Your Scoring Models Are Drifting.
We Catch It Before Your Borrowers Default.
CreditVigil provides independent, continuous monitoring of credit and risk models — anywhere in the world. We detect population shifts, model degradation, calibration failure and disparate impact, and we estimate today's performance before the outcomes that would confirm it have matured.
Gini Coefficient
0.52 → 0.38 (-26.9 %)
Population Stability Index
⚠️PSI breach detected on Q4 2025 cohort — Gini dropped from 0.52 to 0.38
The Problem No One Is Solving
Most lenders are flying blind when it comes to credit model performance. Here's why.
Models Decay Silently
Credit scoring models lose predictive accuracy over time as borrower populations shift and macroeconomic conditions change. Without continuous monitoring, you won’t know until default rates spike.
Scoring Vendors Won’t Tell You
Your scoring vendor built the model — and profits from it. They have zero incentive to flag degradation or recommend a rebuild. You need an independent watchdog.
Spreadsheets Don’t Scale
Quarterly model reviews in Excel are slow, error-prone, and always out of date. By the time the report lands, deterioration has already cost you real money.
CreditVigil: The Check Engine Light
for Your Lending Portfolio
CreditVigil sits as an independent layer between your credit scoring models and your loan portfolio. We ingest performance data, run continuous statistical monitoring, and alert you the moment something drifts out of tolerance.
Think of us as the check-engine light on your dashboard. You don't need to be a mechanic to know something is wrong — you just need the warning in time to act.
Whether you use a single scorecard or a multi-model decisioning stack, CreditVigil gives you the independent oversight your risk committee and regulators demand.
Vintage Loss Curves
Track cumulative default rates by origination cohort. Spot early-warning divergence before it hits your bottom line.
Population Stability Index (PSI)
Detect when your applicant population has shifted away from the development sample your model was trained on.
Gini & KS Tracking
Monitor your model’s ability to separate good borrowers from bad with industry-standard rank-ordering metrics.
Roll Rate Migration
See loans flowing between delinquency buckets in real time. Identify cure-rate drops and delinquency escalation early.
WhatsApp Alerts
Critical threshold breaches delivered instantly to your phone via WhatsApp — the channel your team already lives on.
Board-Ready PDF Reports
Automated monthly performance reports with charts, commentary, and risk flags — ready to share with your board and regulators.
Three Steps. Full Visibility.
Getting started takes minutes, not months.
Connect Your Data
Upload loan-tape CSVs, connect via API, or grant read access to your database. We support all major formats and handle the heavy lifting of data validation.
We Monitor. Continuously.
Our analytics engine runs PSI, Gini, KS, vintage curves, roll rates, and more on every data refresh — flagging drift the moment it crosses your thresholds.
You Act. Before It’s Too Late.
Receive WhatsApp alerts, email digests, and board-ready reports. Know exactly what’s degrading, how fast, and what to do about it.
See What Your Risk Team Has Been Missing
Total Active Loans
127,847
Current Gini
0.38
PSI Score
0.31
30+ DPD Rate
8.2%
Gini & KS Trend (12 Months)
PSI by Month
Recent Alerts
Gini dropped below 0.40
2 hours ago
PSI exceeded 0.25 threshold
6 hours ago
Monthly report generated
1 day ago
Illustrative example using sample data. Your own metrics appear here after your first upload is processed.
What makes this different
Monitoring tools tell you what already happened
By the time a credit model's performance can be measured, the damage is eighteen months old. Everything below exists because that lag is the actual problem.
Performance you can see now, not in eighteen months
Credit outcomes mature over 6 to 24 months, so a measured Gini always describes a model that is no longer in production. We estimate current discrimination and default rate from the score distribution alone, before a single label arrives, and show the measured figure beside it as it catches up.
Confidence-based performance estimation, validated against held-out periods with labels withheld.
Champion versus challenger, done honestly
Two models scored on different populations cannot be compared, and most comparisons quietly do exactly that. We compare only on the accounts both models saw, with paired bootstrap intervals, and refuse to declare a winner where the difference is not significant.
Overlap-only comparison with paired confidence intervals and a minimum effect size before any recommendation.
Direct identifiers are kept out of analytical tables
Source files may contain identifiers during ingestion. Direct identifiers are detected by checksum rather than by column name, replaced with a per-tenant keyed HMAC when they are the subject key, and dropped from the mapping where they are sensitive. Analytical tables store the pseudonym and loan-performance fields. Clinical data arriving at a workspace not configured for it rejects the batch before analytical storage.
Identifier detection uses checksums — Luhn, IBAN, Verhoeff — so a loan reference is less likely to be mistaken for a passport number, and checksums can catch identifiers that column names miss.
