# Summarize a model validation for model risk management

A validation report summary covering conceptual soundness, data, performance, limitations, and findings with severity ratings.

- Topic: Data & Analysis
- Framework: Few-Shot
- Level: Standard
- Tags: Leadership, Write, Summarize, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/summarize-a-model-validation-for-model-risk-management

## Prompt

```text
Help me write a model validation summary for our model risk management function.

**Model:** [name, purpose - e.g. "PD scorecard for small business loans", "CECL loss model", "fraud detection", "AML transaction monitoring"]
**Model tier or risk rating:** [high / medium / low]
**Methodology:** [technique, key variables, vendor or in-house]
**Development data:** [period, sample, exclusions]
**Validation testing performed and results:** [PASTE - e.g. discrimination (AUC, Gini, KS), calibration, stability (PSI), back-testing, sensitivity, benchmarking]
**Use in decisions:** [where model outputs are used and any overrides]
**Issues found:** [LIST]

Write the summary in this structure (aligned with common model risk guidance such as SR 11-7 or equivalent):
1. Overall conclusion: fit for use / fit for use with conditions / not fit for use, with the reason in 2-3 sentences.
2. Conceptual soundness: whether the approach suits the purpose and the key assumptions.
3. Data quality and relevance, including whether the development period reflects current conditions.
4. Outcomes analysis: performance results against thresholds, in a table.
5. Limitations and compensating controls (overlays, overrides, monitoring).
6. Findings table: ID, description, severity (high/medium/low), required action, owner, due date.
7. Ongoing monitoring recommendations with metrics and triggers.
8. For credit decision models: whether fair lending or disparate impact testing was done, and gaps.

Use only the results I provide; mark missing tests [NOT PERFORMED] or [CONFIRM]. Do not overstate performance.
```

## What to fill in

- high / medium / low
- technique, key variables, vendor or in-house
- period, sample, exclusions
- where model outputs are used and any overrides
- LIST
- NOT PERFORMED
- CONFIRM

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.