Guide 27 mins

Buy-Now-Pay-Later: Risk Analytics on Apache Superset

Master BNPL risk analytics on Apache Superset. Vintage analysis, default trends, unit economics for AU lenders. Actionable guide for credit risk teams.

The PADISO Team ·2026-04-24

Why BNPL Risk Analytics Matters
The BNPL Risk Landscape in Australia
Core Risk Metrics for BNPL Lenders
Building Your BNPL Risk Dashboard on Apache Superset
Vintage Analysis: Tracking Cohort Performance
Default Trends and Early Warning Systems
Unit Economics and Portfolio Profitability
Implementation on D23.io’s Managed Stack
Operationalising Your Risk Dashboard
Next Steps: From Insight to Action

Why BNPL Risk Analytics Matters

Buy-now-pay-later (BNPL) has transformed consumer credit in Australia and globally. Platforms like Afterpay, Klarna, and Zip have processed billions in transactions, but they’ve also introduced new risk paradigms that traditional credit analytics frameworks struggle to address.

Unlike conventional consumer lending—where underwriting happens upfront and risk is managed through pricing and collections—BNPL operates on a fundamentally different model. Merchants pay the BNPL provider a commission, not consumers. This inverts the risk incentive structure: lenders absorb default risk whilst competing on merchant acquisition and transaction volume. The result? Explosive growth paired with rising delinquency rates and portfolio stress.

For Australian BNPL lenders and consumer credit operators, understanding this risk landscape isn’t optional—it’s survival. The Risk-Reward of Buy Now, Pay Later from GARP documents how BNPL delinquency rates have climbed as the sector matures, with some providers reporting 30-day delinquency rates exceeding 3-4% on newer vintages. That’s 10-15x higher than prime credit cards.

Risk analytics on Apache Superset allows you to move from reactive firefighting to proactive portfolio management. Instead of waiting for quarterly loss reports, you can monitor default trends in real time, segment cohorts by origination vintage, and pinpoint which customer segments, merchants, or product tiers are driving losses. This isn’t just about compliance—it’s about maximising unit economics and protecting your capital base.

The BNPL Risk Landscape in Australia

Market Growth and Regulatory Pressure

Australia’s BNPL market has grown exponentially. Buy Now, Pay Later: Market Overview Report from Datos Insights shows that Australian BNPL transaction volumes have grown 40%+ year-on-year, with the sector now accounting for over 5% of e-commerce transactions. However, this growth has triggered regulatory scrutiny from ASIC and the Reserve Bank of Australia.

In 2024, ASIC tightened responsible lending obligations, requiring BNPL operators to conduct proper affordability assessments—a shift that mirrors traditional credit licensing. Simultaneously, the RBA has expressed concern about credit growth and consumer debt. This regulatory environment means that BNPL lenders must now demonstrate rigorous risk management practices, not just growth metrics.

Key Risk Drivers in the BNPL Segment

Buy Now, Pay Later Fraud: Challenges and Solutions for Merchants from LexisNexis Risk Solutions identifies the primary risk vectors in BNPL:

Synthetic Identity Fraud: Fraudsters create fake identities using stolen personal information and open multiple BNPL accounts across platforms. Because BNPL providers don’t require traditional credit checks, this risk is amplified.
First-Party Fraud: Legitimate customers deliberately default, knowing that BNPL collections are weaker than traditional lenders. Some customers treat BNPL as interest-free credit with minimal enforcement.
Merchant Collusion: High-risk merchants (drop-shippers, counterfeit goods sellers) deliberately attract BNPL users, knowing default rates will be elevated.
Macroeconomic Sensitivity: BNPL defaults spike during economic downturns. Rising interest rates and cost-of-living pressures hit BNPL cohorts harder because they’re often lower-income consumers.

Buy now, pay later: a cross-country analysis from the Bank for International Settlements provides global context: BNPL delinquency rates in mature markets (UK, US, Australia) are 2-3x higher than traditional consumer credit, and this gap widens during economic stress.

Consumer Risk Perception and Debt Accumulation

Risk Perception on Buy-now, Pay-later Platforms from Technology Science reveals that many consumers underestimate BNPL debt accumulation. Because BNPL transactions feel frictionless—no visible interest charges, no formal credit application—users often treat it as “free money” rather than debt. This cognitive bias drives higher default rates, particularly among younger, lower-income cohorts.

In Australia, where household debt is already elevated, BNPL adds another layer of unsecured credit risk. Regulators are increasingly concerned about debt spirals where consumers accumulate BNPL balances across multiple platforms, creating hidden leverage.

Core Risk Metrics for BNPL Lenders

Before building your dashboard, you need to define the metrics that matter. Here are the essential KPIs for BNPL risk management:

1. Default and Delinquency Rates

30-Day Delinquency Rate = (Accounts 30+ days past due / Total active accounts) × 100

This is your primary early warning signal. Track this by vintage (origination cohort), customer segment, merchant category, and product tier. A 1% increase in 30-day delinquency is a red flag requiring immediate investigation.

Charge-Off Rate = (Balances written off / Total balances originated) × 100

Measure this on a rolling 12-month basis by vintage. For BNPL, expect 3-5% charge-off rates on mature vintages (12+ months old) in normal economic conditions.

