AI Governance and
Model Transparency
Lending teams cannot adopt what they cannot explain. This page answers, plainly, how Voyager AI produces an output, who owns the decision, and what an examiner sees.
Six governance commitments
Each one is observable inside the product, not a statement of intent.
Field-level provenance
Every figure in a credit memo links back to the source document and page. Nothing appears in an output without a traceable origin.
Human-in-the-loop by design
The platform extracts, calculates, and drafts. The credit officer reviews, edits, and approves. No decision is issued without a human.
Examiner-ready audit trail
Extractions, calculations, edits, and approvals are recorded with user and timestamp as the work happens.
Your policy governs the model
Institution-specific credit policy, thresholds, and guardrails drive the output. Files outside those guardrails escalate to a credit officer.
SR 11-7 aligned governance
Model governance is aligned to SR 11-7, with ECOA and Regulation B alignment on decision-supporting outputs. SOC 2 Type I complete, Type II in progress.
Your data stays yours
Single-tenant, logically isolated environments. Customer data never trains shared or external models.
Who owns the decision
Voyager AI does not approve or decline credit. It reads the file, calculates what can be calculated, drafts what can be drafted, and hands a complete, sourced package to the credit officer. The officer edits and approves, and that approval is the decision.
When a file sits outside the institution's own guardrails, the platform escalates it rather than resolving it. That escalation path is where precision matters most, because the deals in the middle of the credit box are the ones a lending team most needs consistent, policy-driven analysis on.
For the platform-level security posture, encryption, and hosting model, see the Security page. For how the category is defined, see AI-Native Lending Intelligence.
Governance and transparency questions
How does Voyager AI explain a lending recommendation?
Every output carries attributable reasoning. Each extracted figure, calculation, and drafted narrative section links back to the source document and page it came from, so a credit officer or examiner can trace any number in a credit memo to the tax return, K-1, rent roll, or debt schedule it was read from.
Where does the human review step sit?
The architecture is human-in-the-loop by design. Voyager AI extracts, calculates, and drafts. The credit officer reviews, edits, and approves. No credit decision is issued by the platform. Files that fall outside the institution's own policy thresholds are escalated rather than auto-resolved.
What can a credit officer override?
Any extracted value, any calculated figure, and any drafted narrative section. Overrides are recorded with the original value, the new value, the user, and the timestamp, so the edit history is part of the record rather than a replacement for it.
What does the audit trail record?
Extractions, calculations, edits, and approvals. Each entry captures what changed, who changed it, when, and which source document supported it. The audit trail is produced as a byproduct of the workflow rather than reconstructed after the fact.
Which frameworks is Voyager AI aligned to?
Model governance is aligned to SR 11-7. Decision-supporting outputs are aligned to ECOA and Regulation B, including attributable reasoning for adverse-action support. A SOC 2 Type I audit is complete and a SOC 2 Type II audit is in progress.
Does customer data train external AI models?
No. Each institution runs in a single-tenant, logically isolated environment, and customer data is never used to train shared or external models.
How is model behavior kept consistent with our credit policy?
The institution's own credit policy, thresholds, and guardrails govern how outputs are produced. Policy documents are loaded into the Policy Library, and the platform applies those rules rather than a generic industry default. Changes to policy change the behavior of the system.
Can examiners review how the platform reached an output?
Yes. The audit trail is examiner-facing. For any credit memo, a reviewer can open the field-level provenance for each figure and see the source document, the calculation applied, and every human edit and approval on the file.
Walk the audit trail with us
Bring a real loan file. We will show you the provenance behind every figure in the memo.