Day 5 · 14:00
The course ends. The pack on Monday does not. This is the one-page operating system you take back to the function.
You will leave able to
- Leave with a checklist for every chart that will see EXCO.
- Know which tool does which job (Excel, R, Quarto, Shiny, AI).
- Name the 30-day habit that keeps the craft from rotting.
On the desk
Everyone can be a data scientist. Not everyone should publish a mean.
Faculty, Hack Your Data · Lagos Business School
The chart checklist
- Tidy: one variable per column, one observation per row.
- Unit of observation named in the subtitle.
- Distribution shown when a mean would hide a tail.
- Colour maps a variable, not a decoration.
- Title is a description; any claim lives in the prose, with a caveat.
- Source file and extract date in the caption.
- The pipeline that built it lives in a repo, not in a sent folder.
Which tool, on purpose
| Job | Default tool | Not this |
|---|---|---|
| Ad-hoc cut, n small, you are alone | Excel / Power Query | A Shiny app |
| Repeatable weekly brief | Quarto + tidyverse | A new workbook each Monday |
| 24/7 operations pulse | Shiny / Posit Connect / BI | A PDF from last Thursday |
| Draft the dplyr you cannot remember | bongo.ai / approved copilot | Paste the customer file into a consumer model |
| Publish the pack of record | Website on main | Email attachments named FINAL_v9 |
The 30-day habit
- Week 1. Recreate one chart from your last EXCO pack in ggplot. Hang both. Ask which one you would sign.
- Week 2. Move the source CSV into a project folder. Knit a one-page Quarto. Send the HTML, not a screenshot.
- Week 3. Add a stop-word list to one qualitative extract (VoC or exit comments).
- Week 4. Put the project on an internal Git host. Invite one colleague. The craft becomes a function’s, not a person’s.
What we did not cover — on purpose
Causal inference beyond the warning. Production machine learning. Credit-risk models. Geospatial. Those are other LBS rooms. This room was communication under uncertainty: tidy data, honest charts, reproducible packs, a copilot you do not blindly trust.
Community
Stay on the alumni channel. Share pipelines, not confidential extracts. If you hire analysts, ask to see a repo. If you *are* the analyst, ask for time to knit rather than time to beautify a slide.