Someone downloads a portal export, renames it, pastes it into a workbook, and hopes the columns did not move. Then Power BI is asked to be the meeting view.
The dashboard is not the failure. The Friday paste is.
Python’s job here is not a chart. It is a feed: pull, clean, load, run again next week without a laptop open.
What the paste is actually costing
The column order changes once. A VLOOKUP breaks. The refresh in Power BI succeeds anyway — on the wrong shape. The number in the room is last week’s accident, not this week’s fact.
A notebook named final_v7 that only runs on one machine is the same problem with better syntax. If the author is out, the meeting waits.
What a pipeline is
One scripted clean. Documented columns. A grain you can name. An output table Power BI or Tableau already knows how to refresh.
APIs replace the portal download when a stable endpoint exists. A scheduled job replaces “hit Run.” A log replaces hope. A failure should arrive before the meeting does — not during it.
That is Python analytics work: Pandas, ETL, jobs, and dashboard feeds. The meeting view stays in Power BI or Tableau unless you ask otherwise.
What Python should not do
Replace the executive dashboard with a Matplotlib gallery. Rewrite DAX that belongs in the semantic model. Become a second copy of the business logic the report already owns.
Cleanup belongs in the pipeline. Measures belong in the model. Pages belong in the meeting. Mixing those three is how every layer lies a little.
A sequence that holds
- Name the grain of the output table — one row equals what
- Write the clean once, in code, not in a paste
- Land it where Power BI already reads
- Schedule it. Watch the failure. Hand it off so someone else can run it
If the source is still Excel and always will be, the script can start there. The point is that Friday is no longer a person.
When this is not the first step
If two dashboards disagree, fix the model before you automate a bad feed. If the KPI is unnamed, write the sentence first. A faster wrong number is not an improvement.
PeekGraph scopes the layer that is lying — paste, model, or definition — on the call. Nothing is billed from a page. The files stay with you at handoff.
Clean the feed once. Stop pasting it every Friday.