DataPilot AI Analyst

Turn messy CSV and Excel data into evidence-backed business decisions.

DataPilot is a downloadable Agent Skill for Codex, Claude Code, and other clients that support SKILL.md. It gives your AI agent a rigorous analytics workflow instead of a generic spreadsheet summary.

What it does

• Profiles CSV, TSV, Excel, and JSON tables
• Finds missing values, duplicate rows, invalid fields, and aggregation risks
• Defines KPIs and checks denominators before calculating them
• Diagnoses sales, marketing, customer, product, and operations performance
• Separates observed evidence from interpretation and causal claims
• Recommends suitable charts without misleading visual choices
• Produces executive summaries, limitations, and measurable next actions

What you receive

• A production-ready SKILL.md
• A deterministic Python dataset profiler
• Business KPI definitions and diagnostic playbooks
• Visualization and quality-control rules
• A reusable management report template

How to use it

1. Install the downloaded Skill in Codex or Claude Code.
2. Upload a CSV or Excel file.
3. Ask a business question, for example:

Analyze this sales workbook. Explain why margin changed, identify data-quality risks, and produce a management-ready action plan.

Real example 1 — Sales diagnosis

Input: Seven monthly channel rows containing one duplicate and one missing revenue value.

DataPilot found:

• The March Search row was duplicated and would have doubled reported revenue.
• March Social revenue was missing, so the total monthly margin could not be reported honestly.
• After deduplication, Search revenue was flat from February to March.
• Social revenue fell 17.4% from January to February, while its margin fell from 30.6% to 23.5%.

Recommended action: Recover the missing source value, fix duplicate generation, and break Social results down by campaign before changing budget.

Real example 2 — Customer churn

Input: A 12-customer sample containing plan, tenure, support, usage, inactivity, and churn fields.

DataPilot found:

• Overall observed churn was 33.3%.
• All three customers with at least three recent support tickets churned.
• All three customers inactive for at least 14 days churned.
• The sample was too small to claim these factors caused churn.

Recommended action: Create a targeted intervention queue, then validate the policy with a larger cohort and a holdout group.

Real example 3 — Marketing performance

Input: Spend, leads, conversions, and attributed revenue across Search, Social, Email, and Affiliate.

DataPilot found:

• Email delivered 9.0x attributed ROAS and a 15.0% lead conversion rate.
• Social generated the most leads but only 4.0% conversion and 1.5x attributed ROAS.
• Blended attributed ROAS was 2.59x.
• Attribution did not prove incremental lift.

Recommended action: Run a limited reallocation test instead of moving the entire budget based on attributed revenue alone.

Why not use a generic prompt?

DataPilot enforces repeatable checks: source inspection, metric definitions, reconciliation, uncertainty, causal guardrails, and recommendation traceability. It is designed to prevent confident conclusions built on duplicated, incomplete, or mismatched data.

Requirements

• An Agent client that supports Skills, such as Codex or Claude Code
• Python 3.8+
• pandas; Excel profiling also requires openpyxl

The buyer uses their own Agent client and model access. No publisher API key is included or required.

Important limitation

Outputs are AI-generated analytical support and may contain errors. Verify material calculations and obtain qualified review before making regulated, financial, legal, medical, employment, or other high-stakes decisions.