Cashflow Risk Model
This document outlines industry-researched risk scoring models for borrowers with connected Shopify, Stripe, and Square accounts. Phase 1+ will incorporate these cashflow-based signals alongside social trust scoring.
Why Cashflow Data?
Traditional credit scores exclude millions of small businesses. Platform data from Shopify, Stripe, and Square provides real-time visibility into business health that traditional lenders lack.
Key advantages:
Real revenue verification vs. self-reported financials
Continuous monitoring vs. point-in-time snapshots
Behavioral signals like refund rates, chargebacks, seasonality
Lower cost than manual underwriting
"Shopify has proprietary transaction data allowing for pre-qualification of merchants, high visibility into cash flows for underwriting." [80]
Industry Models We're Learning From
1. FICO SBSS (Small Business Scoring Service)
The industry standard for SBA loans, scoring 0-300.
Score Components:
Personal credit (most predictive factor)
Business credit bureau data (D&B, Experian)
Revenue trends and profitability
Cash flow and debt-to-income ratio
Business age and payment history
Liens and judgments
Thresholds:
211-300
Excellent
Preferred
191-210
Good
Eligible
161-190
Fair
May qualify
0-160
Poor
Usually rejected
SBA requires minimum 160 SBSS, most lenders want 180+. [Source: Nav FICO SBSS Guide]
2. Kabbage / OnDeck Model
Fintech pioneers using alternative data for instant underwriting.
Kabbage examines 300+ data points including:
Monthly revenue cycles
Customer review scores (Yelp, etc.)
Shipping volumes and patterns
Accounting software data
E-commerce sales trends
Social media activity
OnDeck's proprietary OnDeck Score:
Real-time bank transaction analysis
Accounting software integration
24-hour lending decisions
Lines up to $250K
"SMBs provide information including accounting records, bank accounts, e-commerce revenues, shipping data to enable Kabbage to produce a lending decision." [Source: Business Model Zoo]
3. E-Commerce RBF Model (Clearco/Wayflyer)
Revenue-based financing specifically for e-commerce.
Core Underwriting Dimensions:
Revenue Quality
Stability, predictability, recurring %
Stable revenue = reliable repayment
Growth Trajectory
Month-over-month trends
Sustainable growth vs. unsustainable
Margins
Gross and net profit margins
Higher margins = more resilient
Customer Base
Concentration, repeat rates
Diversified = less risky
Industry
Vertical-specific benchmarks
Context for metrics
Typical Terms:
Funding: $10K - $20M
Fee: 6-19% flat (based on risk)
Payback: % of daily/weekly revenue
Clearco charges 6.5% to 19% depending on amount and repayment timeline. Wayflyer charges 2-8% based on funding amount and business performance.
4. Academic ML Models
Research shows machine learning significantly outperforms traditional scoring.
Best Performing Features (from PMC research):
Loan annuity-to-credit ratio
External credit scores
Social network default status (peers who defaulted)
Regional economic ratings
Address consistency across records
Model Performance:
LightGBM
0.7936
XGBoost
0.7892
CatBoost
0.7890
"Alternative data consistently achieved higher AUC scores across all tested algorithms." Removing alternative variables degraded performance by 4-5 percentage points. [Source: PMC Research]
Risk Signals from Platform Data
Data Currently Available
Stripe
Total revenue, charge count, success rate, MRR (subscriptions), average charge
Square
Total revenue, payment count, refund rate, success rate, average payment
Shopify
Total revenue, order count, average order value, shop metadata
Tier 1: Direct Risk Signals
These metrics directly indicate risk and are already collectible:
Chargeback Rate
Stripe
<0.5%
>1% high risk, >1.5% reject
Refund Rate
Square
<3%
>5% elevated risk
Payment Success Rate
Stripe/Square
>97%
<95% payment issues
MRR % of Revenue
Stripe
Higher = better
<10% = volatile
AOV Consistency
Shopify
Low variance
High variance = unstable
"For most industries, any chargeback rate above 1% means a business might be deemed high-risk. Mastercard fines businesses with chargeback rate of 1.5% or higher." [Source: Stripe]
Tier 2: Derived Metrics
Calculated from raw data for deeper insight:
Revenue Concentration
Top 10% customers / total revenue
>50% = high customer risk
Seasonality Index
StdDev of monthly revenue / mean
High = harder to predict
Customer Retention
Repeat orders / total orders
Low = churn issues
Revenue Velocity
Week-over-week % change
Declining = warning
Days Since Sale
Gap from last transaction
Growing gap = trouble
Gross Margin Proxy
(Revenue - Refunds) / Revenue
Low = tight margins
Tier 3: Behavioral Signals
Patterns that indicate operational health:
Transaction Regularity
Consistent daily/weekly patterns = stable operations
Growth Sustainability
>50% month-over-month may be unsustainable
Platform Tenure
Longer history = more reliable data
Multi-Platform Consistency
Similar revenue across platforms = trustworthy
Implemented Business Health Score Model
Based on FinRegLab research findings, we've implemented a four-component scoring model that prioritizes cash flow stability over absolute revenue amounts. This aligns with research showing volatility is the strongest predictor of default.
