1. Auto approve / continue
Passes eligibility, affordability and risk rules, then continues to contract/disbursement checks.
LendGo platform
Decision engine demo
Shows the ideal automated lending flow: approve clear low-risk applications, decline hard-stop applications early, and send only edge cases to review.
2. Auto decline
Hard stops such as no consent, underage, confirmed tamper or strong compliance hit stop the journey.
3. Edge-case review
Borderline, missing or policy-sensitive cases are the exception path while thresholds are finalised.
Current result
Auto approve / continue
Account AUTO-APPROVE is processed through transparent rules. Every stop, continue or review decision has visible reason codes for audit and customer-service follow-up.
Affordability
R 8 500
Existing instalments
R 2 500
Arrears
R 0
How automation works
Low-cost gates run first, then richer checks are used only when the applicant is still eligible.
1. Intake
Consent, identity, amount and declared income.
2. Evidence
Bank, statement, fraud and bureau inputs.
3. Features
Affordability, arrears, debt and behaviour fields.
4. Decision
Approve, decline or route edge case.
5. Audit
Reason codes, event log and reviewer notes.
Feature ratios
Flags are visible and explainable.
Loan-to-affordability
Review line: 0,293
Instalment-to-income
Review line: 2,308
Debt-to-income
Review line: 5,472
Arrears-to-income
Review line: 3,058
Business value
Decision proof that can become delivery logic
The demo shows how each application can be routed immediately, while keeping the evidence and reason codes needed for audit, customer support and later model improvement.
Version
Proof
Sample rows
5
Hard-stop rules
0No flags in this group.
High-risk reasons
0No flags in this group.
Edge-case reasons
0No flags in this group.
Validation limits
Proof result, not a final production ML claim.
AUC
0.875
Gini
0.750
KS
0.750
Rows
5
These metrics prove the mechanics only. Production validation needs a larger labelled repayment dataset with paid, paid late, failed collection, defaulted and settled outcomes.
Next data needed
Needed before reliable trained-model accuracy.
Paid on time
Paid late
Failed collection
Defaulted or written off