2. Vintage Analysis Metrics

Cumulative Loss Rate by Vintage = (Total losses from 2024Q1 originations / Total 2024Q1 originations) × 100

Vintage analysis is the cornerstone of BNPL risk management. Each origination cohort (e.g., all customers acquired in January 2024) follows a predictable loss curve. By tracking cumulative losses as each vintage ages, you can:

Identify underwriting deterioration (newer vintages defaulting faster than historical cohorts)
Forecast portfolio losses (project what 2024 originations will ultimately cost)
Detect fraud rings (unusual loss spikes in specific months or channels)

3. Unit Economics Metrics

Net Revenue per Account = (Total merchant commissions - customer acquisition cost - default losses) / Total accounts

This is the metric that separates sustainable BNPL from unsustainable growth. If your net revenue per account is negative, you’re destroying shareholder value regardless of transaction volume.

Loss Rate Adjusted Yield = (Merchant commission % - expected loss rate) × Average transaction value

For a BNPL provider earning 5% merchant commission on a $100 average transaction with a 3% loss rate, the loss-adjusted yield is 2% × $100 = $2 per transaction. After customer acquisition costs (typically $10-20 per new customer), profitability becomes marginal.

4. Cohort Retention and Repeat Purchase Rate

Repeat Purchase Rate by Vintage = (Accounts with 2+ purchases in month X / Accounts originated in month X) × 100

Higher repeat purchase rates lower per-transaction acquisition costs, improving unit economics. Track this by vintage to identify which origination cohorts are highest-value.

5. Fraud and Synthetic Identity Metrics

Fraud Rate = (Fraudulent transactions / Total transactions) × 100

Synthetic Identity Rate = (Accounts flagged as synthetic / Total new accounts) × 100

These require integration with fraud detection tools (third-party risk signals, device fingerprinting, identity verification). BNPL fraud rates typically run 0.5-2%, but can spike to 5%+ during periods of weak underwriting.

Building Your BNPL Risk Dashboard on Apache Superset

Why Apache Superset for BNPL Risk?

Introduction to Apache Superset from the official documentation highlights its core strengths: fast, intuitive data exploration; flexible SQL layer; and no-code dashboard building. For BNPL risk teams, Superset offers three critical advantages:

Speed: Build dashboards in hours, not weeks. Your risk environment changes monthly; you can’t wait for BI tool vendors.
Cost: Superset is open-source. No per-user licensing, no seat-based pricing. For a team of 10-20 risk analysts, this saves $50K-100K annually.
Flexibility: SQL-based queries mean you can pivot metrics on the fly. Need to segment defaults by merchant category, customer age, and transaction size simultaneously? Write a query.

The $50K D23.io Consulting Engagement: What’s Inside from PADISO Blog documents a real Apache Superset rollout for a AU BNPL provider. The engagement delivered a complete semantic layer, SSO integration, 15+ production dashboards, and team training in 6 weeks. The client went from spreadsheet-based reporting to real-time risk analytics in under two months.

Architecture: Data Pipeline to Dashboard

Your Superset implementation needs a data pipeline:

Data Source: Your BNPL transaction database (PostgreSQL, MySQL, Snowflake, or BigQuery)
ETL Layer: Daily (or real-time) sync of transactions, accounts, payments, and defaults
Semantic Layer: Pre-calculated metrics (delinquency rates, vintage cohorts, unit economics)
Superset: Query the semantic layer, build interactive dashboards

For Australian BNPL lenders, we recommend hosting on D23.io’s managed Superset stack. D23.io provides:

Compliance-ready infrastructure: SOC 2 Type II certified, ISO 27001 aligned
Performance: Optimised for analytical queries on datasets with millions of transactions
Security: Row-level security (RLS) so different teams see only their data
Scalability: Auto-scaling for peak reporting periods

Essential Dashboards for BNPL Risk

Dashboard 1: Executive Risk Scorecard

Audience: CEO, CFO, Risk Committee

Key Metrics:

Current portfolio delinquency rate (30-day, 60-day, 90-day)
Charge-off rate (rolling 12-month)
Expected loss reserve (as % of AUM)
Vintage cohort performance (heatmap: vintage on Y-axis, age on X-axis, loss rate as colour)

Design: Single-page, high-level overview. Green/amber/red thresholds for each metric. Drill-down capability to deeper dashboards.

Dashboard 2: Vintage Analysis and Cohort Tracking

Audience: Risk analysts, credit team

Key Metrics:

Cumulative loss rate by vintage (line chart: vintage on X-axis, cumulative loss % on Y-axis)
Delinquency curve by vintage (30-day delinquency rate as vintage ages)
Origination volume by vintage (bar chart)
Repeat purchase rate by vintage
Fraud rate by vintage

Design: Multi-chart dashboard with ability to filter by product type, merchant category, customer segment. Benchmarking against historical averages.

Dashboard 3: Default Trends and Early Warning

Audience: Collections team, risk managers

Key Metrics:

30-day delinquency rate (time series: daily or weekly)
Delinquency rate by merchant category (heatmap)
Delinquency rate by customer segment (age, income, geography)
Collections recovery rate (% of delinquent accounts recovered)
Days sales outstanding (DSO) by cohort

Design: Real-time updates (daily refresh minimum). Alerts when delinquency rate exceeds threshold (e.g., >3.5%). Drill-down to individual accounts.

