A single API to score, explain, and approve SME loan applications. Three-model ML ensemble. 115 features. SHAP explainability. Sub-200ms decisions.
48 million Nigerian SMEs drive half the GDP. Yet 90% can't access formal credit. The infrastructure to assess them simply doesn't exist.
Four in five SMEs have zero formal credit history. Traditional scorecards assign them maximum risk by default — punishing the unbanked for being unbanked.
Each application costs lenders ₦50,000+ and takes weeks. Only 12% of SME applications reach a final decision. The rest simply abandon the process.
When SMEs are declined, they receive no explanation. No recourse. No path to approval. This erodes trust and violates emerging data protection mandates.
Every application flows through a deterministic pipeline — from raw data to explainable, auditable decision.
Bank statements via Open Banking (Mono, Okra), credit bureau reports (CRC, First Central, XDS), mobile money, POS data, and BVN/NIN/CAC verification.
7 Data CategoriesRaw data transforms into 115 predictive signals across the "5 C's of Credit" — capacity, capital, character, collateral, and conditions.
115 FeaturesLogistic Regression for interpretability. Random Forest for noise robustness. XGBoost for accuracy. Weighted consensus produces a single probability.
3-Model ConsensusProbability maps to a 0–1,000 YADEM Score with five risk bands (A–E), each with deterministic routing: auto-approve, manual review, or decline.
5 Risk BandsSHAP TreeExplainer decomposes every decision into additive feature contributions. Top factors surfaced in plain language for applicants and regulators.
NDPA CompliantKarma Blacklist, device fingerprinting, and velocity checks execute in parallel. A fraud flag overrides any favourable credit decision.
Parallel ExecutionScore, band, decision, explanation, fraud result, and recommended terms — delivered as a single JSON response in under 200ms.
<200ms E2EEvery component is architected for thin-file, cash-heavy, informal-economy borrowers that Western scoring systems ignore.
Three complementary models vote with optimized weights. No single point of model failure. AUC-ROC: 0.9928 on validation.
115 features derived from financial ratios, bureau trajectories, mobile money velocity, psychometric assessments, and sector signals.
Every decision decomposed into human-readable factor contributions. "Cashflow stability contributed +45 points to your score."
Karma Blacklist lookup, device fingerprinting, and application velocity analysis — running in parallel with scoring.
Consent management, data minimization, right-to-explanation, retention enforcement, and DPIA generation built in from day one.
RESTful endpoints with Swagger docs, API key auth, rate limiting, and consent middleware. Integrate in hours, not months.
Score thin-file borrowers using mobile money patterns, POS terminal data, e-commerce ratings, and utility payment consistency.
Disparate impact ratio, equal opportunity, and calibration checks across gender, region, and sector — enforced before every model deployment.
Six configurable business rules: CBN affordability test, sector risk overrides, active default blocks, and repeat borrower incentives.
Integrate YADEM into any lending platform with a single POST. Score, explanation, and fraud check returned together.
curl -X POST https://api.yadem.ai/v1/score \ -H "X-API-Key: ydm_live_..." \ -d { "bvn": "22345678901", "business_sector": "retail_fmcg", "business_age_months": 48, "requested_loan_amount_ngn": 2000000, "financial": { "avg_monthly_revenue_6m": 850000, "cashflow_volatility_6m": 0.15 } }
Validated on synthetic Nigerian SME data. Production metrics will be published after pilot deployment.
Join our private beta. Integrate the YADEM engine and start approving applications in minutes, not weeks.
Request Early Access →