# Building a system that assembles and checks a credit application, and explains a decision inside a bank

Source: https://www.bankingnewsai.com/agentic-banking/build/credit-underwriting
Last updated: Sep 17, 2026

Credit is the use case with the clearest rulebook and the least room for autonomy. The scoring model itself is a model in the model-risk sense and stays under validation; what the newer patterns add is everything around it: assembling the file from documents and systems, checking it for completeness, drafting the decision memo and producing the reasons a declined applicant is owed. The design that survives an exam keeps the decision with a person or a validated model, and uses the language model to prepare and explain, never to decide on its own.

## The brief (default answers)

| Choice | Answer |
| --- | --- |
| Who is affected | Customers, directly |
| Reversibility | Not really |
| Stakes | High |
| Knowledge | Both |
| Verifiability | By judgment |
| Steps | Known |
| Jurisdiction | United States |
| Delivery route | Through a cloud platform |

- **Pattern:** Workflow — Several model calls orchestrated by your code, in a sequence or a graph you designed.
- **Tier:** 1, Tier 1: a person decides — High stakes, hard to undo, customer-facing. The model prepares, drafts and checks; a person decides and acts, and the record shows it.
- **Workflow shapes:** Parallelization, Evaluator and optimizer
- **Human involvement:** The model prepares, drafts and checks. A person decides and acts, and the record shows who.
- **Knowledge:** Retrieval over the governed document set, with a citation on every answer and a refusal when nothing relevant is found.; Tool calls to the system of record for anything live. Never retrieval for a balance, a case status or a limit.; Documents to reason with, systems to check against: the workflow decides which question goes where before the model answers.

## Decomposition

| Step | Owner | Note |
| --- | --- | --- |
| Collect and classify documents | model | Extraction against a schema, with a confidence per field. |
| Verify income, identity and existing exposure | system | Tool calls to the core, the bureau and the KYC store; never retrieval. |
| Score and price | system | The validated scoring model, outside the language model entirely. |
| Draft the decision memo | model | From the file and the score; every number traceable to its source. |
| Decide | human | The underwriter, with the memo; the record shows the reviewer. |
| Adverse-action notice | system | Reasons generated from the decision record and checked against the required list. |

## Control layer weights (0–4)

| Layer | Weight |
| --- | --- |
| governance | 1 |
| identity | 1 |
| actions | 0.9 |
| data | 1 |
| models | 1 |
| runtime | 0.8 |
| observability | 1 |
| oversight | 1 |

## Documents that apply

- [ECOA / Regulation B adverse action (15 U.S.C. 1691(d); 12 CFR 1002.9)](https://www.bankingnewsai.com/ai-regulation/documents/cfpb-ecoa-regulation-b-adverse-action) — Specific principal reasons for any adverse credit action, whatever the model.
- [FCRA adverse action and credit-score disclosures (15 U.S.C. 1681m, 1681g(f))](https://www.bankingnewsai.com/ai-regulation/documents/cfpb-fcra-adverse-action-key-factors) — Key factors behind a credit score must be disclosed.
- [SR 26-2](https://www.bankingnewsai.com/ai-regulation/documents/fed-sr-26-2) — A credit model is a model: validation and monitoring scaled to materiality.
- [Regulation (EU) 2024/1689](https://www.bankingnewsai.com/ai-regulation/documents/eu-ai-act-regulation-2024-1689) — Credit scoring of natural persons is high-risk in the EU (Annex III 5(b)).
- [EBA Guidelines on loan origination and monitoring (EBA/GL/2020/06)](https://www.bankingnewsai.com/ai-regulation/documents/eba-gl-2020-06-loan-origination-monitoring) — EU rulebook for automated creditworthiness models; staff must be able to override.
- [CFPB Chatbots in Consumer Finance (issue spotlight, 2023)](https://www.bankingnewsai.com/ai-regulation/documents/cfpb-chatbots-in-consumer-finance-2023) — Customer-facing AI must not block access to a person or give wrong answers about rights.
- [SR 23-4](https://www.bankingnewsai.com/ai-regulation/documents/fed-sr-23-4) — US third-party risk management, including the model provider.
- [SB 26-189](https://www.bankingnewsai.com/ai-regulation/documents/co-sb26-189) — Colorado: notice, explanation and human review for consequential automated decisions from Jan 1, 2027.
- [BCBS Third-Party Risk Principles (Dec 2025)](https://www.bankingnewsai.com/ai-regulation/documents/bcbs-third-party-risk-principles-2025) — Nth-party supply chains and concentration on cloud providers.
- [BCBS ICT Risk Management Report (June 2026)](https://www.bankingnewsai.com/ai-regulation/documents/bcbs-ict-risk-management-range-of-practices-2026) — How supervisors look at ICT and cloud dependencies.
- [OCC Bulletin 2026-13](https://www.bankingnewsai.com/ai-regulation/documents/occ-bulletin-2026-13) — For a national bank, the OCC's copy of the 2026 model-risk guidance.
- [NIST AI RMF 1.0](https://www.bankingnewsai.com/ai-regulation/documents/nist-ai-100-1) — The voluntary Govern, Map, Measure, Manage frame for everything model-risk guidance leaves out.