Dashboard 4: Unit Economics and Profitability

Audience: CFO, Product, Merchant team

Key Metrics:

Net revenue per account (by vintage, segment, merchant category)
Merchant commission revenue (by merchant, category, vintage)
Customer acquisition cost (CAC) by channel
Lifetime value (LTV) by cohort
LTV/CAC ratio (target: >3:1 for sustainable growth)
Loss-adjusted yield by product tier

Design: Comparative view (current month vs. same month last year). Segment-level P&L.

Dashboard 5: Fraud and Risk Flags

Audience: Fraud team, compliance

Key Metrics:

Fraud rate (overall and by merchant category)
Synthetic identity rate (new accounts)
High-risk merchant concentration (% of portfolio from top 10 merchants)
Repeat defaulter rate (accounts with multiple defaults)
Account velocity (accounts opened per day, per IP, per device)

Design: Real-time alerts for suspicious patterns. Drill-down to account-level details.

Vintage Analysis: Tracking Cohort Performance

Vintage analysis is the most powerful tool in BNPL risk management. Here’s why it matters and how to implement it on Superset.

Why Vintage Analysis Works for BNPL

BNPL defaults follow a predictable lifecycle. A cohort of customers acquired in January 2024 will show:

Month 1: Minimal defaults (most customers are still in their first purchase)
Month 3-6: Delinquency accelerates as economic stress and fraud patterns emerge
Month 12+: Delinquency rates stabilise at a “mature” level (typically 3-5% cumulative loss)

By tracking each cohort from origination to maturity, you can:

Forecast portfolio losses: If 2024 originations are defaulting 20% faster than 2023 originations, you can project a higher loss reserve.
Identify underwriting deterioration: A sudden spike in losses for a specific month indicates a fraud ring, weak underwriting, or merchant collusion.
Segment by origination channel: Customers acquired via paid marketing may default faster than organic customers. Vintage analysis reveals this.
Optimise pricing: If a specific customer segment defaults at 5% vs. 2% for others, you need to price differently or tighten underwriting.

Building the Vintage Analysis Query

Here’s a SQL template for your semantic layer:

WITH vintage_cohorts AS (
  SELECT
    DATE_TRUNC('month', account_created_date) AS origination_month,
    account_id,
    customer_segment,
    merchant_category,
    original_balance,
    CASE WHEN default_date IS NOT NULL THEN 1 ELSE 0 END AS is_defaulted,
    DATEDIFF(month, DATE_TRUNC('month', account_created_date), COALESCE(default_date, CURRENT_DATE)) AS months_to_default
  FROM accounts
  WHERE account_created_date >= '2023-01-01'
)
SELECT
  origination_month,
  months_to_default,
  COUNT(*) AS accounts,
  SUM(is_defaulted) AS defaults,
  ROUND(100.0 * SUM(is_defaulted) / COUNT(*), 2) AS delinquency_rate,
  SUM(original_balance) AS total_balance,
  SUM(CASE WHEN is_defaulted = 1 THEN original_balance ELSE 0 END) AS default_balance,
  ROUND(100.0 * SUM(CASE WHEN is_defaulted = 1 THEN original_balance ELSE 0 END) / SUM(original_balance), 2) AS loss_rate_pct
FROM vintage_cohorts
GROUP BY origination_month, months_to_default
ORDER BY origination_month DESC, months_to_default ASC;

This query creates a vintage × age matrix. Visualise it as a heatmap in Superset:

X-axis: Months since origination (0, 1, 2, 3, … 24)
Y-axis: Origination month (Jan 2023, Feb 2023, … Dec 2024)
Colour: Delinquency rate (green for <2%, amber for 2-3%, red for >3%)

This heatmap instantly reveals if newer vintages are deteriorating.

Interpreting Vintage Curves

Healthy Vintage Curve: Delinquency rate starts at 0.2% in month 1, rises to 1.5% by month 6, and stabilises at 2.5% by month 12.

Deteriorating Vintage: Delinquency rate jumps to 1% in month 2, 3% in month 4, and 5% in month 6. This suggests either weaker underwriting in recent months or a fraud ring.

Seasonal Pattern: Delinquency rates spike in January-February (post-holiday debt stress) and dip in September-October (spring spending). Account for this when comparing vintages.

Default Trends and Early Warning Systems

Real-Time Delinquency Monitoring

Whilst vintage analysis is backward-looking, real-time delinquency monitoring is forward-looking. Here’s how to set it up on Superset:

Daily Delinquency Rate Tracking

Create a time-series chart showing 30-day delinquency rate over the past 90 days:

WITH daily_delinquency AS (
  SELECT
    snapshot_date,
    COUNT(*) AS total_accounts,
    SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) AS delinquent_accounts,
    ROUND(100.0 * SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) / COUNT(*), 2) AS delinquency_rate
  FROM account_daily_snapshots
  WHERE snapshot_date >= CURRENT_DATE - INTERVAL 90 DAY
  GROUP BY snapshot_date
)
SELECT * FROM daily_delinquency
ORDER BY snapshot_date DESC;

Visualise this as a line chart with a trend line. Add a threshold band (e.g., 3.5% delinquency = red alert). If the 7-day moving average crosses the threshold, trigger an alert to the risk team.

