# Prompt 5 — Explore a Metric Across Dimensions with an AI-Assisted Tree

> Adapted from Chapter 10 (Asha's escalation-page pass; parallel to Fig 10.3's plant-0477 backorder example). The pattern for the Decomposition Tree AI visual.
>
> **Verified against**: Power BI Copilot, June 2026 surface (Decomposition Tree with the AI *High Value* / *Low Value* operations)

## Crystallize

Decomposition Tree has no Crystallize prompt in the conversational sense. The Crystallize work is:

- **The analyzed measure** (in the Asha case: *Escalation rate*)
- **The Explain-by field set** (*Customer tier*, *Product line*, *Feature area*, *Assigned account manager*)
- **The initial filter state** on the report page — this is the load-bearing Crystallize choice most analysts skip. The tree's AI walk optimises from within whatever filter is set; if you haven't audited your slicer state, the walk you're about to quote is scoped to your view, not the audience's.

## Scope

No conversation in this surface. The stand-in: write down which slicers on the report page are set to non-default values before starting the walk. If you cannot list them, you have not audited your own filter state.

## Assemble

The tree's *High Value* walk from the root — in the Asha case: *All escalations → Enterprise tier → Identity product line → SSO feature area → Customer X*. Presented in Power BI as a smooth left-to-right cascade that reads as an AI-discovered path.

## Refine (per-visual move)

- **Ch 10 filter-clear re-walk** (the per-visual move for Decomposition Tree): Clear all cross-filters on the page and re-walk the path from the root. If the AI-picked dimension at any level differs from the original walk, the original was filter-dependent. In the Asha case, the second-level pick changed from *Identity product line* to *Ticket age* once an analyst-set Q3 filter was cleared — the walk about to be quoted was scoped to her analyst view, not the audience's.
- **Ch 3 (Big Idea)**: If you quote the path at all, quote it with the filter state named explicitly (*"under a filter to Q3 only, the AI-selected high-escalation path is Enterprise → Identity → SSO → Customer X"*), not as an unconditional finding.

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

## Ship decision

Do not quote a Decomposition Tree path in an async report; the audience will read it under a different filter context than the one that produced it. Save the analysis as follow-up material if leadership asks the question live; if they do, re-walk the tree in the meeting under the filter state the audience is actually looking at.

---

**When to use this prompt**: You have a metric and want to explore which dimensional slice contains the extreme values. The filter-clear re-walk is what distinguishes the discipline from quoting a filter-dependent path as an unconditional finding.

**Dataset**: [asha-escalation-drivers](/datasets/asha-escalation-drivers/) — download CSV + JSON to reproduce the tree walk in Power BI. Shared across all four Chapter 10 AI-visuals prompts.