Coreunda

The problem

Most risk tools give you a score. Not a reason.

A single blended ECL number is easy to report and impossible to defend. When a regulator asks why a segment's provisioning moved, or a declined customer asks why, "the model said so" is not an answer — and increasingly, it's not a legal one either.

Fair-lending obligations and the EU AI Act's explainability requirements are pushing credit decisioning toward the same standard good engineering already wants: show the work, not just the answer.

Reg B / ECOA, US

Adverse action notices require specific, accurate reasons — not a generic score.

EU AI Act, Art. 86

Credit scoring is named high-risk; affected individuals get a right to explanation.

IFRS 9 / CECL

Provisioning has to be auditable back to the scenario assumptions that produced it.

The approach

One calculation. One record. Every view drawn from it.

Coreunda never recomputes a number for a different audience. A single decision record — versioned, content-hashable, reproducible — feeds the internal dashboard, the regulatory report, and the explainability export alike.

01 — Portfolio & macro data

What you already have

Your loan-level portfolio, plus macro history from open sources — unemployment, GDP, inflation, and more where relevant. Governance decides which variables drive the model.

02 — Scenario engine

Five named futures, not one guess

A vector autoregression fit to your macro history, simulated forward by resampling real historical shocks — not an assumed bell curve. Worse, Bad, Base, Good, Better, each a full monthly path.

03 — Decision record

Every dollar traced to its scenario

PD × EAD × LGD, computed monthly, per contract, per scenario, then blended by governance-approved weights. The full breakdown is retained — not just the blended total.

04 — Behavioral signal

Each customer's actual repayment history

PD is seeded from a contract's real payment behavior, not just its profile at origination — and a customer's standing on one product informs the read on their others. Current on a mortgage while missing a card payment reads differently than current everywhere.

Transparency in practice

Every contract's loss curve, in the open.

Pick any contract, any month, any scenario, and see exactly what fed the number — the same view your team, your auditor, and eventually your customer's notice would draw from. No separate "explainability layer" bolted on after the fact.

Worse Base Better

Illustrative — the real thing is one chart among the full monthly, per-scenario breakdown Coreunda produces for every contract.

Built for

Two verticals underserved by generic ECL software.

Both run on the same transparent engine — what differs is the loss mechanics each product actually needs.

In development — first

Unsecured lending & BNPL

Fixed-installment and revolving credit lines with no collateral. Loss given default modeled as a recovery-rate blend sensitive to the same macro scenarios driving PD.

SME & trade finance

Adds a collateral-based LGD waterfall — cure, restructure, repossession — plus trade-specific macro drivers like FX volatility and trade volume.

Where the line sits

Coreunda supplies the evidence. You supply the judgment.

Complete, reproducible decision records for every contract — factors, weights, and the exact reason codes behind a scenario shift.
Jurisdiction-aware export templates for adverse-action and explainability requirements.
No claim to being your compliance department: your legal and compliance teams make the final regulatory call, the same way they already do with bureau data today.

Why this split works

A transparent parametric model — not an approximated black-box explanation — means exact factor decomposition is possible in the first place. Coreunda's job is to make sure nothing is hidden. Your compliance team's job is deciding what a given regulator needs to see.

The loss forecast that shows its work.