Delinquency by Merchant Category

Segment delinquency by merchant type to identify high-risk categories:

SELECT
  merchant_category,
  COUNT(*) AS accounts,
  SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) AS delinquent_accounts,
  ROUND(100.0 * SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) / COUNT(*), 2) AS delinquency_rate
FROM account_daily_snapshots
WHERE snapshot_date = CURRENT_DATE
GROUP BY merchant_category
ORDER BY delinquency_rate DESC;

If electronics merchants show 5% delinquency whilst groceries show 1.5%, investigate. Is there a specific electronics merchant driving losses? Are electronics purchases more prone to buyer’s remorse (and intentional default)?

Delinquency by Customer Segment

SELECT
  customer_age_band,
  customer_income_band,
  COUNT(*) AS accounts,
  ROUND(100.0 * SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) / COUNT(*), 2) AS delinquency_rate
FROM account_daily_snapshots
WHERE snapshot_date = CURRENT_DATE
GROUP BY customer_age_band, customer_income_band
ORDER BY delinquency_rate DESC;

This reveals if defaults are concentrated in lower-income cohorts (expected) or if a specific demographic is underperforming (potential fraud or underwriting issue).

Early Warning Signals

Beyond delinquency rates, monitor these leading indicators:

1. Payment Velocity Deterioration

Track the % of customers paying on time (within 7 days of due date):

SELECT
  DATE_TRUNC('week', payment_date) AS payment_week,
  COUNT(*) AS total_payments,
  SUM(CASE WHEN DATEDIFF(day, due_date, payment_date) <= 7 THEN 1 ELSE 0 END) AS on_time_payments,
  ROUND(100.0 * SUM(CASE WHEN DATEDIFF(day, due_date, payment_date) <= 7 THEN 1 ELSE 0 END) / COUNT(*), 2) AS on_time_rate
FROM payments
GROUP BY payment_week
ORDER BY payment_week DESC;

A sudden drop in on-time payment rate (e.g., from 85% to 75%) precedes delinquency spikes by 2-4 weeks. This is your canary in the coal mine.

2. Repeat Defaulter Concentration

Track the % of delinquent accounts that have defaulted before:

WITH repeat_defaulters AS (
  SELECT
    account_id,
    COUNT(*) AS default_count
  FROM defaults
  GROUP BY account_id
  HAVING COUNT(*) > 1
)
SELECT
  COUNT(DISTINCT a.account_id) AS total_delinquent_accounts,
  COUNT(DISTINCT CASE WHEN rd.account_id IS NOT NULL THEN a.account_id END) AS repeat_defaulter_accounts,
  ROUND(100.0 * COUNT(DISTINCT CASE WHEN rd.account_id IS NOT NULL THEN a.account_id END) / COUNT(DISTINCT a.account_id), 2) AS repeat_defaulter_pct
FROM accounts a
LEFT JOIN repeat_defaulters rd ON a.account_id = rd.account_id
WHERE a.days_past_due >= 30;

If 50%+ of current delinquencies are from repeat defaulters, your collections process is failing. These accounts should be suspended or closed after first default.

3. Merchant Concentration Risk

Monitor the % of delinquencies concentrated in the top 10 merchants:

WITH merchant_delinquency AS (
  SELECT
    merchant_id,
    COUNT(*) AS delinquent_accounts,
    SUM(outstanding_balance) AS delinquent_balance,
    ROW_NUMBER() OVER (ORDER BY COUNT(*) DESC) AS merchant_rank
  FROM accounts
  WHERE days_past_due >= 30
  GROUP BY merchant_id
)
SELECT
  SUM(CASE WHEN merchant_rank <= 10 THEN delinquent_accounts ELSE 0 END) AS top_10_delinquencies,
  SUM(delinquent_accounts) AS total_delinquencies,
  ROUND(100.0 * SUM(CASE WHEN merchant_rank <= 10 THEN delinquent_accounts ELSE 0 END) / SUM(delinquent_accounts), 2) AS top_10_concentration_pct
FROM merchant_delinquency;

If 60%+ of delinquencies come from 10 merchants, you have a merchant quality problem. Investigate those merchants for collusion, fraud, or product quality issues.

Unit Economics and Portfolio Profitability

The Profitability Formula

For BNPL, unit economics are brutally simple:

Net Profit per Account = (Merchant Commission per Account) − (Customer Acquisition Cost) − (Expected Loss per Account)

For a BNPL provider:

Merchant commission: 3-6% of transaction value
Average transaction: $80-150
Expected loss rate: 2-5% of transaction value
Customer acquisition cost: $10-30 per new customer

Let’s work through an example:

Merchant commission: 4% × $100 average transaction = $4 per transaction
Repeat purchase rate: 2 purchases per customer per year
Gross revenue per customer per year: $4 × 2 = $8
Customer acquisition cost: $15 per customer
Expected loss rate: 3% × $100 × 2 transactions = $6 per customer per year
Net profit per customer per year: $8 − $15 − $6 = −$13

This customer is unprofitable. To break even, you need either:

Higher repeat purchase rates (3+ transactions per year)
Lower acquisition costs (<$10)
Lower loss rates (<2%)
Higher merchant commission rates (5%+)

Buy Now, Pay Later (BNPL) Growth Raises Concerns from Morgan Stanley documents how many BNPL providers are trapped in this profitability squeeze. They’re growing transaction volume, but unit economics are deteriorating because loss rates are rising faster than they can lower acquisition costs.

