reinforcing the vertical-level product fit concern.
Partner × Creator Tools (14.6%), Paid Search × Ecommerce (14.59%), and Organic × Marketplace (13.9%) round out the top elevated-risk combinations, all exceeding 1.5pp above the company average.
Effect sizes (Cohen's d) distinguish statistically significant patterns from operationally meaningful ones. Large effect sizes indicate features that genuinely separate churners from retained customers and should drive intervention design.
Churned accounts had a mean health score of 69.5 vs. 82.9 for retained accounts, a 13.4-point gap. This large negative effect size confirms health score is the single most powerful churn signal in the dataset. An account scoring below 70 is in high-risk territory and should trigger immediate intervention. This metric is already functioning as designed, but thresholds for automated alerts should be set around the 70-point boundary.
Churned accounts scored 55.8 on onboarding vs. 70.6 for retained customers, a 14.8-point gap. This large effect size indicates onboarding quality sets the long-term trajectory. Accounts scoring below 60 in the first 30 days are at severe risk and require proactive outreach. The effect size magnitude justifies building a dedicated intervention playbook for low-onboarding cohorts.
Churned accounts submitted an average of 6.9 support tickets in 90 days vs. 3.4 for retained accounts, a 3.5-ticket gap. This large positive effect size confirms that high support volume is not merely engagement — it is a distress signal. Accounts exceeding 6 tickets in 90 days should be flagged for CSM intervention, not just support queue management.
Churned accounts had 1.4 unresolved tickets on average vs. 0.5 for retained accounts, a 0.9-ticket gap. This large effect size indicates that leaving tickets unresolved is not a minor operational inefficiency — it is a direct churn driver. Any account with 2+ unresolved tickets should receive escalated attention. The magnitude is nearly as large as total ticket volume, suggesting resolution speed matters as much as reducing overall support load.
Churned accounts took 24.6 days to reach value vs. 18.5 for retained accounts, a 6.1-day gap. This medium effect size confirms that slow activation is a meaningful churn risk factor. Accounts not reaching value within 20 days should trigger proactive onboarding support. While the effect is smaller than support or health metrics, it represents an early intervention point before other issues compound.
Churned accounts logged in 12.2 days per month vs. 14.9 for retained accounts, a 2.7-day gap. The small effect size indicates login frequency is a secondary signal, not a root cause. Driving login frequency without addressing underlying satisfaction or product fit is unlikely to prevent churn. This metric is better used as a monitoring indicator than an intervention target.
Tenure (15.2 vs. 15.6 months), account value ($628 vs. $659 MRR), and expansion rate (0.06 vs. 0.08) show negligible differences between churned and retained cohorts. Effect sizes are all small, indicating these dimensions do not meaningfully separate churn risk. Churn is not primarily driven by customer maturity, contract size, or recent expansion behavior. Interventions should focus on experience metrics (health, onboarding, support) rather than demographic or revenue attributes.
Evidence: Churned accounts averaged a health score of 69.5 vs. 82.9 for retained customers (Cohen's d = -1.24, the largest effect in the dataset). Accounts below 70 are in high-risk territory.
High Impact Easy
First Step: Configure automated alerts to trigger CSM outreach when any account's health score drops below 70. Prioritize accounts that also have unresolved tickets or low onboarding scores for immediate intervention.
Evidence: Churned accounts had 1.4 unresolved tickets vs. 0.5 for retained accounts (Cohen's d = 0.83). Unresolved tickets correlate with poor health at r = -0.40. Currently, 6.38% of accounts have 3+ unresolved tickets.
High Impact Medium
First Step: Establish a dedicated escalation queue for any account with 2+ unresolved tickets and health score below 75. Assign a single owner responsible for resolution within 48 hours.
Evidence: Churned accounts scored 55.8 on onboarding vs. 70.6 for retained accounts (Cohen's d = -0.89). Onboarding score correlates with long-term health at r = 0.43, indicating early deficits compound over time.
