What we build
A decisioning stack that scales with your policy — and earns the regulator's trust.
Each capability is a production component — not a proof-of-concept — wired into your stack, documented for your risk committee, and monitored continuously.
Multi-source data ingest
Bureaus, core banking, bank statements, payroll, collateral, ID, and identity-network APIs — pulled in parallel, with fallbacks when a source is slow or missing.
Feature engineering that doesn’t stop at raw data
DSR, LTV, cash-flow stability, utilisation trajectory, income pattern, aging curves, payment velocity — the features risk models actually need, computed before scoring.
Ensemble scoring across model families
Credit, affordability, fraud, and behavioural models run in parallel and are blended into a single explainable score — with per-model contribution logged on every decision.
Policy + tier + pricing in one engine
Exclusions, concentration limits, tier mapping, risk-based pricing — applied consistently, versioned, and overridable through a controlled policy workflow.
Reason codes, audit trail, customer-facing explanations
Every decision produces top-feature reason codes, a full model trail for analysts, and plain-language explanations in multiple languages — ready for regulators and applicants.
Drift monitoring + scheduled retraining
Feature drift, population stability, and portfolio performance monitored continuously. Scheduled retraining on a cadence or triggered by threshold breach.
Lending products we decision for
One engine, every lending motion.
Same ensemble scoring, policy engine, and explainability layer — tuned per product line. Shared features, per-product models, portfolio-level analytics. Every product listed below runs on the same stack; only the model weights, policy rules, and integrations change.
SME lending
Working-capital, term loans, invoice financing, and trade finance for small and mid-market businesses. Ensemble scoring across bureau + banking + industry risk.
Personal loans
Unsecured consumer lending — payroll, debt-consolidation, education. Affordability-first models with payment-history signals and income-stability features.
Mortgages
Secured home loans with LTV-aware decisioning, collateral valuation, affordability stress-testing, and regulator-ready audit trails for each approval.
Auto finance
New and used vehicle financing — dealer-channel integrations, collateral OCR, residual-value modelling, and segment-specific risk layers.
BNPL
Short-tenor buy-now-pay-later decisioning in under 2 seconds — network-level fraud checks, lightweight affordability, and real-time merchant-side signals.
Credit cards
Consumer and business card issuance — utilisation-aware scoring, rewards-tier decisioning, and dynamic credit-limit reviews on the same engine.
Model families we deploy
No single model covers every decision. So we ensemble.
Each model family covers a distinct risk — blending their outputs into a single score gives you coverage, resilience, and transparency a scorecard alone can't match.
Gradient-boosted tree ensembles combining bureau data with alternative signals — cash flow, utility payments, mobile-wallet behaviour. Calibrated to your portfolio, not a global template.
DSR, income-stability, and expense modelling combined with regulatory policy caps. Blends deterministic rules (hard floors) with learned patterns (stability, volatility).
Real-time signals — device fingerprint, IP velocity, application network graph — detect synthetic identities, ring-fraud, and application stacking.
Applicant trajectory vs similar-cohort history — early-warning for deterioration and favourable bias for improving cohorts. Useful for thin-file applicants.
Data sources wired into every model
Every signal that moves the decision — integrated.
Pulled in parallel, normalised into a single feature schema, versioned alongside the model that consumes them.
Explainability, not just predictions
A score alone doesn't pass compliance. A trail does.
Every approve, refer, or decline is accompanied by top-feature reason codes, full feature and model-version provenance, and a customer-facing explanation — generated at decision time, indexed for audit, and available in the languages your market speaks.
- SHAP-style top-feature contributions per decision
- Full feature + model-version provenance logged
- Customer-facing explanations (multi-language)
- Aligned to MAS FEAT, SR 11-7, local equivalents
Compliance & model governance
Built to pass model risk review — not just to ship.
Regulator-ready from day one. Delivery includes documentation, back-testing, drift monitoring, and governance workflows your risk committee, internal audit, and external regulator will all want to see.
Model Risk Management (MRM)
Aligned to SR 11-7, OCC 2011-12, and MAS FEAT — formal model inventory, tiered review cadence, independent validation before production.