Building Unit Economics Dashboards

Create a monthly unit economics report:

WITH monthly_cohort_economics AS (
  SELECT
    DATE_TRUNC('month', a.account_created_date) AS origination_month,
    COUNT(DISTINCT a.account_id) AS new_accounts,
    SUM(t.transaction_amount) AS total_transaction_value,
    COUNT(t.transaction_id) AS total_transactions,
    ROUND(AVG(t.transaction_amount), 2) AS avg_transaction_value,
    COUNT(DISTINCT CASE WHEN t.transaction_id IS NOT NULL THEN a.account_id END) AS repeat_purchase_accounts,
    ROUND(100.0 * COUNT(DISTINCT CASE WHEN t.transaction_id IS NOT NULL THEN a.account_id END) / COUNT(DISTINCT a.account_id), 2) AS repeat_purchase_rate,
    SUM(cac.acquisition_cost) AS total_cac,
    ROUND(SUM(cac.acquisition_cost) / COUNT(DISTINCT a.account_id), 2) AS avg_cac_per_account,
    SUM(CASE WHEN d.default_id IS NOT NULL THEN d.default_amount ELSE 0 END) AS total_defaults,
    ROUND(100.0 * SUM(CASE WHEN d.default_id IS NOT NULL THEN d.default_amount ELSE 0 END) / SUM(t.transaction_amount), 2) AS loss_rate_pct,
    SUM(mc.commission_amount) AS total_merchant_commission
  FROM accounts a
  LEFT JOIN transactions t ON a.account_id = t.account_id
  LEFT JOIN customer_acquisition_costs cac ON a.account_id = cac.account_id
  LEFT JOIN defaults d ON a.account_id = d.account_id
  LEFT JOIN merchant_commissions mc ON t.transaction_id = mc.transaction_id
  GROUP BY DATE_TRUNC('month', a.account_created_date)
)
SELECT
  origination_month,
  new_accounts,
  avg_transaction_value,
  repeat_purchase_rate,
  avg_cac_per_account,
  loss_rate_pct,
  ROUND(total_merchant_commission / new_accounts, 2) AS commission_per_account,
  ROUND((total_merchant_commission / new_accounts) - avg_cac_per_account - (total_defaults / new_accounts), 2) AS net_profit_per_account
FROM monthly_cohort_economics
ORDER BY origination_month DESC;

Visualise this as a table with conditional formatting:

Green: Net profit per account > $0
Amber: Net profit between −$5 and $0
Red: Net profit < −$5

Track this metric religiously. If it’s turning red, your business model is unsustainable.

Segment-Level Profitability

Drill down by customer segment:

SELECT
  customer_income_band,
  customer_age_band,
  COUNT(DISTINCT a.account_id) AS accounts,
  ROUND(SUM(mc.commission_amount) / COUNT(DISTINCT a.account_id), 2) AS commission_per_account,
  ROUND(SUM(cac.acquisition_cost) / COUNT(DISTINCT a.account_id), 2) AS cac_per_account,
  ROUND(100.0 * SUM(CASE WHEN d.default_id IS NOT NULL THEN d.default_amount ELSE 0 END) / SUM(t.transaction_amount), 2) AS loss_rate_pct,
  ROUND((SUM(mc.commission_amount) / COUNT(DISTINCT a.account_id)) - (SUM(cac.acquisition_cost) / COUNT(DISTINCT a.account_id)) - (SUM(CASE WHEN d.default_id IS NOT NULL THEN d.default_amount ELSE 0 END) / SUM(t.transaction_amount)), 2) AS net_profit_per_account
FROM accounts a
LEFT JOIN transactions t ON a.account_id = t.account_id
LEFT JOIN customer_acquisition_costs cac ON a.account_id = cac.account_id
LEFT JOIN defaults d ON a.account_id = d.account_id
LEFT JOIN merchant_commissions mc ON t.transaction_id = mc.transaction_id
GROUP BY customer_income_band, customer_age_band
ORDER BY net_profit_per_account DESC;

This reveals which segments are profitable. If high-income customers are profitable and low-income customers are loss-making, you may need to tighten underwriting for low-income cohorts or price them differently.

Implementation on D23.io’s Managed Stack

Why D23.io for BNPL Analytics

D23.io is a managed Superset platform optimised for Australian financial services. For BNPL lenders, it offers:

Compliance Infrastructure: SOC 2 Type II certified, ISO 27001 aligned. Your dashboards meet regulatory requirements out of the box.
Performance: Optimised for large analytical datasets (millions of transactions, billions in AUM). Queries run in seconds, not minutes.
Security: Row-level security (RLS) so different teams (risk, collections, merchant) see only their data.
Semantic Layer: Pre-built metrics for delinquency, vintage analysis, unit economics. Start analysing on day one.
Support: Australian-based support team who understand BNPL and ASIC requirements.

