Dashboard Guide – SaaSGuard BI Layer¶

SaaSGuard ships four Apache Superset dashboards that translate model outputs into actionable CS workflows. Each dashboard is backed by the marts.mart_customer_risk_scores dbt model, which materializes rule-based risk scores for all 5,000 active customers.

Quick Access¶

Dashboard	URL	Primary User	Refresh
Customer 360	`:8088/superset/dashboard/customer-360/`	CSM	On demand
Churn Heatmap	`:8088/superset/dashboard/churn-heatmap/`	VP CS / CRO	Daily
Risk Drill-Down	`:8088/superset/dashboard/risk-drilldown/`	CS Team Lead	Daily
Uplift Simulator	`:8088/superset/dashboard/uplift-simulator/`	VP CS / Finance	Weekly

Dashboard 1 — Customer 360¶

What it shows: A complete risk snapshot for a single customer account. Used by a CSM to prepare for an EBR (Executive Business Review), renewal conversation, or intervention call.

Charts¶

Chart	Type	Business Question
Churn Score KPI	Big Number + Trend	What is this customer's current risk level?
Risk Flag Breakdown	Horizontal Bar	Which specific signals are driving the risk?
Usage Trend (90d)	Line Chart	Is engagement accelerating or decelerating?
Open Support Tickets	Table	What issues need immediate resolution?
GTM Opportunity	KPI Card	Is there a renewal or expansion in flight?

How to use¶

Set the customer_id filter (top of dashboard) to the account you're reviewing.
Read the Churn Score first — this is the headline risk number (0–100%).
Check Risk Flag Breakdown to understand why the score is high.
low_events → product adoption intervention needed
support_overload → escalate open tickets before renewal call
onboarding_at_risk → trigger onboarding check-in for new accounts
Review the Usage Trend — a declining slope in the last 30d is a leading indicator.
Cross-reference the GTM Opportunity — a high-risk customer with an open renewal needs coordinated CS + Sales alignment immediately.

Business narrative¶

"At SaaSGuard, a CSM preparing for an account call traditionally spent 15 minutes gathering data from four different tools. The Customer 360 dashboard consolidates churn probability, usage cadence, support status, and GTM signals into a 30-second review. This directly supports the CS team's ability to intervene before churn, rather than reacting after it."

Dashboard 2 — Churn Heatmap¶

What it shows: Portfolio-level view of churn risk concentration across customer segments (plan_tier × industry). Answers the VP CS question: "Where is our ARR most at risk this quarter?"

Charts¶

Chart	Type	Business Question
Risk Tier Heatmap	Heatmap	Which segment has the highest concentration of risk?
Risk Tier Distribution	Donut	What % of our portfolio is at HIGH or CRITICAL risk?
ARR at Risk by Plan Tier	Stacked Bar	Where is revenue risk concentrated by tier?
Churn Rate by Industry	Horizontal Bar	Which verticals show the highest attrition?
KPI Summary Row	Big Numbers	Portfolio-level headline metrics
Score Distribution	Histogram	What does the full risk distribution look like?

How to read the heatmap¶

Dark red cells (high churn score + large count) require segment-level CS strategy changes.
Enterprise + FinTech is typically the highest-revenue cell — even a 1% improvement here has outsized ARR impact.
Starter + Early Stage cells often show high churn rates but low ARR impact per customer; these warrant product-led retention (in-app nudges, onboarding automation) rather than CSM time.

Business narrative¶

"The Churn Heatmap gives the VP of Customer Success a portfolio risk posture in 30 seconds. Combined with the Uplift Simulator, it enables data-driven resource allocation: assign senior CSMs to the high-ARR/high-risk quadrant, and use automated playbooks for the high-count/low-ARR segments."

Dashboard 3 — Risk Drill-Down¶

What it shows: Ranked list of at-risk customers (HIGH + CRITICAL) with recommended CS actions, usage metrics, and correlation analytics. The CS team's daily queue prioritisation tool.

Charts¶

Chart	Type	Business Question
At-Risk Customer Table	Table w/ conditional formatting	Which customers need outreach today?
Churn Score vs Events (scatter)	Scatter Plot	How strong is the usage-decay signal?
Usage Decay Funnel	Funnel	How many customers are at each engagement stage?
Support Load vs Churn	Bar	Do high-ticket-volume customers churn more?
Onboarding Activation Gate	Bar	Does early integration predict 90d retention?

Reading the customer table¶

Sorted by ARR at risk (descending) — highest-impact interventions appear first.
recommended_action column mirrors the domain rule in PredictionResult:
🔴 CRITICAL — escalate same day, EBR within 7 days
🟠 HIGH RISK — outreach within 48 hours
🟡 MEDIUM — add to watch list, next weekly review
The top_risk_drivers column shows the two most active risk signals — copy this into the AI summary prompt for context.

Key analytic findings¶

Usage decay signal: customers with events_last_30d < 5 have 3.2× higher churn probability than active users. The scatter chart makes this correlation visible.
Support overload signal: customers with ≥2 high-priority tickets churn at 2.1× the base rate — often the last signal before cancellation.
Onboarding activation gate: ≥3 integration connects in first 30 days reduces first-90-day churn by 63% (SHAP finding).

Business narrative¶

"The Risk Drill-Down is the daily CS standup tool. The team lead filters to CRITICAL + HIGH risk, assigns the top 10 accounts by ARR impact, and each CSM uses the Customer 360 dashboard for their specific accounts. The onboarding activation gate finding directly informs CS playbook design: every new customer should be guided to 3+ integrations in their first month."

