← All use cases

Analyze datasets and ship reports

Convert messy data into clean analysis and visualizations.

ZeroTwo Work can tidy data, merge sources, dig into a question, model the outcome, and deliver a readable report backed by charts and noted caveats.

Intermediate
1h

Related links

Skills & Plugins

  • Spreadsheets
    Checks CSV, TSV, and Excel files, formulas, and joins, and builds tables or charts ($spreadsheets).
  • Google Drive
    Reads the Google Sheets and other files you point it to in Drive.
  • Data Analytics
    Bundled plugin for gathering context, validating data, and building reports, dashboards, and notebooks (desktop app).

Starter prompt

Do homes close to the highway sell for less than similar homes farther away? I've attached a sales CSV and a separate parcel-attributes file. Before joining them, check that the keys are unique and report the match rate, and never fill in a value you can't find. Then fit a simple regression with sensible controls, chart the main comparison, and write a one-page plain-language report with the caveats spelled out.

Introduction

The aim is an artifact someone can act on, produced through a loop of import, tidy, transform, visualize and model. ZeroTwo Work should yield analysis others can review and rerun.

ZeroTwo Work

Define your use case

Pick one specific question, then attach CSV or Excel files or name a Sheet by mentioning @google-drive. Analysis runs in a Python sandbox, and the desktop app can also work on files on your computer.

Running example: Property values near the highway

The question is how much lower valuations are near the highway; the work is making inputs trustworthy and documenting joins.

Example result

In a fictional sample, nearby homes averaged $500,000 versus $600,000 farther out, a gap that held at $94,000 after dropping the priciest outlier. The report notes the small sample and that causation isn't established.

Import the data

Have ZeroTwo inventory formats, meaning, key columns and quality issues first, with no conclusions yet.

Tidy and merge the inputs

Profile the merge before running it: key uniqueness, nulls, formatting, trial joins and match rates, then the safest strategy. Discuss derived keys' tradeoffs first.

Explore with charts

Compare groups, outliers and missingness against the question, and save useful charts.

Model the question

Begin with a transparent baseline such as regression with controls, stating target, features, leakage risks, evaluation and plain-language meaning. A weak model still shows where the problem lies.

Communicate the result

Deliver the format the audience needs (memo, spreadsheet, PDF or Jupyter notebook), with honest caveats.

Optional: set up a Python environment

For reusable scripts, ask for a small reproducible setup, such as a notebook or a script with pinned package versions, and keep the source files unchanged. It isn't required for attached files.

Suggested prompts

Prompts cover inventorying files, checking the merge before joining, charting, modeling and drafting the report.

Related use cases