Weighted Formula
Why these weights? See FinRegLab Research for the research basis.
Component Breakdown
1. Revenue Stability (35% weight)
The strongest predictor per FinRegLab research. Measures month-over-month revenue consistency using Coefficient of Variation (CV).
How It Works:
Group all orders by calendar month
Sum revenue per month to create a time series:
[$8,200, $9,100, $7,800, ...]Calculate CV:
(standard deviation / mean) × 100Lower CV = more stable = higher score
Example Calculation:
What It Measures:
Predictability of cash flow for repayment planning
Resilience to seasonal/market fluctuations
Business model sustainability
< 15%
100
Excellent
Very predictable revenue
15-25%
85
Strong
Minor month-to-month variation
25-40%
70
Good
Normal business fluctuations
40-60%
50
Fair
Noticeable revenue swings
60-80%
30
Weak
Volatile cash flow
≥ 80%
15
Poor
Highly unpredictable
Requires 3+ months of data. Limited data defaults to 40 pts (Fair).
Why This Is Weighted Highest (35%): FinRegLab's study of 38,000+ small business loans found balance volatility (a cash flow stability measure) was the single strongest predictor of loan default.
2. Order Consistency (25% weight)
Transaction frequency and regularity. JPMorgan Chase Institute research shows businesses with steady transaction patterns have higher survival rates.
How It Works:
Group all orders by week (Sunday-Saturday boundaries)
Count orders per week to create a time series:
[12, 15, 11, 14, 13, ...]Calculate CV of weekly order counts
Example Calculation:
What It Measures:
Regular customer demand vs. sporadic sales
Operational consistency (fulfillment capacity)
Business model predictability
< 20%
100
Excellent
Very predictable weekly volume
20-35%
85
Strong
Minor week-to-week variation
35-50%
70
Good
Normal seasonal/promotional effects
50-70%
50
Fair
Noticeable demand swings
70-90%
30
Weak
Unpredictable order flow
≥ 90%
15
Poor
Highly irregular (feast or famine)
Requires 4+ weeks of data. Limited data defaults to 40 pts (Fair).
Why This Matters: A business with 50 orders one week and 5 the next is harder to underwrite than one with steady 25-30 orders weekly, even if total volume is similar.
3. Business Tenure (20% weight)
Track record matters, but less than combined cash flow metrics. Calculated from the date of first verified order.
36+
100
3+ years
24-35
85
2+ years
12-23
70
1+ year
6-11
50
6+ months
3-5
30
< 6 months
< 3
15
Very New
4. Growth Trend (20% weight)
Future capacity indicator. Measures momentum by comparing the first half of the data period to the second half.
How It Works:
Find the actual data span (first order date to last order date)
Split at the midpoint of the actual data range
Sum revenue in each half
Calculate growth rate:
((recent - prior) / prior) × 100
Example Calculation:
Why We Use Actual Data Midpoint: The comparison is based on when orders actually exist, not an arbitrary time window. This ensures both halves contain meaningful data even if the business is new.
Why Moderate Growth Scores Highest:
10-30% growth is sustainable and indicates healthy demand
50%+ growth may be unsustainable (flash sales, one-time orders)
Extreme growth often precedes corrections
Edge Cases:
Less than 45 days of order history: Score 40 (Fair - insufficient data)
Zero prior revenue but has recent sales: Score 60 (new business with traction)
Zero revenue in both periods: Score 40 (Fair - insufficient data)
+10% to +30%
100
Healthy Growth
Sustainable momentum
+30% to +50%
85
Fast Growth
Good but watch for volatility
0% to +10%
75
Stable
Mature, predictable business
+50% or more
60
May be volatile
Could be unsustainable spike
0% to -10%
50
Minor Decline
Seasonal or temporary dip
-10% to -25%
30
Declining
Concerning trend
Below -25%
15
Significant Decline
Business may be struggling
Requires 45+ days of order history. Limited data defaults to 40 pts (Fair).