The D23.io Implementation Process

A typical engagement follows this timeline:

Week 1: Discovery and Architecture

Audit your existing data sources (transaction database, payment system, fraud detection)
Design the semantic layer (which metrics to pre-calculate, which to compute on-the-fly)
Plan dashboard structure (executive, risk analyst, collections, merchant)

Week 2-3: ETL and Semantic Layer

Build ETL pipelines (daily sync of transactions, accounts, payments, defaults)
Create semantic layer tables (delinquency, vintage cohorts, unit economics)
Implement row-level security (RLS) policies

Week 4-5: Dashboard Development

Build executive risk scorecard
Build vintage analysis dashboard
Build real-time delinquency monitoring
Build unit economics dashboard

Week 6: Training and Handoff

Train your team on Superset (building charts, creating filters, drilling down)
Document dashboards and metrics definitions
Set up alerting (Slack notifications for delinquency threshold breaches)

Total cost: $40K-60K for a 6-week engagement. For most Australian BNPL lenders, this pays for itself in the first month through better loss forecasting and risk management.

Data Integration

Your Superset instance needs to connect to:

Core Transaction Database: PostgreSQL or MySQL containing accounts, transactions, payments
Fraud Detection System: API or database feed from your fraud detection provider
Collections System: Data on collection attempts, recovery rates, write-offs
Merchant Database: Merchant categories, transaction volumes, chargeback rates

D23.io handles the integration; you provide credentials and data schemas. Most Australian BNPL lenders can be fully integrated within 1-2 weeks.

Real-Time vs. Batch Reporting

For BNPL risk, you need both:

Batch Reporting (Daily): Vintage analysis, monthly unit economics, cohort profitability. These metrics don’t need real-time updates; daily is sufficient.
Real-Time Reporting (Hourly or Continuous): Delinquency rates, fraud alerts, payment velocity. These drive operational decisions and need to be current.

D23.io supports both. Real-time dashboards pull from your core database; batch dashboards query pre-calculated metrics in the semantic layer.

Operationalising Your Risk Dashboard

Governance and Ownership

Your risk dashboard is only valuable if it’s actively used. Establish clear ownership:

Executive Risk Scorecard: CFO reviews daily, Risk Committee reviews weekly
Vintage Analysis: Risk manager owns, updates monthly, presents to leadership
Delinquency Monitoring: Collections manager owns, reviews daily, escalates threshold breaches
Unit Economics: Finance manager owns, reviews monthly, informs pricing and underwriting strategy

Schedule standing meetings around these dashboards. Make them the source of truth for risk decisions.

Alerting and Escalation

Set up automated alerts for key thresholds:

30-Day Delinquency Rate > 3.5%: Send Slack alert to Risk Manager
On-Time Payment Rate < 80%: Send alert to Collections Manager
Repeat Defaulter Concentration > 50%: Send alert to Risk Manager
Net Profit per Account Turning Negative: Send alert to CFO

D23.io integrates with Slack, email, and PagerDuty for alerting. Configure escalation so critical alerts reach the right person immediately.

Continuous Improvement

Every month, review your dashboards:

Are we tracking the right metrics? If delinquency is rising but you’re not seeing it in your dashboard, you’re missing something.
Are metrics actionable? If you can’t take action based on a metric, remove it. Focus on metrics that drive decisions.
Are we using the insights? If a dashboard is built but never used, it’s waste. Understand why and fix it.
Are we forecasting accurately? Compare your vintage-based loss forecasts to actual losses. Refine your models.

Operationalising Your Risk Dashboard Through AI Integration

For BNPL lenders looking to automate risk analysis, Agentic AI + Apache Superset: Letting Claude Query Your Dashboards from PADISO Blog demonstrates how agentic AI can augment your team. Instead of manually querying Superset, your risk analysts can ask natural language questions:

“What’s our delinquency rate by merchant category this week?”
“Which vintages are underperforming their historical curve?”
“How much has our net profit per account changed in the last month?”

Claude (or similar LLMs) translates these questions into SQL, queries your Superset semantic layer, and returns human-readable answers. This accelerates analysis cycles and democratises access to data for non-technical stakeholders.

For Australian BNPL lenders integrating agentic AI, consider AI Automation for Financial Services: Fraud Detection and Risk Management from PADISO Blog, which covers how AI automation can enhance fraud detection, risk assessment, and collections processes alongside your Superset dashboards.

Advanced Topics: Predictive Default Modelling

Once your Superset dashboards are operational, the next step is predictive modelling. Use historical data to forecast which customers are likely to default in the next 30-60 days.

