When I moved countries, my financial life crossed the border faster than my financial history.
That experience is one reason I started Goscore. A person can have income, habits, obligations, and years of responsible financial behaviour—and still arrive in a new market looking strangely empty to the systems expected to assess them.
Open banking offered a practical possibility: with consent, current transaction data could help build a more relevant picture.
But access to more data is not the same as a better decision.
The data is rich. The meaning is not automatic.
A bank transaction has an amount, a date, text, and context that may be incomplete. Turning it into “income,” “rent,” “gambling,” “subscription,” or “financial stress” requires interpretation.
That interpretation can be useful. It can also be wrong in very ordinary ways.
A transfer from a family member may look like income. A reimbursed business expense may look like spending. A shared household account may describe two people as one. A seasonal worker may look unstable when the pattern is entirely normal for the job.
The model does not know the person. It sees evidence prepared through several technical and commercial layers.
This is why I become nervous when a fintech presentation jumps from “we have transaction data” to “we understand affordability.” There is a lot of product work in that arrow.
Start with the decision boundary
Before choosing signals, I ask what the system is allowed to decide.
Is it verifying a stated income? Prioritising a manual review? Detecting a possible inconsistency? Estimating affordability? Making or materially influencing a lending decision?
The closer we get to a consequential decision, the more important governance, data quality, fair treatment, explanation, and human review become.
The European Banking Authority’s loan origination and monitoring guidelines connect creditworthiness assessment with governance, consumer protection, and prudent credit risk. That is the right frame. The product is not a score floating in space. It is part of how an institution grants and monitors credit.
The EBA has also examined creditworthiness practices among non-bank lenders, including the use of alternative data. More information may improve an assessment, but it also expands the questions around relevance, soundness, and consumer protection.
Explainability begins before the model
Teams often treat explainability as a feature added after a model is trained: create a reason code, write a sentence, show a chart.
I think it begins earlier.
Can we explain why this input belongs in the decision? Can we show how the category was produced? Can a user correct a classification? Can the credit team understand what changed the result? Can the company defend the signal as relevant rather than merely predictive?
A technically explainable model can still support an inexplicable product.
The strongest credit systems I have seen do not try to remove judgement from the room. They help people use evidence consistently, reveal uncertainty, and preserve the path from input to outcome.
Inclusion is an outcome, not a headline
It is tempting to describe alternative data as automatically inclusive. Sometimes it can help people who are poorly served by traditional files. That possibility matters enormously to me.
But inclusion should be measured.
Who receives a better outcome? Who becomes newly visible? Who is excluded by data access, account type, language, income pattern, or the way categories are defined? Does the system reduce unnecessary rejection, or simply create a more detailed reason to reject the same people?
These are not objections to innovation. They are the work required to make the innovation real.
For a new financial-data product, I would test at least four layers:
- Data reality: availability, consent, coverage, latency, and messy descriptions.
- Signal validity: stability, relevance, leakage, bias, and behaviour across customer groups.
- Decision value: improvement over the current process, not over an imaginary baseline.
- Customer consequence: explanation, correction, recourse, and the cost of being wrong.
Open banking is infrastructure. Trust is the product.
The European Commission’s work on PSD3 and the Payment Services Regulation aims, among other things, to improve the functioning of open banking while protecting consumers and addressing fraud. The infrastructure continues to mature.
That gives fintechs more room to build. It does not decide what deserves to be built.
The opportunity is not to collect the maximum amount of financial data. It is to use the minimum evidence needed to make a decision more current, more understandable, and more responsible.
Open banking can help a lender see a person who was previously invisible.
The hard part is making sure the system sees them accurately.