Privacy-Safe Display Labels:
≥30%
"Accelerating"
10-30%
"Growing"
0-10%
"Stable"
-10% to 0%
"Slight decline"
<-10%
"Declining"
Privacy-First Display
We show qualitative tiers instead of exact numbers:
$8,500/month revenue
"Revenue: Strong"
Exact figures are sensitive
180 orders/month
"Orders: Steady"
Protects competitive info
+12% growth
"Trend: Growing"
Qualitative is sufficient
36 months active
"Tenure: 3+ years"
Ranges work equally well
Component Tier Labels
85-100
Excellent
70-84
Strong
55-69
Good
40-54
Fair
25-39
Weak
0-24
Poor
Loan Affordability (Second Indicator)
The Business Health Score measures how healthy a business is, but it doesn't answer a critical question: Can this business afford this specific loan?
A business with excellent stability could still be requesting 10x their monthly revenue—that's risky regardless of their health score. Rather than combining these into a single score (where good health could mask dangerous loan size), we display them as two separate indicators.
Why Two Indicators?
The Problem with Additive Scoring: If we combined health and affordability into one score, a business with:
Excellent stability (35 pts)
Great order consistency (25 pts)
Long tenure (20 pts)
Healthy growth (20 pts)
...would score 100/100 even if requesting a loan equal to 6 months of revenue. That's misleading.
The Solution: Display two independent signals, similar to how Kiva shows both "borrower trustworthiness" and "field partner risk" separately.
Loan Affordability Tiers
Based on Loan-to-Revenue Ratio = Loan Amount ÷ Average Monthly Revenue
Comfortable
< 0.5x
Loan is less than 2 weeks of revenue
Manageable
0.5x - 1x
Loan is less than 1 month of revenue
Stretched
1x - 2x
Loan equals 1-2 months of revenue
High Burden
> 2x
Loan exceeds 2 months of revenue
Privacy-Safe Display
We show relative sizing, not exact revenue:
< 0.25x
"< 1 week revenue"
0.25x - 0.5x
"~1-2 weeks revenue"
0.5x - 1x
"~2-4 weeks revenue"
1x - 2x
"~1-2 months revenue"
> 2x
"> 2 months revenue"
Example Displays
Scenario 1: Healthy business, reasonable loan
Scenario 2: Healthy business, large loan
Scenario 3: Newer business, small loan
This allows lenders to make informed decisions. A Grade A business with "High Burden" affordability is a different risk profile than a Grade A business with "Comfortable" affordability.
Risk Grades and Funding Terms
Grade Mapping
80-100
A
Low
Strong revenue, stable, quality metrics
65-79
B
Moderate-Low
Good fundamentals, minor concerns
50-64
C
Moderate
Acceptable with conditions
40-49
D
Elevated
Requires monitoring
<40
E/HR
High/Reject
Insufficient data or high risk
Funding Parameters by Grade
A
5-6x monthly
6-8%
8-10%
B
4-5x monthly
8-10%
10-12%
C
3-4x monthly
10-14%
12-15%
D
2-3x monthly
14-18%
15-18%
E/HR
Not eligible
-
-
Conditions by Grade
A
Standard terms
B
Quarterly data refresh
C
Monthly data refresh, revenue verification
D
Weekly monitoring, personal guarantee may be required
Implementation Phases
Phase 0 (Current)
Social trust scoring only
Gather baseline repayment data
No cashflow scoring required
Phase 1 (Next)
Add optional platform connections
Display credit scores to borrowers
Show scores to lenders (informational)
Continue gathering repayment correlation data
Phase 2 (Future)
Mandatory platform connection for larger loans
Risk-adjusted funding limits
Automated underwriting decisions
Blend social trust + cashflow scores
Key Sources
Industry & Platform Models
Academic Research
Risk Thresholds
Fintech Cost & Performance
Next: Social Trust Scoring | Risk & Defaults | Lender Warnings
Last updated