Analytical Models
This page documents the data structures produced by Elysium Nexus analytical engines. These are computed outputs, not raw data imports.
Business Health Index (BHI)
Proprietary five-component weighted composite score (0-100) measuring entity financial health.
Components: Liquidity, Volatility, Revenue Stability, Expense Ratio, Payment Behavior.
Each component draws from multiple financial inputs (balance ratios, transaction patterns, revenue trends, expense growth, payment timeliness) and is weighted according to Elysium's proprietary model.
Bands: critical (0-34), poor (35-49), fair (50-64), good (65-79), excellent (80-100).
API output includes: overall score, band, per-component scores, trend direction, seasonal adjustment, confidence level, and scoring method.
Fraud Signals
Multi-signal detection engine operating on configurable time windows.
Signal categories:
- Velocity anomalies — Unusual spikes in transaction count or value
- Behavioral pattern shifts — Dormancy-to-activity transitions
- Structuring detection — Patterns consistent with threshold evasion
- Concentration analysis — Unusual value distribution
- Flow analysis — Pass-through and layering pattern detection
- Peer deviation — Statistical outlier detection against peer cohort
Each signal produces explainability metadata: contributing factors, mitigating factors, and evidence items. Overall alert severity is computed from triggered signals.
Risk Scores
Composite risk score (0-100) combining multiple proprietary components including inverse health scoring, churn probability, default probability, and volatility assessment.
Risk bands: low, medium, high, critical.
Churn and default probabilities are computed via proprietary statistical models with documented coefficients available under NDA.
Cashflow Forecast
Forward-looking weekly cash position projections with configurable confidence intervals.
Outputs:
- Weekly projected inflows, outflows, net flows, and closing balances
- Confidence intervals (90%, 95%, or 99%)
- Minimum projected balance and timing
- Shortfall probability
- Working capital gap detection
- Risk flags
Method: Trend extrapolation with seasonal decomposition and volatility-adjusted confidence bands.
Markov Credit State Model
Four-state discrete Markov chain for credit migration analysis.
States: Healthy, Watch, Critical, Default — mapped from BHI score ranges.
Outputs: State predictions at 1, 3, 6, and 12-month horizons; steady-state distribution; expected time-to-default; top at-risk customers; trajectory classification (improving/stable/deteriorating).
The model uses Bayesian posterior estimation with proprietary prior parameters calibrated for financial services.
Next-Best-Action Recommendations
Rule-based recommendation engine generating prioritized actions for relationship managers.
Action types include: credit line adjustments, cash runway warnings, churn risk retention offers, upsell opportunities, payment reminders, FX hedge recommendations, failed payment recovery, and revenue decline alerts.
Key properties:
- All credit, retention, and treasury actions require human approval
- Cash runway warnings and payment reminders are auto-sent (no financial action triggered)
- Actions are ranked by priority and confidence
- ML re-ranking is available for improved personalization
Peer Benchmarking
Percentile ranking across multiple financial metrics against an anonymized peer cohort.
Metrics assessed: Business health, burn rate, liquidity, revenue growth, payment behavior, and cash runway.
Performance labels: top_quartile, above_average, average, below_average, bottom_quartile.
Minimum peer cohort size is enforced for statistical significance.
Seasonal Adjustment
Industry-specific monthly adjustment profiles that compensate for expected seasonal variations, preventing false-negative health signals during lean periods.
Industries covered include agriculture, retail/commerce, hospitality/tourism, construction, services/technology, manufacturing, and others relevant to the deployment market.