The questions we get asked
They come from risk, credit and IT leadership. The answers are direct, including when the answer is no.
Frequently asked questions
What is NEXCLARA, in one sentence?
A decision layer that reads the data your institution already holds, computes a credit risk, explains its recommendation factor by factor, and returns the decision to your systems. It replaces none of your tools: it connects to them.
Is NEXCLARA a credit bureau?
No, and that is a structural position. NEXCLARA builds no credit database for its own account. The credit bureau supplies the data; NEXCLARA produces the decision from data you legitimately hold. No prior agreement with a bureau is needed to start.
Does NEXCLARA lend money?
No. NEXCLARA grants no credit and carries no credit risk. It is a technology provider, which places it outside the scope of heavy financial licensing.
Who makes the final decision?
Your credit officer, within their delegation limit. NEXCLARA produces a substantiated recommendation; it never binds the institution. Any departure from the recommendation requires written grounds, recorded and statistically usable — and that gap is in fact the richest indicator in the system.
Does the platform suit a microfinance institution as well as a bank?
The same engine serves both, with different configuration. A bank integrates NEXCLARA with its core banking system and targets portfolio performance; a microfinance institution uses assisted entry, with no technical integration, and targets clients it could not assess at all.
We have no statistical history. Is that a blocker?
No, and it is a major design point. Most statistical approaches need thousands of historical files. NEXCLARA starts from an expert scorecard that states your real credit policy, calibrated with your teams. The learned model then takes over, gradually, under your control.
Who decides the share of the statistical model?
Your risk department, from the interface, with no development. The weighting between expert scorecard and learned model is a versioned setting, simulable before activation, and reversible.
What happens when a value is missing?
It is flagged as missing, never replaced by an average. Substituting an average manufactures information that does not exist. Absence weighs on the confidence level, and a file that is too incomplete cannot produce an automatic approval: it is routed to review.
How does your explainability differ from others?
It is computed in the same act as the decision, not rebuilt afterwards by a second model. An explanation computed separately can diverge from the decision it claims to justify; this one structurally cannot.
Can a decision taken two years ago be reconstructed?
Yes. Every decision references the exact policy, scorecard and model versions that produced it, and the chain can be replayed identically with the data of that day. It is an audit and litigation requirement, not a convenience.
What happens if a data source is unavailable?
The decision is produced in degraded mode, explicitly flagged as such, stating what was missing. No external source is blocking: the system neither stops nor invents.
How much work for our IT team?
It depends on the mode, and one of them requires none: assisted entry works with no technical integration. For the others, we hand over a complete integration contract — specification, formats with examples, field mapping, error codes, and a test environment populated with synthetic data.
Do we have to change our core banking system?
No. NEXCLARA connects to what exists, including older systems. Adapters translate into an internal pivot model; no adapter contains decision logic, and adding a system does not change the engine.
Does the decision come back into our systems?
Yes, through the same pivot model in reverse: synchronous response, outbound notification, return file, query interface, analytical export, signed decision document. A decision that does not return forces double entry, and double entry is the leading cause of a decision tool being abandoned.
What if delivery of the decision fails?
The decision is kept. The mechanism includes a persistent queue, spaced retries, a consultable failure log and manual replay. A decision that has been produced is never lost because its delivery failed.
Where is our data hosted?
Wherever your security policy requires. Four modes exist on one codebase: shared and dedicated at NEXCLARA on certified European hosting, on your premises in your own data centre or private cloud, or hybrid — processing on your side, supervision with no personal data at all.
Is deployment on our premises genuinely possible?
Yes, and it is a planned, documented and tested mode, not a negotiated exception. The architecture uses no proprietary service beyond standard components, and operation on a closed network is verified automatically on every change.
How do you guarantee one institution cannot see another's data?
Isolation is applied at three levels — database, application and audit — and no query executes without an institution context. In dedicated mode, isolation becomes physical rather than merely logical.
What about the Banking Commission submission requirement?
It applies, and we arrive with the file assembled: responsibility matrix, data location, subcontracting chain, security arrangements, continuity plan, audit rights, reversibility plan, service levels, incident notification.
How do we leave if we decide to stop?
With your data, in standard formats, through a documented and tested procedure. The reversibility plan is the piece a regulator scrutinises most and the one most often missing from a vendor. Leaving is not a negotiation.
Can the audit log be altered?
No. The log is append-only with cryptographic chaining, and no application path — administrator included — allows an entry to be changed or deleted. Its integrity is verified, and a break raises an alert.
Do you use artificial intelligence?
A statistical model intervenes in one of the five computation stages, weighted as you choose, and always alongside the expert scorecard that serves as guardrail and as reference for explanation. NEXCLARA is not an “AI company”: the product is the decision capability; learning is one mechanism within it.
How do you address algorithmic bias?
Through full explainability, which makes every contribution visible and therefore contestable; through human supervision of the final decision; and through tracking the human override rate, which reveals a calibration flaw or a blind spot in the data long before it becomes systemic.
How is consent managed?
It is traced by purpose and by data source. Alternative sources, when enabled, each require a specific, traced and verifiable consent. The consent chain must be documented and defensible: that is a requirement of the applicable framework, not an option.
Who is the data controller?
The client institution. NEXCLARA acts as a technical processor, and that qualification is materialised in the architecture, not only in the contract.