i.
Supervised internal ratings
Years mapping the ECB-supervised internal-ratings approach at EU D-SIB banks — frameworks vetted by the world's most demanding supervisor.
EUR 100bn+ of loan pools · decades of through-cycle data
The problem
Without a reference point, risk is invisible.
The earth turns at a thousand miles an hour and nobody in the room feels it. Private credit sits in the same chair: returns feel about right and risk feels contained only because there is nothing to measure them against. The absence of a benchmark does not remove risk. It hides it.
3½ years
in development with an ECB-supervised DACH bank of EUR 100bn+, with input from one of the largest global financial sponsors
500m+
data points behind the rating agency credit default models used for calibration
2006
the year the independent challenge framework entered institutional use — tested through 2008, 2011 and Covid-19
≤ 1 notch
from regulator-approved IRB ratings for most of a live portfolio; nearly all within two
The triangulation
Three roads. One grade.
Each road was built independently. KINN19 is where they converge — every output tested, outliers flagged, rating dispersion explained. Validation is not a feature of the engine; it is the engine.
ii.
Rating agency models
Systematic calibration against a rating agency rating engine and credit default models in use by insurers for risk-based capital.
500m+ data points · multiple credit cycles
iii.
A crisis-tested framework
An acquired quantitative rating framework in institutional use since 2006 challenges every output the engine produces.
Proven through 2008, 2011 and Covid-19
Where three independent methods converge on the same grade, the output stops being an opinion.
How it works
AI reads. Rules decide.
From source document to benchmark rating, every step is visible. Identical inputs produce identical outputs, and every grade traces back to the page it came from — the guarantee we give every supervisor.
Ingest
AIFinancial statements and loan documents — PDF, Word, Excel, scanned — in any language.
Extract
AI32 financial fields recognised and cross-checked against hundreds of thousands of data points.
Reconcile
RulesRating-agency definitions and analyst-grade adjustments, applied as versioned deterministic rules.
Compute
ModelsPD, LGD and expected loss — structure-aware across seniority, collateral and covenants.
Benchmark
OutputMapped to the familiar external scale, with a Transparency File for every loan.
Only possible now.
A year ago this could not have been built. Reading unit-level financials and loan documentation reliably, at portfolio scale, in any format, has only become possible with the most recent advances in document AI. KINN19 puts that intelligence where it belongs — reading — and leaves the rating to rules a regulator can interrogate. Leading AI and credit experts and a leading university are joining the programme.
Data sovereignty
Your data never really leaves.
Validation without exposure. KINN19 validates the risk without ever holding the identity — the data belongs to the institution, cryptographically and not just contractually.
Inside the institution
Every exposure is anonymised and encrypted at source; each real name becomes a dummy code. The mapping table never leaves.
KINN19 validation
Validation, benchmarking and the challenge layer run only on coded, anonymised data — no names, no identifying fields.
Back inside the perimeter
Results return under the same codes and map to real exposures only inside the institution's own systems.
Encrypted, dummy-coded validation flows arrive with the year-end release.
Private beta
Now running with early adopters.
Over the coming six months KINN19 is running with a selected group of European banks and insurers. We are admitting a small number of further institutions. See it in forty-five minutes, on twenty to fifty of your own loans, under dummy codes.
Zam Khan · [email protected] Roger Thomson · [email protected]