# Prompt 3 — Surface Drivers for a Metric

> Adapted from Chapter 10 (Asha worked example, escalation drivers). The pattern for the Key Influencers AI visual.
>
> **Verified against**: Power BI Copilot, June 2026 surface (Key Influencers visual)

## Crystallize

> "What drives escalation on customer support tickets?"

Run against the Key Influencers visual with three Explain-by fields (Customer tier, Product line, Ticket age). The Crystallize work happens in the **choice of Explain-by fields**, not in a conversational exchange — Key Influencers is a no-conversation surface.

The discipline: choose Explain-by fields that you have a hypothesis about, not the fields most readily at hand. Explain-by selection is the prompt.

## Scope

No explicit scope turn — the visual just runs. Treat the field selection as your own contract with the data; if Key Influencers returns something surprising, ask whether your Explain-by set is the right one before accepting the surprise as the answer.

## Assemble

Key Influencers returns a ranked driver list — for the Asha case: (1) Ticket age > 7 days, 4.1x; (2) Enterprise tier, 2.3x; (3) Identity product line, 1.8x.

## Refine (per-visual move + Part I principles)

- **Ch 10 confound check** (the per-visual move for Key Influencers): Filter to the top-ranked driver's narrowest meaningful slice and re-run. In the Asha case, filtering to Identity-only made the Enterprise driver disappear — revealing that the original ranking implied two independent levers where there was actually one (the SSO/federation feature area within Identity tickets, confounded with customer tier through population overlap). **Run this check before treating any Key Influencers ranking as actionable.**
- **Ch 3 (Big Idea) and Ch 6 (color)**: If the Smart Narrative paragraph or the visual's headline foregrounds a driver the confound check disqualified, edit to name the specific finding the data actually supports.

See the [companion to Chapter 10](/ch/10/) for the full AI-visuals treatment.

## Ship decision

Ship the rewritten driver list when the confound check correctly isolates the operational lever. A Key Influencers ranking without the confound check is *vibe charting*, regardless of how confident the visual looks.

---

**When to use this prompt**: You have a metric and three or more hypotheses about what drives it, and you want a ranked influence list. The confound check is what distinguishes the discipline from accepting the visual's first answer.

**Dataset**: [asha-escalation-drivers](/datasets/asha-escalation-drivers/) — download CSV + JSON to reproduce the Key Influencers pass end-to-end. This dataset is shared across all four Chapter 10 AI-visuals prompts (Key Influencers, Anomaly Detection, Decomposition Tree, Smart Narrative).