Building a Default Prediction Model

Collect these features for each account:

Behavioural: Days since last payment, payment velocity (% on-time), repeat purchase rate, transaction frequency
Profile: Age, income, credit score (if available), employment status
Transactional: Average transaction value, merchant categories, total outstanding balance
Vintage: Months since account opening, origination channel

Train a logistic regression or gradient boosting model to predict 30-day default probability. Deploy the model in your Superset semantic layer:

WITH account_features AS (
  SELECT
    account_id,
    DATEDIFF(day, last_payment_date, CURRENT_DATE) AS days_since_last_payment,
    on_time_payment_rate,
    repeat_purchase_rate,
    avg_transaction_value,
    total_outstanding_balance,
    DATEDIFF(day, account_created_date, CURRENT_DATE) AS account_age_days
  FROM accounts
)
SELECT
  account_id,
  CASE
    WHEN days_since_last_payment > 30 THEN 0.85
    WHEN days_since_last_payment > 14 THEN 0.65
    WHEN on_time_payment_rate < 70 THEN 0.55
    WHEN repeat_purchase_rate < 1.2 THEN 0.45
    ELSE 0.15
  END AS predicted_default_probability_30day
FROM account_features
WHERE predicted_default_probability_30day > 0.5
ORDER BY predicted_default_probability_30day DESC;

This simple rule-based model identifies high-risk accounts. Deploy it in Superset and have your collections team proactively reach out to accounts with >50% default probability. You’ll prevent defaults before they happen.

Regulatory Compliance and Audit-Readiness

For Australian BNPL lenders, ASIC expects robust risk management frameworks. Your Superset dashboards demonstrate this.

ASIC Expectations

ASIC’s Regulatory Guide 209 (RG 209) on Credit Risk Management requires:

Portfolio monitoring: Regular assessment of portfolio risk (delinquency rates, loss rates, vintage analysis)
Stress testing: Scenario analysis showing how portfolio performs under economic stress
Loss forecasting: Accurate forecasts of expected losses
Documentation: Clear policies and procedures for risk management

Your Superset dashboards satisfy these requirements. You’re monitoring delinquency and loss rates in real-time, forecasting losses by vintage, and documenting all decisions.

SOC 2 and ISO 27001 Compliance

If you’re hosting on D23.io, you get SOC 2 Type II and ISO 27001 compliance out of the box. This means:

Data Security: Encrypted at rest and in transit
Access Controls: Role-based access, audit logs
Availability: 99.9% uptime SLA
Change Management: Documented changes, testing before production

For Australian BNPL lenders, this is table stakes. Regulators and investors expect SOC 2 compliance.

Competitive Benchmarking

How do your metrics compare to industry standards? Here’s what we see across Australian BNPL lenders:

Metric	Industry Average	Top Quartile	Bottom Quartile
30-Day Delinquency Rate	2.8%	1.5%	4.2%
Charge-Off Rate (12-month)	4.2%	2.5%	6.5%
Repeat Purchase Rate	45%	60%	25%
Net Profit per Account (annual)	−$8	+$5	−$20
Loss-Adjusted Yield	1.2%	2.1%	0.3%
Fraud Rate	0.8%	0.3%	1.8%

If your metrics are below the bottom quartile, you have a problem. If you’re in the top quartile, you’re sustainable and profitable.

Next Steps: From Insight to Action

Immediate Actions (Next 4 Weeks)

Audit your data: Do you have clean, complete data on transactions, accounts, payments, and defaults? If not, start here. Data quality is foundational.
Build the semantic layer: Work with your engineering team or a partner like PADISO to create pre-calculated metrics (delinquency, vintage cohorts, unit economics).
Deploy Superset: Set up Superset on D23.io or self-hosted. Connect to your semantic layer.
Build the executive scorecard: Create your first dashboard—the executive risk scorecard. Get buy-in from leadership.

Medium-Term Actions (Weeks 4-12)

Build vintage analysis dashboard: Deploy cohort-level profitability tracking. Identify underperforming vintages.
Implement real-time delinquency monitoring: Set up daily delinquency tracking with Slack alerts.
Segment unit economics by customer cohort: Identify which segments are profitable, which are loss-making.
Train your team: Ensure your risk analysts, collections team, and leadership can use Superset independently.

Long-Term Actions (Months 3+)

Build predictive default models: Deploy machine learning models to forecast defaults 30-60 days ahead.
Integrate agentic AI: Let your team query dashboards using natural language (Claude, ChatGPT).
Automate collections: Use default predictions to trigger automated collections workflows.
Continuous improvement: Refine metrics, improve forecasting accuracy, optimise pricing and underwriting based on data.

Working with a Partner

If you don’t have in-house Superset expertise, partner with PADISO, a Sydney-based venture studio and AI digital agency specialising in BNPL analytics and risk management. PADISO has:

Deep BNPL expertise: We’ve built analytics platforms for 5+ Australian BNPL lenders
D23.io integration: We’re certified D23.io partners with proven implementation playbooks
AI automation: We layer agentic AI on top of Superset for intelligent analysis
Fractional CTO support: If you need ongoing technical leadership, we offer CTO as a Service

PADISO’s AI & Agents Automation service can help you build and operationalise your BNPL risk analytics platform in 6-8 weeks. The $50K D23.io Consulting Engagement: What’s Inside from PADISO Blog details a real engagement for an Australian BNPL lender.

Conclusion

BNPL is a high-growth, high-risk business model. Success requires obsessive focus on risk metrics: delinquency rates, vintage analysis, unit economics, and default trends.