Fair-lending + outcome equity
Disparate-impact analysis and proxy-feature review on every deploy. ECOA-style testing in the US; fairness metrics aligned to MAS FEAT principles in SG.
Ongoing monitoring
PSI, CSI, and feature-level drift tracked continuously. Portfolio performance vs. expected outcomes reported monthly. Threshold-based alerts wake the risk team.
Champion / challenger
Every new model ships as a challenger alongside the champion. Traffic-split control, shadow-book comparison, and one-click rollback via the control plane.
Back-testing + validation
Out-of-time back-tests across economic regimes, stability over vintages, stress scenarios aligned to your ICAAP inputs. Full validation report per model version.
Reg-reporting ready
IFRS 9 ECL model inputs, Basel PD/LGD/EAD exports, MAS FEAT transparency disclosures. Pre-formatted outputs your treasury and compliance teams can file without rework.
Why Axccelerate for risk decisioning
Not a scorecard.
A decisioning stack.
A scorecard gives you one score. Our stack gives you ensemble scoring, policy orchestration, audit trails, drift monitoring — the infrastructure a real lender actually needs.
Pricing
Priced to the product line, not the applicant volume.
Risk deployments are custom — we scope against your policy, products, and integrations before quoting.
Glossary
The vocabulary behind every decision.
A quick reference for the acronyms that show up in risk decisioning — the terms your risk team, regulator, and model documentation will all use.
- DSR
- Debt-service ratio
Monthly debt obligations as a fraction of income. Most lending policies cap DSR between 0.55 and 0.65 for unsecured consumer credit, lower for SMEs.
- LTV
- Loan-to-value ratio
Loan amount divided by collateral value. Lower LTV = lower risk; mortgages typically sit below 0.80, secured SME lending varies by collateral class.
- PSI
- Population Stability Index
Measures how much an input-feature distribution has shifted between training and production. PSI > 0.25 usually triggers re-investigation or retraining.
- CSI
- Characteristic Stability Index
Per-feature version of PSI — flags exactly which variables have drifted. Useful for pinpointing the cause of a model-performance decline.
- SHAP
- SHapley Additive exPlanations
Game-theoretic method that assigns each feature a signed contribution to an individual prediction. The backbone of our per-decision reason-code generation.
- KYC / KYB
- Know Your Customer / Know Your Business
Identity verification and risk assessment performed on individual applicants (KYC) or corporate entities including UBO discovery (KYB).
- PEP
- Politically Exposed Person
Applicants whose public role or close associates trigger heightened AML screening and enhanced due diligence under FATF guidance.
- AML / CTF
- Anti-Money-Laundering / Counter-Terrorism Financing
The regulatory regime governing sanctions screening, transaction monitoring, and suspicious-activity reporting across financial services.
- FEAT
- Fairness · Ethics · Accountability · Transparency
The Monetary Authority of Singapore's principles for the use of AI and data analytics in financial services. Drives our explainability and governance defaults.
- IFRS 9 ECL
- Expected Credit Loss
Accounting standard requiring lenders to provision for credit losses using forward-looking inputs — PD, LGD, EAD — across 3 staging buckets.
- STP
- Straight-Through Processing
Applications approved or declined automatically with no manual review — the main KPI for decisioning efficiency. Typical healthy ranges: 65-85% depending on product.
- MRM
- Model Risk Management
Formal discipline for identifying, assessing, monitoring, and controlling the risks that arise from using models. SR 11-7 is the canonical US reference.
- ICAAP
- Internal Capital Adequacy Assessment Process
A bank's own evaluation of the capital it needs, typically submitted annually to the regulator. Our risk outputs feed the PD/LGD inputs to this process.
- PD / LGD / EAD
- Probability of Default · Loss Given Default · Exposure at Default
The Basel risk parameters used to compute capital requirements and IFRS 9 provisions. Our models produce them directly or feed the feature store that derives them.
Your decisioning chain, engineered.
30-minute scoping with a senior engineer and a risk-systems operator. You'll leave with a model plan, integration sketch, and realistic timeline — not a sales pitch.