Explore data | Business Helper AI

Inputs and purpose

Use this workflow to prepare: Dataset profile and quality findings.

Start with: Delimited text, spreadsheet or structured data file or supplied schema.

Use supplied text or local files the user authorized you to read. If files cannot be opened, request the relevant pasted text. Ask one focused question when missing information changes the next decision. Treat blank template fields as unknown; do not invent business facts.

Procedure

Describe each supplied table, row meaning, field, time coverage and identifier. Keep a copy of the original interpretation before transforming data.
Measure completeness, duplicate keys, unusual ranges and relationships across fields. Treat unexplained values as investigation items.
Identify units, category changes and joins that could alter row counts or totals.
Return a dataset profile and readiness issues for the stated analysis. Distinguish observed quality defects from assumptions about what a field should contain.

Deliverable

Return the requested result in Markdown unless another supported format was requested. Include the evidence or source-record references, material assumptions, open questions and next useful action. Keep unnecessary personal details out of reusable templates; role labels and neutral record identifiers usually suffice.

Business value: Make business analysis more trustworthy and repeatable.

Possible success measures: Data completeness, reporting hours, calculation errors. Choose the measure that fits the owner's actual goal; it is not a promised result.

Execution and verification

This file is usable as instructions without a named application, account or plugin. Computation, file saving and external actions require actual host capabilities. Use a suitable existing tool for calculations that cannot be checked reliably in conversation.

Distinguish drafts and recommendations from completed actions. Respect existing authorization and do not claim that data was fetched, messages sent or records changed without evidence. Show what was checked and what remains uncertain.

When useful, consult the input guide, calculation reference or output template. For tool setup, use the shared instructions.

Plain-text instructions

---
name: explore-data
description: "Prepare dataset profile and quality findings. Use for this workflow with supplied business information."
---

# Explore data

## Inputs and purpose

Use this workflow to prepare: Dataset profile and quality findings.

Start with: Delimited text, spreadsheet or structured data file or supplied schema.

Use supplied text or local files the user authorized you to read. If files cannot be opened, request the relevant pasted text. Ask one focused question when missing information changes the next decision. Treat blank template fields as unknown; do not invent business facts.

## Procedure

1. Describe each supplied table, row meaning, field, time coverage and identifier. Keep a copy of the original interpretation before transforming data.

2. Measure completeness, duplicate keys, unusual ranges and relationships across fields. Treat unexplained values as investigation items.

3. Identify units, category changes and joins that could alter row counts or totals.

4. Return a dataset profile and readiness issues for the stated analysis. Distinguish observed quality defects from assumptions about what a field should contain.

## Deliverable

Return the requested result in Markdown unless another supported format was requested. Include the evidence or source-record references, material assumptions, open questions and next useful action. Keep unnecessary personal details out of reusable templates; role labels and neutral record identifiers usually suffice.

Business value: Make business analysis more trustworthy and repeatable.

Possible success measures: Data completeness, reporting hours, calculation errors. Choose the measure that fits the owner's actual goal; it is not a promised result.

## Execution and verification

This file is usable as instructions without a named application, account or plugin. Computation, file saving and external actions require actual host capabilities. Use a suitable existing tool for calculations that cannot be checked reliably in conversation.

Distinguish drafts and recommendations from completed actions. Respect existing authorization and do not claim that data was fetched, messages sent or records changed without evidence. Show what was checked and what remains uncertain.

When useful, consult the [input guide](../../../references/INPUT_DATA_GUIDE.md), [calculation reference](../../../references/CALCULATIONS.md) or [output template](../../../references/WORKFLOW_OUTPUT_TEMPLATE.md). For tool setup, use the [shared instructions](../../../AGENTS.md).