Skip to content
Risk intelligence

A score you cannot explain is a liability, not an asset.

Scoring EnginesParametric and versionedBuilt on Sinergy

Scoring Engines are parametric financial and behavioural model families. They are deliberately separate from each other and from the credit workflow, so each view can be tuned, governed and audited for the portfolio it actually serves.

The risk problem

One number cannot carry two different questions.

Whether a business can afford an obligation and whether it has been behaving like a business that will meet one are different questions with different data behind them. Collapsing them into a single opaque score makes the model harder to improve and impossible to defend in a review.

What it does

Two families, and the inputs and controls they share.

Capacity and behaviour are different questions with different data behind them. Collapsing them into one score is what makes a model impossible to defend, so the figure keeps them apart.

Shared inputs
Alternative data
01Financial models

Measure capacity

Statements, ratios, cash cycle and sector parameters combine into an analysis, score and internal rating appropriate to the product and segment.

ControlsPolicy thresholds

Cut-offs, weights, overrides and segment rules are set and versioned by the risk team, not shipped in a release.

ControlsGovernance

The signals and weights behind a result stay attached to it, for investigation, internal audit and external review.

02Behavioural models

Keep the view current

Payment patterns, utilisation and operating signals update the picture continuously, so a rating reflects the last quarter rather than the last filing.

ControlsCalibration

Models are refit against what the group actually saw in its own book, in the segments it actually lends to.

How it runs

From raw input to a decision that can be replayed.

01Ingest

Financial statements, bureau data, transactional feeds and internal history resolve onto one counterparty.

02Derive

Features are computed and versioned, so a model always knows which definition it was trained on.

03Score

Financial and behavioural families run separately and produce distinct, labelled outputs.

04Apply

Credit policy turns the outputs into a limit, a price and a decision inside Credit Lifecycle.

05Observe

Realised performance flows back and becomes the training set for the next calibration.

Where it sits

Underneath the decision, not inside it.

Scoring produces the view. Credit Lifecycle decides what to do with it, and Early Warning watches what happens next.

01Data02Underwriting03Operate04Control05Structure06Distribute
Deployment

Usable without the rest of the platform.

Standalone

Scores into your own workflow

The engines can serve an existing origination system through defined interfaces, without adopting Credit Lifecycle.

Data

Your feeds, your residency

Model inputs, storage location and retention follow the institution's own data model and controls.

Review

Built for a model committee

Versions, training sets, performance and drift are recorded in a form a model risk function can actually review.

Questions

The ones that come up first.

Are these machine-learning models or rule-based?

Parametric: the structure and the parameters are explicit and set by the risk team, and every output exposes the signals and weights behind it. That is a deliberate constraint, because a result that cannot be reconstructed cannot be defended in a review.

Can we bring our own models?

Yes. The engines can run alongside an institution's existing models, and policy decides how the outputs are combined.

What data is required to start?

Financial statements and repayment history are enough for a first calibration. Transactional and behavioural feeds improve the behavioural family and are usually added second.

How often are models recalibrated?

On observed performance rather than a fixed calendar. The trigger, the training window and the approval are recorded with the version.

Scoring Engines

Bring a portfolio and the outcomes it produced.

The useful demo is a calibration on your own book.

Request a demo