Isolation enforced by the database, not by our code
Every table holding your data has row-level security keyed on your tenant, forced even for the table owner, and no application role holds a bypass. A bug in our application cannot show your book to another customer, because the database will not return the rows.
Internal automated checks mapped to SOC 2 Trust Services Criteria and HIPAA Security Rule technical controls — engineering evidence, not a certification.
Evidence, not screenshots
Every computation, access and configuration change lands in a hash-chained append-only trail that can be verified independently. Regulatory packs assemble the line items your supervisor expects — and a pack missing a required metric says so on its front page rather than filing short.
Regulatory pack templates are structured against SR 26-2 and EU AI Act Annex III item lists. That is documentation structure, not a compliance attestation.
Fits the way you already work
Any file format, any column names — mapping is inferred, not configured. Source IP allowlisting per credential, webhooks, Slack, Teams, email and WhatsApp for alerts, a documented API, and governed agent access for teams building on top of it.
CSV, Excel, Parquet and JSON in; API, webhook and scheduled report out.
Plans That Scale With Your Portfolio
Lite
For MFIs & SACCOs
- Gini & KS monitoring
- PSI drift detection
- Monthly PDF reports
- Email alerts
- Portfolio health dashboard
Pro
For digital lenders
- Everything in Lite
- Real-time monitoring dashboard
- PSI & CSI automated alerts
- WhatsApp & Slack alerts
- Multi-model comparison
- Vintage analysis
- Custom alert thresholds
Enterprise
For banks
- Everything in Pro
- Champion / challenger analysis
- Regulatory compliance reports
- Board-ready presentations
- Dedicated account manager
- Custom integrations
- On-premise deployment option
All plans include a free portfolio health check to get started. No credit card required.
Contact us for pricing details →Built for Lenders Everywhere
Digital Lenders
FairMoney, Carbon, Tala, Branch — you’re disbursing thousands of loans daily using AI scoring. We tell you when that AI starts making bad decisions.
Neobanks
Kuda, OPay, PalmPay — your overdraft and lending products need the same monitoring as dedicated lenders.
MFIs & SACCOs
170+ MFIs in Kenya, hundreds of SACCOs — you need board-ready reports without building a data team.
BNPL & Asset Finance
Lipa Later, M-Kopa — you have credit portfolios but don’t think of yourselves as lenders. You still need monitoring.
Banks with Digital Arms
NCBA, KCB, Co-op Bank — your digital lending channels disburse billions but monitoring is still manual.
Telcos with Embedded Credit
MTN MoMo, Safaricom — your mobile money platforms enable billions in credit through partners. Who’s watching the models?
Why Your Scoring Vendor Can't Monitor Themselves
Your Scoring Vendor
- Sells you the score AND judges its own performance
- No incentive to flag degradation
- Single-vendor view
- Monitors what’s convenient
- Reports on their terms
CreditVigil
- Independent — we don’t sell scores
- Vendor-neutral — monitors ANY model
- Compares models head-to-head
- Monitors what regulators need
- Alerts in real-time
No auditor should audit their own books. No scoring vendor should grade their own model. That's why CreditVigil exists.
Regulators Are Coming. Be Ready.
🇳🇬Nigeria
The CBN’s July 2025 Digital Lending Regulations require all licensed digital lenders to demonstrate ongoing credit model monitoring, including evidence of model validation, performance tracking, and drift detection. Lenders who can’t produce these reports risk license suspension.
🇰🇪Kenya
The CBK is actively licensing digital credit providers under the 2022 framework and demanding quarterly model performance reports. MFIs and SACCOs must now demonstrate to their boards that credit scoring models are performing as expected — or explain why they’re not.
🇿🇦South Africa
SARB’s model risk management expectations are tightening, aligned with international SR 11-7 standards. Banks and non-bank lenders using AI/ML models for credit decisions must maintain independent model validation — not self-assessments by the model vendor.
Find Out What Your Models Are Hiding
Send us a CSV of your loan portfolio and we'll deliver a comprehensive health check report within 72 hours. No commitment, no credit card, no sales pitch — just actionable insight into your scoring model performance.
- Vintage loss curve analysis
- Gini coefficient calculation
- Score band performance breakdown
- Delinquency roll rate snapshot
- Top 3 risk recommendations
Free Portfolio Health Check
Upload your portfolio data and receive a board-ready PDF report with vintage curves, Gini trends, and risk recommendations — completely free.
Request Your Free Health Check →Sensitive application data is protected with authenticated encryption, with transport security enforced by the deployed environment. Confidentiality agreements can be supported where appropriate. We've reviewed portfolios worth over $200M across 4 African markets.
Built by Credit Risk Practitioners, Not Just Engineers
CreditVigil is built by Ampersands AI — a team with deep experience in CECL provisioning, Basel capital frameworks, and loss forecasting across African lending markets. We understand the nuances of credit risk because we've lived them.