## Evals

- A golden dataset of at least 300 real cases with expected outputs, including adversarial inputs: wrong documents, unusual formats, prompts that try to change the task.
- A judge model scoring against a written rubric (accuracy, completeness, tone, citation present), calibrated against a human-scored sample every month. Threshold set from the human sample, not guessed.
- Citation checks: every factual claim resolves to a passage in the governed set; unsupported claims below 2% of answers.
- Escalation evals: the cases that must reach a person do, on a held-out set, with precision and recall both reported.
- Trace evals per step, not only end to end: which step fails, how often, at what cost, so a prompt or model change can be judged step by step.
- The same suite reruns on every prompt change, model version change and retrieval change; a regression blocks the release. That is what ongoing monitoring and outcomes analysis mean in model-risk terms.

## Human gates

- A named accountable executive signs off the use case before build, and the inventory records the tier.
- A person reviews every output before it reaches a customer or a system of record; the review is logged with the reviewer's identity.
- Independent validation before launch and outcomes analysis after it, on the model-risk schedule for a material model.
- Adverse outcomes to a customer carry the specific reasons the law requires, produced from the decision record, not reconstructed afterwards.
- A kill switch and a rollback path, tested before launch.
- The customer is told they are dealing with an AI system and can reach a person in one step.
- Adverse-action explainability is tested as a launch gate: if the system cannot produce specific principal reasons, it does not decide.

## What an examiner will ask

- Where is this system in your inventory, what tier did you assign, and who signed it off?
- What counts as a model here, and what does your validation cover for the parts that are not?
- Show me the data lineage behind the retrieval set and the training or tuning data.
- What can the system do without a person, and where is that written down?
- How do you know it is still working: which evals run, how often, and what happened the last time one failed?
- What did you do about the vendor: due diligence, contract, exit plan, concentration?
- Walk me through one wrong output from production and what the customer, if any, saw.
- Who can switch it off, and has that been tested?
- Produce the adverse-action reasons for this declined applicant from the decision record.
- Show me how a customer reaches a person, and how long it took the last ten who tried.

## What the board should hear

- What it does: assembles and checks a credit application, and explains a decision; the model prepares, drafts and checks. a person decides and acts, and the record shows who.
- Pattern: workflow (parallelization, evaluator and optimizer); tier 1: a person decides.
- Rules it answers to: 4 documents across US, each linked in the brief.
- How we know it works: a golden dataset, automated checks on every release, and human review at the level the tier demands.
- What could go wrong and who answers: the accountable owner, the kill switch, the escalation route.

## Pitfalls

- Letting the language model influence the score, which makes the scoring model unvalidatable.
- Reconstructing adverse-action reasons after the fact instead of from the decision record.
- Forgetting that a fraud-driven decline is still an adverse action.

Change any answer on the interactive map: https://www.bankingnewsai.com/agentic-banking/build/credit-underwriting

---

Canonical page: https://www.bankingnewsai.com/agentic-banking/build/credit-underwriting
Part of [BankingNewsAI](https://www.bankingnewsai.com/) — a free daily brief on AI in banking, an AI regulation tracker (19 authorities, 166 documents) and AI-strategy profiles of the 100 largest US banks. Markdown versions of every reference page: append `.md` to the page URL; index at https://www.bankingnewsai.com/llms.txt.