# Building a system that tests models and decisions for disparate treatment and impact, and documents the results inside a bank

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

Fair-lending work is analytical and evidentiary: the outputs are tests, comparisons and documentation that will be read by an examiner or a court. That makes it a workflow with an evaluator, not an agent. Models help by drafting the analysis plan, explaining a model's features in plain language and assembling the file; the statistical tests run in code, and a person owns the conclusion.

## The brief (default answers)

| Choice | Answer |
| --- | --- |
| Who is affected | Customers, via a person |
| Reversibility | With cost |
| Stakes | High |
| Knowledge | Both |
| Verifiability | By judgment |
| Steps | Known |
| Jurisdiction | United States |
| Delivery route | Provider API |

- **Pattern:** Workflow — Several model calls orchestrated by your code, in a sequence or a graph you designed.
- **Tier:** 2, Tier 2: act with approval — Material but recoverable. The model may prepare and, within limits, act, with a person approving anything that leaves the bank or touches a customer.
- **Workflow shapes:** Prompt chaining, Evaluator and optimizer
- **Human involvement:** The model prepares and may act within limits; a person approves anything that reaches a customer or cannot be undone.
- **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 |
| --- | --- | --- |
| Define protected classes and the comparison design | human | Counsel and compliance set the frame. |
| Run disparate-impact tests | system | Deterministic statistics on the decision data. |
| Explain features and interactions | model | Plain-language explanations of what the tests found, cited to the outputs. |
| Draft the memo | model | From the results; nothing asserted beyond them. |
| Conclude and remediate | human | Signed, with the actions taken. |

## Control layer weights (0–4)

| Layer | Weight |
| --- | --- |
| governance | 1 |
| identity | 1 |
| actions | 0.9 |
| data | 1 |
| models | 0.55 |
| runtime | 0.8 |
| observability | 0.7 |
| oversight | 0.6 |

## Documents that apply

- [Regulation B final rule on disparate impact (April 2026)](https://www.bankingnewsai.com/ai-regulation/documents/cfpb-regulation-b-final-rule-2026) — The current US position on disparate impact under Regulation B.
- [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) — Adverse-action reasons are also the audit trail for discrimination testing.
- [Joint Statement on Automated Systems (CFPB, DOJ, EEOC, FTC)](https://www.bankingnewsai.com/ai-regulation/documents/cfpb-joint-statement-automated-systems-2023) — Four agencies on automated systems and discrimination law.
- [Regulation (EU) 2024/1689](https://www.bankingnewsai.com/ai-regulation/documents/eu-ai-act-regulation-2024-1689) — Bias examination is a data-governance duty for high-risk systems (Article 10).
- [SR 26-2](https://www.bankingnewsai.com/ai-regulation/documents/fed-sr-26-2) — US model risk management as revised in April 2026.
- [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.
- [FSB AI monitoring report (Oct 2025)](https://www.bankingnewsai.com/ai-regulation/documents/fsb-monitoring-ai-adoption-vulnerabilities-2025) — Concentration on a small number of model suppliers.
- [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 150 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

- Per-action approval by a competent reviewer for anything customer-facing or irreversible; sampled review for the rest.
- Validation proportionate to materiality, with monitoring for drift on inputs and outputs.
- An escalation route to a person that the customer can reach in one step.
- Monthly review of the evals and the exception log by the accountable owner.
- 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.

## What the board should hear

- What it does: tests models and decisions for disparate treatment and impact, and documents the results; the model prepares and may act within limits; a person approves anything that reaches a customer or cannot be undone.
- Pattern: workflow (prompt chaining, evaluator and optimizer); tier 2: act with approval.
- 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

- Using a model to run or interpret the statistics without the code path to reproduce them.
- Drift in the bias tests after a model change with no rerun.
- Treating an explainability tool's output as the legal explanation.

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

---

Canonical page: https://www.bankingnewsai.com/agentic-banking/build/fair-lending
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.