Underwriting examples must preserve the reason behind a decision—not just the score.

Start here

Does any of this sound familiar?

• Your lending team documents how application facts, policy rules, and reviewer judgment affect a credit decision.
• An adverse action requires an accurate explanation that is specific to the decision rather than a generic model label.
• You are considering AI support but need the institution to retain control over credit decisions and notices.

Underwriting examples are more than application rows paired with approve or decline. The trail can include policy version, data considered, exceptions, human review, reasons, notice, and correction.

Records contain sensitive financial information and may carry consumer-protection and contract duties. Define the AI task and authority to use each record; a plausible generated explanation is not proof of accuracy.

The problem

A score cannot stand in for a defensible explanation

An application may go to manual review because income documentation conflicts with a verified field. A reviewer checks policy and resolves the discrepancy. Without policy version, reason, and review stage, a dataset can confuse a preliminary recommendation with the actual decision.

Calling a model complex does not excuse inaccurate adverse-action reasons. Reuse also raises questions about applicant-data authority, representativeness, historical disparities, and consumer impact. Preserve the ability to reproduce and challenge decisions.

For credit decisions, a useful example includes the reason a reviewer can defend—not just the model’s answer.

The solution

Evaluate the decision trail, not only prediction accuracy

Define the lending task and obtain legal and fair-lending review before selecting or moving records.

DataSupply partners only with labs that meet its top 0.01% credibility standard. We help assess whether a qualified buyer may be a fit and negotiate terms that reflect the data's potential value, including exclusivity where relevant. We also help you work through diligence questions about rights, privacy, security, and compliance, then present a high-level inventory of permitted records, not the dataset. Fit is specific to each situation; no buyer or value is guaranteed.

What to inventory before any buyer conversation

• Choose a bounded use case Separate document organization from underwriting recommendations and final decisions. Define output users, what it may influence, escalation, prohibited uses, and criteria for human intervention.
• Retain the rationale and chronology For an authorized example, capture policy and model versions, input categories, decision stage, reviewer action, reason, notice, and correction. Keep approved, denied, withdrawn, incomplete, and unresolved cases distinct.
• Review explanation and fairness Test whether reasons are specific, accurate, and traceable. Reviewers should assess errors, inconsistent treatment, proxy effects, and missing or disputed data; record findings and approvals.

Set the boundaries before discussing access.

Obtain counsel, compliance, privacy, fair-lending, model-risk, and security approval. Check notices, contracts, bureau-data rights, purpose, and retention. Use least privilege, approved environments, deletion controls, and human accountability; de-identification alone does not settle rights or bias.

What could make a permitted example useful?

A focused evaluation may test a narrow administrative task or decision-trail quality; it is not a historical loan-book sale. Rights, representativeness, data quality, safeguards, and preparation determine feasibility. No payment is assured.

A practical first step.

Map one workflow and its reason codes without export. Ask legal and fair-lending teams to classify company, applicant, and restricted third-party data and identify authority needed.

datasupply.ai can discuss possible fit and buyer questions without receiving your dataset. You decide whether to pursue any introduction. No buyer, license, or payment is guaranteed.

Documented example / what it proves

The CFPB has stated that algorithmic complexity does not remove explanation duties

The CFPB’s Circular 2022-03 addresses adverse-action notices for algorithmic credit decisions. It states that ECOA and Regulation B require specific reasons and that the requirements apply regardless of technology; a creditor may not use an algorithm that prevents accurate reasons. Read Consumer Financial Protection Bureau, Circular 2022-03.

The circular supports decision-linked reasons but is not a license transaction, permission to reuse applications, or assurance that an explanation method meets every duty. Seek advice for the product and jurisdiction.

The important limit: The CFPB circular substantiates the explanation requirement described here; it does not establish a closed AI data license or authorize reuse of borrower information.