The layer underneath every page on this site.
Goscore turns consented financial data into explainable features, and delivers them into the decision systems a lender already runs. No core replacement, no change of record system, and the bank keeps policy and the final decision.
From a consented account to a decision your systems can act on.
Each stage below is a boundary a review can inspect: what enters it, what leaves it, and what is retained.
Consent and access
The applicant consents to share account data under PSD2 / open banking. Goscore reads what the consent covers, for the purpose it was given, and nothing else. Traditional credit information continues to come from the sources you already use.
Normalisation
Raw transactions arrive as inconsistent free text across banks. They are cleaned, deduplicated and given a consistent structure before anything is inferred from them.
Categorisation
Merchants are resolved from free-text strings and transactions are categorised — income, essential spend, obligations, discretionary, transfers — using a taxonomy that means the same thing across banks and over time.
Feature layer
Categorised activity becomes stable, named features: verified income and its stability, recurring obligations, disposable income, affordability at a given instalment, and behavioural risk indicators.
Decision support
Features become a risk indication with the drivers that produced it, plus the evidence pack behind it. This is an input to your decision, not a decision.
Delivery
One API response per applicant, a batch file for a historical backtest, or a standalone review flow where no integration exists yet.
Three ways in, none of which move your core.
Backtest
A file of historical applications with the outcomes you already know. Nothing connects to a live system, and this is where almost every engagement starts.
API
A request per applicant from your existing decision flow, returning features, indication and reason drivers. Deployed as a challenger, a second look after a decline, or an additional layer.
Standalone flow
A hosted consent and review flow for teams that want evidence before an integration project exists. Useful for underwriting queues and pilots.
A decision you can reconstruct afterwards.
- Reason drivers
- Every indication carries the signals that produced it, in language a customer-facing team can use.
- Decision trace
- Source → feature → threshold → outcome, retained so a decision can be reconstructed after the fact.
- Human review
- Nothing here removes a reviewer. The evidence pack exists to make human judgment faster, not to replace it.
- Bank policy
- Thresholds, cut-offs and the final decision remain yours. Goscore does not set credit policy.
- Scope of data
- Features are derived for the purpose the consent was given for. Consent can be withdrawn, and the applicant is told what was used.
Regulated in Norway, and written down.
Goscore AS is an authorised credit reference agency supervised by Datatilsynet, registered in Norway (company no. 922707324). The formal commitments — legal basis, sharing and processors, transfers outside the EEA, retention, security and automated decision-making — are set out in the privacy policy rather than summarised here.
What we answer in writing, before a test.
- 01 Deployment model and hosting region, and which options exist for data residency.
- 02 Which aggregators and data sources are used in your market, and their coverage.
- 03 What is retained, for how long, and how deletion is evidenced.
- 04 Access control, logging and separation between environments.
- 05 Sub-processors, and the data processing agreement that governs them.
- 06 Model governance: versioning, monitoring, drift, and how changes are released.
- 07 How a historical backtest is performed, and what leaves your building to run it.
- 08 Incident handling and notification, and who is accountable for what.
These are the questions a bank’s security, data and model-governance functions ask. We answer them per engagement, in writing, before any data moves — rather than publishing a claim here that your reviewers cannot test.
Start with two weeks and a sample.
A data proof shows you the categorisation, the features and the trace on data you recognise — before anyone talks about integration.
Scope a 2-week data proof →