Dashboard 4 — Uplift Simulator¶

What it shows: What-if analysis estimating the ARR recovery potential from targeting the top-N at-risk customers with CS interventions. Enables data-driven budget allocation and ROI conversations with finance.

Charts¶

Chart	Type	Business Question
Cumulative ARR Recovery	Line Chart	What's the incremental return on each additional account?
Intervention ROI Table	Table	Break-even analysis at 10/25/50/100 accounts
Segment Uplift Opportunity	Bar	Which segments offer the best recovery ROI?
KPI Summary (recoverable ARR)	Big Numbers	Headline: total recoverable ARR across scenarios
Early-Stage Intervention	Bar	Is onboarding-stage intervention worth prioritising?

How to use the simulator¶

Read the cumulative recovery chart — the steep early slope shows that the top 10–20 accounts account for a disproportionate share of recoverable ARR (Pareto principle).
Use the ROI table to answer: "If we invest $X in CS outreach, what's the return?"
Each account intervention is estimated at $600 (2 CSM hours × $300/hr)
At 15% effectiveness (Forrester benchmark): recoverable_arr_15pct / intervention_cost
Filter by plan_tier to build a tiered CS coverage model:
Enterprise: high-touch, 1:1 CSM
Growth: pooled CSM + AI summaries
Starter: product-led playbooks + email automation

Business narrative¶

"On $200M ARR with a 5% churn rate, $10M is at risk annually. Targeting the top 50 at-risk customers by ARR impact — a 3-hour CSM investment each — could recover $580K at 15% effectiveness for a 4:1 ROI. This dashboard makes that conversation concrete, turning churn prediction into a capital allocation decision."

Data Freshness & Refresh Schedule¶

Data Layer	Refresh Trigger	Staleness Tolerance
`raw.*` tables	Synthetic data (daily in production)	24h
`marts.mart_customer_churn_features`	`dbt run` (triggered by raw refresh)	24h
`marts.mart_customer_risk_scores`	`dbt run --select mart_customer_risk_scores`	24h
Superset dashboard cache	Superset refresh schedule (hourly)	1h

To refresh manually:

docker compose exec dbt dbt run --select mart_customer_risk_scores

Known Limitations¶

Rule-based scores vs. ML model: mart_customer_risk_scores.churn_score uses a rule-based approximation. For calibrated probabilities, use POST /predictions/churn. The dashboard scores are correlated but not identical to the XGBoost outputs.
No real-time scores: Scores update daily with the dbt run. For fresh scores, trigger the API endpoint and refresh the page.
Customer 360 requires customer_id filter: Without a filter, the chart returns the first customer alphabetically. Set the filter before using the dashboard.
Uplift assumptions: The 10/15/20% churn reduction rates are industry benchmarks (Forrester 2023). Your actual intervention effectiveness should be measured and substituted once the program has run for one quarter.

Dashboard 5 — Propensity Quadrant¶

Purpose: Maps every active expansion candidate into one of four quadrants based on rule-based churn risk and expansion proxy score. CS can instantly see which customers to expand, which to rescue, and which are in conflict.

Data source: marts.mart_propensity_quadrant (backed by mart_customer_risk_scores + mart_customer_expansion_features)

URL: :8088/superset/dashboard/propensity-quadrant/

Primary users: VP CS, Head of Revenue

Refresh: Daily (run dbt run --select mart_propensity_quadrant)

Quadrant Definitions¶

Quadrant	Churn risk	Expansion signal	Action
Flight Risk	≥ 0.50	≥ 0.50	Senior Exec intervention — restore health before any upsell motion
Growth Engine	< 0.25	≥ 0.50	Active upgrade conversation — highest ROI segment
Churn Candidate	≥ 0.50	< 0.25	Retention priority — defer expansion until account health restored
Stable Base	all others	all others	Nurture — product-led expansion signals, no active outreach needed

SQL Chart Config (Superset)¶

-- Scatter plot: churn_probability (X) vs upgrade_propensity_proxy (Y)
-- Color by quadrant_label, size by expected_arr_uplift_proxy
SELECT
    customer_id,
    churn_probability,
    upgrade_propensity_proxy,
    quadrant_label,
    expected_arr_uplift_proxy,
    plan_tier,
    was_contacted,
    days_since_last_outreach
FROM marts.mart_propensity_quadrant
ORDER BY expected_arr_uplift_proxy DESC

Chart type: Scatter plot - X-axis: churn_probability [0, 1] - Y-axis: upgrade_propensity_proxy [0, 1] - Color dimension: quadrant_label - Size dimension: expected_arr_uplift_proxy - Add reference lines at X=0.5 and Y=0.5 to draw quadrant borders

Refresh Command¶

# Rebuild the quadrant mart after dbt run
docker compose exec dbt dbt run --select mart_propensity_quadrant
# Or full refresh:
docker compose exec dbt dbt run && dbt test --select mart_propensity_quadrant

Notes¶

The expansion proxy score is rule-based, not the ML model score. It uses a weighted sum of premium_feature_trials_30d (0.30), feature_limit_hit_30d (0.25), feature_request_tickets_90d (0.15), has_open_expansion_opp (0.20), and mrr_tier_ceiling_pct (0.10).
For free-tier customers, feature_limit_hit_30d carries more weight than mrr_tier_ceiling_pct (which is always 0).
The was_contacted and days_since_last_outreach columns enable CS managers to see which Flight Risk and Growth Engine customers have already been contacted.
To get the ML-scored version, use the /predictions/customers/{id}/360 API endpoint which returns is_flight_risk as a machine-readable boolean.