High Impact Easy
First Step: Within 30 days of signup, identify all accounts scoring below 60 on onboarding and trigger a structured CSM call focused on activation barriers and use case alignment.
Evidence: Creator Tools segment churns at 15.04%, 2.92pp above average and 4.03pp above B2B SaaS. The segment shows elevated churn across multiple acquisition channels, with Paid Search × Creator Tools reaching 20.0% churn rate.
High Impact Hard
First Step: Conduct 10 exit interviews with recently churned Creator Tools accounts to identify feature gaps or workflow mismatches. Use findings to prioritize product roadmap adjustments for this vertical.
Evidence: Paid Search customers churn at 13.8%, 3.47pp above Outbound (the best-performing channel). This represents a 33.6% relative increase in churn risk. The Paid Search × Creator Tools combination reaches 20.0% churn.
Medium Impact Medium
First Step: Compare keyword targeting, ad copy, and landing page messaging for Paid Search vs. Outbound to identify expectation mismatches. Test revised messaging that aligns more closely with the Outbound value proposition.
Evidence: Churned accounts took 24.6 days to reach value vs. 18.5 for retained accounts (Cohen's d = 0.64). Time-to-value correlates negatively with health score (r = -0.27), indicating activation speed sets baseline trajectory.
Medium Impact Medium
First Step: Map the activation journey for the median 19-day cohort and identify the top 3 friction points delaying value realization. Prioritize product or process changes that eliminate these bottlenecks.
Evidence: Starter plan churns at 14.72%, 2.6pp above average and 5.56pp above Enterprise. This represents a 60.7% relative increase vs. Enterprise churn. With 224 churned accounts, Starter contributes disproportionate risk.
Medium Impact Hard
First Step: Survey recent Starter plan churns to identify whether price, feature limits, or support access drove the decision. Test a revised Starter tier with adjusted pricing or one additional high-value feature.
Evidence: Churned accounts submitted 6.9 tickets vs. 3.4 for retained accounts (Cohen's d = 0.85). Support volume correlates with unresolved tickets at r = 0.79, indicating capacity constraints as volume rises.
Medium Impact Easy
First Step: Configure alerts to flag any account submitting 6+ tickets in a 90-day window. Route these accounts to a senior support engineer with authority to escalate product issues directly to engineering.
Evidence: Health score (Cohen's d = -1.24), onboarding score (Cohen's d = -0.89), support tickets (Cohen's d = 0.85), and unresolved tickets (Cohen's d = 0.83) show large effect sizes. A predictive model could identify at-risk accounts before they reach critical health thresholds.
Medium Impact Medium
First Step: Build a Gradient Boosting classifier using health_score, onboarding_score, support_tickets_90d, unresolved_tickets, and time_to_value_days as primary features. Validate against a holdout set and deploy weekly churn risk scores to the CSM dashboard.
Evidence: Outbound customers churn at 10.33%, the lowest rate among all channels and 1.79pp below average. This suggests stronger pre-qualification or intent signaling during the sales process.
Low Impact Medium
First Step: Document the Outbound sales qualification criteria and compare against Paid Search and Organic self-serve flows. Identify qualification gaps and test adding 1-2 friction points (e.g. use case selection, activation checklist) to self-serve signups.
One in eight customers will churn within 90 days.
HIGH Severity
The baseline 12.12% churn rate translates to 788 accounts at immediate risk. With segment concentrations in Creator Tools (15.04%), Starter plan (14.72%), and Paid Search (13.8%), a significant portion of the customer base sits in elevated-risk categories. If average MRR is $655, the monthly MRR at risk across these 788 accounts approaches $516K (788 accounts × $655), or roughly $1.55M over the full 90-day churn window.
Mitigation: Deploy health score alerts, unresolved ticket SLAs, and onboarding outreach programs within 30 days to stabilize the highest-risk cohorts. Prioritize accounts in multiple high-risk segments (e.g. Starter plan + Creator Tools + Paid Search).
