CREDIT DECISIONS
Credit scoring: explain every decision, factor by factor
Credit scoring with TIMi estimates the probability that a borrower repays as agreed, and breaks each score down variable by variable, so a decision can be explained and documented.
Repayment propensity
Modeler compares borrowers who repaid as agreed with those who did not.
White-box by design
Every score is broken down variable by variable, for one applicant.
On your infrastructure
Applicant data stays on your own servers.
HOW IT WORKS
How credit scoring works.
Three steps, in one engine, on a standard server with no cluster.
1. Prepare the data
Anatella gathers the applicant’s financial and declared data, the repayment history available and relevant behavioral data, then structures them into variables.
2. Build the model
Modeler compares the profile of the applicant with that of previous borrowers who repaid as agreed or not, and estimates a probability of repayment. AutoML tests several approaches and keeps the one with the best balance between accuracy and interpretability.
3. Score and explain
Each applicant gets a score with the factors that explain it variable by variable, such as debt level, income regularity or credit history.
GETTING IT RIGHT
What a credit scoring model needs to hold up.
A score is only useful if it can be defended to the applicant and documented for internal or regulatory control.
Explain each decision individually
Every decision can be broken down factor by factor for that applicant, not only described for the population as a whole.
Check which variables the model relies on
The white-box report lists every variable the model uses, so your compliance team can verify that none of them is a variable you are not allowed to use for credit decisions.
Validate with the metrics risk teams know
Modeler reports the lift curve, and Anatella can produce ROC and KS curves on a scored dataset, two measures common in credit risk analysis.
EXPLAINABILITY
A score gives a level of risk. The factors give the reasons.
A score alone tells you a level of risk, not what justifies it. In a field where decisions must be explained to the applicant and documented for control, a white-box model breaks the score down factor by factor.
For credit and risk teams
A decision can be justified factor by factor instead of being presented as a number without detail.
For transparency and compliance
Automated credit decisions are an area where transparency rules apply, such as GDPR and the AI Act. A variable-by-variable explanation gives you the documentation, and the regulatory assessment stays with your compliance team.
For model owners
Modeler’s built-in Drift Report shows when a model loses accuracy, so you update it when needed instead of on a fixed calendar.
DATA
What data a credit scoring model uses.
Anatella connects to more than 80 sources, so this data can be gathered without prior custom development.
Financial data
Income, expenses and debt level.
Credit and repayment history
When available.
Declared data
Professional situation and seniority.
Past decisions and outcomes
Repayment results already observed, as a learning base.
PROOF
Same models, applied to risk.
We do not yet have a published customer story dedicated to credit scoring. These deployments apply the same models to risk in banking and insurance.
98% accuracy on business bankruptcy within 6 months
Built from public balance sheets, with 24,000 variables including many year-over-year evolutions.
€250 million more could be invested safely
An individual-level model of cash-out probability showed that AXA could invest an additional €250 million without increasing liquidity risk.
FAQ
Common questions.
Is a white-box model as accurate as a black-box model?
Accuracy depends on the available data, not only on the type of model. Modeler tests several approaches and keeps the one with the best balance between accuracy and interpretability, without sacrificing one for the other on principle.
Can the model justify an individual decision to an applicant?
Yes, this is one of the benefits of a white-box model: each decision can be broken down factor by factor for that applicant, not only for the general population.
Do we need a history of payment defaults?
Yes, a history of repayments and defaults is needed to train the model. Without it, a provisional score based on business rules can come before the predictive model.
Does it only apply to consumer credit?
No. The same principle applies to business credit or other financial commitments, with variables adapted to the context.
RELATED USE CASES
Keep exploring.
SEE FOR YOURSELF
Test it on your own data.
Run TIMi on your own data with a free trial, or talk to us about your case.
