Definition
Alternative credit scoring assesses a borrower's creditworthiness without relying solely (or in addition) on traditional bureau scores — FICP in France, Schufa in Germany, FICO in the US.
It combines alternative data: AIS bank flows, BNPL payment history, digital behaviour, telecom data, company records, invoices. Enabled by PSD2 (which opened access to flows via AISPs), it can assess in a few seconds profiles that bureaus cover poorly.
The limits of traditional bureau scoring
In France, unlike the US or the UK, there is no positive score (FICO-style). The tools are negative: FICP (repayment incidents, over-indebtedness), FCC (cheque and card incidents) and each bank's own internal scoring. As a result, first-time borrowers, freelancers, newcomers and young people without a long history are invisible or penalised. Alternative scoring fills this gap.
The data used
- Bank flows (AIS): recurring income, fixed charges, average balances, savings rate, overdrafts, committed share of spending — the number-one source.
- BNPL / payment history: on-time repayments at Alma, Klarna, FLOA.
- Telecom data: payment regularity (mostly in Africa, rarer in Europe for GDPR reasons).
- Behavioural data: account age, number of banks, geographic stability.
- Self-declared: contract type, seniority, family situation.
- Company data (B2B): SIREN, age, sector, invoices, outstandings (Pappers, Heron Data).
The methodology: supervised ML
A model is typically a gradient boosting one (XGBoost, LightGBM), sometimes a neural network, trained on default history ("did the customer default at 12 months?"). It produces a score or probability of default, yielding a binary decision (threshold) or multi-tier one (rate by score). Explainability is mandatory for consumer credit (SHAP values, refusal reasons), under the right to explanation in Article 22 of the GDPR.
Major use cases
- BNPL (Klarna, Alma, FLOA, Younited Pay): instant scoring at checkout for fee-free 3x/4x instalments.
- Instant consumer credit: Younited, FLOA, Cetelem decide in 3 to 5 minutes, versus 3 to 5 days at a bank.
- SME / freelance business credit: Karmen, Defacto, October qualify unstable income via flows and invoices.
- Earned wage access (EWA): Stairwage, Rosaly estimate the employee's available cash.
- Embedded finance: scoring built into a merchant's checkout via API.
What alternative scoring is not
- Not a replacement for FICP: checking the FICP remains mandatory for consumer credit (Lagarde law); scoring complements it.
- Not FICO-style "positive scoring": France has refused a national positive register for 30 years; each player has its own proprietary model.
- Not exempt from the right to explanation: Article 22 of the GDPR requires explaining automated decisions and allowing human intervention.
- Not magic: the best models gain +5 to +15 Gini points over the traditional bureau — useful, not miraculous.
Quality criteria
- Gini / AUC: ability to separate good and bad payers (target Gini > 60% for solid consumer credit).
- Approval rate: share of accepted applicants, to be weighed against the default rate.
- Default rate: defaults at 12 months among the accepted (2 to 8% on BNPL).
- Latency: < 2 seconds for a checkout integration.
- Stability: resistance to drift over time (otherwise retraining).
In the PSD2 ecosystem
PSD2 was the catalyst for alternative scoring in Europe: before, only banks had the flows; with AIS, any lender can (with consent) retrieve 12 to 24 months of history. With FIDA (2027+), outstanding credits, savings and insurance will be added, for an even more accurate score.
Concrete examples
- B2C players: Algoan (a scoring building block sold to banks and fintechs), FLOA (BNP), Younited Credit, and the in-house models of Klarna and Alma.
- Younited: built its business on alternative scoring (AIS flows + self-declared data); a credit institution authorised by the ACPR since 2011, listed on Euronext Paris.
- Algoan: an SDK/API plugged into an AISP (Bridge, Tink) that returns a score and its reasons, sold to BNP, Cofidis and neobanks.
- Heron Data: B2B focus (US and UK SMEs), qualifying flows for credit, factoring and leasing.
- EWA: Rosaly and Stairwage score the employee on their salary flows to advance part of the net pay.
- GDPR / ethics limit: the use of behavioural data (geolocation, social networks) is highly contested in Europe; serious players stick to AIS, self-declared data and the bureau, explicitly consented.
- Evolution: use of LLMs for transaction context, models that are transparent by design, and pressure from the AI Act (2026), which classifies credit scoring as high-risk AI.