Support capacity constraints create unresolved ticket backlogs.
HIGH Severity
The strong correlation between total tickets and unresolved tickets (r = 0.79) indicates support resolution speed degrades as volume increases. 6.38% of accounts have 3+ unresolved tickets, representing sustained support failures. Churned accounts had nearly 3× the unresolved ticket count of retained accounts (1.4 vs. 0.5).
Mitigation: Increase support staffing or implement intelligent routing to prevent high-volume accounts from overwhelming the queue. Establish escalation paths for accounts with 2+ unresolved tickets and health scores below 75.
Creator Tools product-market fit gap.
MEDIUM Severity
Creator Tools shows 15.04% churn (2.92pp above average) with elevated rates across multiple acquisition channels. The Paid Search × Creator Tools combination reaches 20.0% churn, nearly double the company baseline. This cross-channel consistency suggests the product does not adequately serve this vertical's workflows or needs.
Mitigation: Conduct exit interviews and feature gap analysis specific to Creator Tools. Consider whether to invest in vertical-specific features or reduce acquisition spend for this segment until product improvements are deployed.
Paid Search CAC payback at risk due to elevated churn.
MEDIUM Severity
Paid Search customers churn at 13.8%, 3.47pp above the best-performing channel (Outbound). If Paid Search CAC is comparable to other channels, the 33.6% higher churn rate severely damages payback period and lifetime value. The Paid Search × Creator Tools segment reaches 20.0% churn, amplifying the concern.
Mitigation: Audit keyword targeting, ad copy, and landing page expectations to reduce intent mismatches. Test tighter qualification in the signup flow or redirect Paid Search spend toward segments with stronger retention (Ecommerce, B2B SaaS).
Onboarding deficits compound over time.
MEDIUM Severity
The 14.8-point gap in onboarding scores between churned and retained accounts (Cohen's d = -0.89) persists throughout the customer lifecycle due to the moderate correlation between onboarding and long-term health (r = 0.43). Accounts scoring below 60 rarely recover to above-average health, creating a cohort locked into eventual churn.
Mitigation: Implement proactive outreach for accounts scoring below 60 within the first 30 days. Do not wait for health scores to decline months later — intervene immediately post-onboarding to correct activation deficits.
| Column | Mean | Median | Std Dev | Min | Max | Skewness | Outliers |
|---|---|---|---|---|---|---|---|
| months_active | 15.59 | 16.0 | 7.58 | 1.0 | 40.0 | 0.11 (symmetric) | 11 (0.17%) |
| mrr | $655.50 | $524.99 | $484.47 | $51.83 | $6,564.07 | 2.52 (highly right-skewed) | 334 (5.14%) |
| onboarding_score | 68.78 | 69.0 | 17.18 | 0.0 | 100.0 | -0.20 (symmetric) | 27 (0.42%) |
| monthly_login_days | 14.59 | 15.0 | 6.81 | 0.0 | 38.0 | 0.08 (symmetric) | 32 (0.49%) |
| support_tickets_90d | 3.79 | 2.0 | 4.26 | 0.0 | 39.0 | 2.04 (highly right-skewed) | 387 (5.95%) |
| unresolved_tickets | 0.59 | 0.0 | 1.19 | 0.0 | 15.0 | 3.35 (highly right-skewed) | 415 (6.38%) |
| time_to_value_days | 19.20 | 19.0 | 9.72 | 1.0 | 62.0 | 0.34 (symmetric) | 60 (0.92%) |
| net_revenue_expansion | 0.08 | 0.08 | 0.22 | -0.40 | 0.91 | 0.12 (symmetric) | 41 (0.63%) |
| health_score | 81.24 | 82.7 | 11.56 | 23.3 | 99.0 | -1.14 (highly left-skewed) | 220 (3.38%) |
| churned_next_90d | 0.12 | 0.0 | 0.33 | 0.0 | 1.0 | 2.32 (highly right-skewed) | 788 (12.12%) |