# Explain what drives a model's predictions to non-technical stakeholders

Translates feature importance, SHAP values, and model behavior into a plain-English explanation leaders can trust and act on.

- Topic: Data & Analysis
- Framework: Few-Shot
- Level: Standard
- Tags: Leadership, Sales, Explain & Teach, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/explain-what-drives-a-model-s-predictions-to-non-technical

## Prompt

```text
Help me explain what drives my model's predictions to a non-technical audience.

**What the model does and how it is used:** [e.g. "scores leads weekly; sales calls the top 20%"]
**Model type:** [e.g. gradient boosted trees, logistic regression]
**Performance in plain terms:** [e.g. "top 20% of scores contain 55% of conversions"]
**Interpretability outputs:** [PASTE feature importance, SHAP summary, coefficients, partial dependence descriptions, or example explanations for individual predictions]
**Audience and their likely concerns:** [e.g. "sales leaders who think the model ignores relationships"]

Produce:
1. **What the model does** in 2-3 sentences, with no jargon.
2. **The top drivers** in plain English, with direction ("customers who logged in fewer than twice in the last month are much more likely to cancel"). Group technical features into business concepts.
3. **Two or three worked examples** of individual predictions explained simply.
4. **How good it is**, framed against the current approach or a baseline, and how often it will be wrong.
5. **What it does not mean:** clarify that drivers are associations, not proven causes, unless causal work was done.
6. **Limits and safeguards:** where the model is weaker (segments, new customers), and when a human should override it.
7. **Likely questions** from this audience with short answers.

Rules: do not overstate accuracy or claim causation. Avoid terms like SHAP, log-odds, or AUC in the main text; put any needed technical notes in a short appendix.
```

## What to fill in

- e.g. "scores leads weekly; sales calls the top 20%"
- e.g. gradient boosted trees, logistic regression
- e.g. "top 20% of scores contain 55% of conversions"

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