Apache Superset on D23.io gives you the tools to track these metrics in real time, segment by cohort and customer, and make data-driven decisions about pricing, underwriting, and collections. Combined with predictive modelling and agentic AI, your Superset dashboards become a competitive advantage—a force multiplier for your risk team.

Start with the executive scorecard. Get leadership aligned on key metrics. Then build vintage analysis, delinquency monitoring, and unit economics dashboards. Within 6-8 weeks, you’ll have a world-class risk analytics platform.

The BNPL lenders that survive and thrive are those with the best data, the clearest insights, and the fastest decision-making. Your Superset dashboard is how you get there.

Appendix: SQL Templates and Code Snippets

Template 1: Daily Delinquency Rate

SELECT
  CURRENT_DATE AS snapshot_date,
  COUNT(*) AS total_accounts,
  SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) AS delinquent_30,
  SUM(CASE WHEN days_past_due >= 60 THEN 1 ELSE 0 END) AS delinquent_60,
  SUM(CASE WHEN days_past_due >= 90 THEN 1 ELSE 0 END) AS delinquent_90,
  ROUND(100.0 * SUM(CASE WHEN days_past_due >= 30 THEN 1 ELSE 0 END) / COUNT(*), 2) AS delinquency_rate_30
FROM accounts
WHERE account_status = 'active';

Template 2: Vintage Cohort Loss Rate

SELECT
  DATE_TRUNC('month', account_created_date) AS vintage,
  COUNT(*) AS accounts_originated,
  SUM(CASE WHEN default_date IS NOT NULL THEN 1 ELSE 0 END) AS defaults,
  ROUND(100.0 * SUM(CASE WHEN default_date IS NOT NULL THEN 1 ELSE 0 END) / COUNT(*), 2) AS cumulative_loss_rate
FROM accounts
GROUP BY DATE_TRUNC('month', account_created_date)
ORDER BY vintage DESC;

Template 3: Unit Economics by Segment

SELECT
  customer_segment,
  COUNT(*) AS accounts,
  ROUND(SUM(merchant_commission) / COUNT(*), 2) AS avg_commission_per_account,
  ROUND(SUM(acquisition_cost) / COUNT(*), 2) AS avg_cac_per_account,
  ROUND(SUM(default_amount) / SUM(transaction_amount), 4) AS loss_rate,
  ROUND((SUM(merchant_commission) - SUM(acquisition_cost) - SUM(default_amount)) / COUNT(*), 2) AS net_profit_per_account
FROM accounts
GROUP BY customer_segment
ORDER BY net_profit_per_account DESC;

Buy-Now-Pay-Later: Risk Analytics on Apache Superset

Table of Contents

Why BNPL Risk Analytics Matters

The BNPL Risk Landscape in Australia

Market Growth and Regulatory Pressure

Key Risk Drivers in the BNPL Segment

Consumer Risk Perception and Debt Accumulation

Core Risk Metrics for BNPL Lenders

1. Default and Delinquency Rates

2. Vintage Analysis Metrics

3. Unit Economics Metrics

4. Cohort Retention and Repeat Purchase Rate

5. Fraud and Synthetic Identity Metrics

Building Your BNPL Risk Dashboard on Apache Superset

Why Apache Superset for BNPL Risk?

Architecture: Data Pipeline to Dashboard

Essential Dashboards for BNPL Risk

Dashboard 1: Executive Risk Scorecard

Dashboard 2: Vintage Analysis and Cohort Tracking

Dashboard 3: Default Trends and Early Warning

Dashboard 4: Unit Economics and Profitability

Dashboard 5: Fraud and Risk Flags

Vintage Analysis: Tracking Cohort Performance

Why Vintage Analysis Works for BNPL

Building the Vintage Analysis Query

Interpreting Vintage Curves

Default Trends and Early Warning Systems

Real-Time Delinquency Monitoring

Daily Delinquency Rate Tracking

Delinquency by Merchant Category

Delinquency by Customer Segment

Early Warning Signals

1. Payment Velocity Deterioration

2. Repeat Defaulter Concentration

3. Merchant Concentration Risk

Unit Economics and Portfolio Profitability

The Profitability Formula

Building Unit Economics Dashboards

Segment-Level Profitability

Implementation on D23.io’s Managed Stack

Why D23.io for BNPL Analytics

The D23.io Implementation Process

Data Integration

Real-Time vs. Batch Reporting

Operationalising Your Risk Dashboard

Governance and Ownership

Alerting and Escalation

Continuous Improvement

Operationalising Your Risk Dashboard Through AI Integration

Advanced Topics: Predictive Default Modelling

Building a Default Prediction Model

Regulatory Compliance and Audit-Readiness

ASIC Expectations

SOC 2 and ISO 27001 Compliance

Competitive Benchmarking

Next Steps: From Insight to Action

Immediate Actions (Next 4 Weeks)

Medium-Term Actions (Weeks 4-12)

Long-Term Actions (Months 3+)

Working with a Partner

Conclusion

Appendix: SQL Templates and Code Snippets

Template 1: Daily Delinquency Rate

Template 2: Vintage Cohort Loss Rate

Template 3: Unit Economics by Segment

Further Reading and Resources