Creditworthiness scoring: declarative, open banking or statements, how to choose

In short: there are three ways to assess a customer's creditworthiness: declarative data (fast but unverifiable, and now insufficient under CCD2), open banking (real banking data with consent, the benchmark) and the upload of OCR-processed statements (the same analysis, for the 30 to 40% of customers who refuse the bank connection). The right solution combines the last two to cover 100% of customers.**

With the CCD2 directive, applicable in November 2026, creditworthiness assessment based on verifiable data becomes mandatory for all consumer credit, BNPL included. The self-declaration accepted yesterday exposes you to an inspection by your national supervisor tomorrow (ACPR in France, BaFin in Germany, DNB in the Netherlands). Here is how to compare the approaches, and the criteria for choosing the tool.

The three approaches, honestly compared

• Data. Declarative: What the customer declares · Open banking (PSD2): Real banking flows, at the source · Statement upload (OB-like): OCR-processed PDF/photo statements.
• Reliability. Declarative: Unverifiable, falsifiable · Open banking (PSD2): Maximum · Statement upload (OB-like): Equivalent after extraction.
• CCD2 compliance. Declarative: no, insufficient on its own · Open banking (PSD2): yes · Statement upload (OB-like): yes.
• Customer friction. Declarative: Low · Open banking (PSD2): Connection consent required · Statement upload (OB-like): Simple file upload.
• Coverage. Declarative: 100% · Open banking (PSD2): ~60-70% (connection refusals) · Statement upload (OB-like): The remaining 30-40%.
• Fraud. Declarative: High (fake supporting documents) · Open banking (PSD2): Near zero · Statement upload (OB-like): Forgery detection required.

The conclusion is self-evident: open banking by default, statement upload as the alternative, declarative data for pre-qualification only. That leaves choosing the tool. Five criteria.

1. Customer coverage: what happens when the connection is refused?

This is the most underestimated criterion. 30 to 40% of customers refuse to connect their bank account. If your solution stops there, you lose those files or you fall back on declarative data, and therefore outside CCD2. Demand an alternative of equal reliability, known as open banking-like: statement upload (PDF or photo), OCR extraction, and the same scoring engine as open banking, for a consistent, auditable decision.

Question to ask: "A customer refuses the bank connection: what journey do you offer them, and is the scoring identical?"

2. Depth of analysis, not just a number

A balance tells you nothing. The analysis must categorise transactions (salaries, rent, existing loans), calculate the decision indicators (disposable income, debt ratio), and spot warning signs: rejected direct debits, gambling, repeated micro-loans, a recently opened account, international transfers. And detect the hidden multiple loans that declarative data never shows.

Question to ask: "Show me the full output for a file: categorisation, disposable income, alerts."

3. CCD2 compliance and traceability

CCD2 requires an assessment based on verifiable data and the retention of evidence (3 years in case of inspection). Your solution must produce a complete audit trail: which data, which calculation, which decision, on which date. A score without exportable justification does not protect you in an inspection by the supervisor, and the AI Act adds the requirement for model explainability.

Question to ask: "In case of a supervisory inspection of a file from 2027, what exactly can I produce?"

4. Adaptation to your business

A generic score treats a €15,000 consumer loan, an €80-a-month energy subscription and a pay-in-4 BNPL the same way. Yet the predictive signals differ by sector. Check whether the model can be adapted or retrained on your target and your data, and whether it covers B2B if your customers are businesses.

Question to ask: "Can the model be retrained on my portfolio, and with what results at comparable players?"

5. Monitoring over time, not only at granting

Creditworthiness is a moving picture. The best solutions maintain access to the data (with consent) for several months after granting, to detect signs of fragility early, such as repeated rejections, a drop in incoming flows, a chronic overdraft, and act before the default rather than suffer it. This is also the logic behind the preventive forbearance expected by regulators.

Question to ask: "After granting, can you monitor how the account evolves and alert me before the first incident?"

How Meelo meets these 5 criteria

Meelo combines Open Banking (PSD2 aggregation across multiple aggregators) and OB Like: the customer who refuses the connection uploads their last three statements, which are OCR-processed and then run through the same scoring engine, for 100% customer coverage. The creditworthiness analysis output includes a categorised statement, disposable income, debt ratio and warning signs; account access is maintained for six months for early detection. Scores can be adapted by sector (consumer credit, BNPL, energy, with payment defaults halved and 88% of good customers retained), a complete audit trail, explainable AI. Decision in 2 to 5 seconds. To see the output on a real file, you can schedule a demo.

FAQ

How do you assess a customer's creditworthiness?

By analysing their real banking flows: recurring income, fixed expenses, existing loans, payment incidents, disposable income. Open banking (with consent) or the analysis of uploaded statements replaces declarative data, which is unverifiable and falsifiable.

What does the CCD2 directive require for consumer credit?

From November 2026, a mandatory creditworthiness assessment based on verifiable data before any consumer credit (BNPL included), with retention of evidence. The customer's self-declaration is no longer enough in case of inspection.

What should you do when a customer refuses to connect their bank account?

Offer them the option to upload their statements (PDF or photo): OCR-processed and then analysed by the same engine as open banking, they provide an assessment of equal reliability. This is the so-called "open banking-like" approach.

Does creditworthiness scoring work for businesses?

Yes: analysing business banking flows and financial data (balance sheets, payment behaviour) makes it possible to assess the default risk of a B2B customer, alongside business verification (KYB).

‍