How it works
Three steps on the surface, five underneath
How it works
Three steps, about twenty seconds.
01Upload a CSV or spreadsheet
Drop in the file you already have. A sales export, a list of orders, survey responses, ad spend by campaign. Plotwell reads the header row and the data underneath.
02Plotwell understands it and builds a dashboard
It works out which columns are dates, categories and numbers, picks the charts that actually fit, and lays out a clean dashboard. No modelling, no metric setup, no chart wizard.
03Read the plain-language takeaways
Next to the charts, Plotwell writes what changed, what looks off and what to look at. Every figure is read from your data, so you can act on it, not guess.
Every number is computed from your rows
The charts are built by aggregating your actual data, in code. The model chooses which columns to chart and writes the takeaways. Any takeaway that cites a figure Plotwell cannot read from the computed output is dropped before you see it. Plotwell describes your data. It does not invent it.
In detail
The five stages
Written out because the trustworthy part of this product is not the model call. It is what happens on either side of it.
- 01Parsing
Getting the columns out
A CSV is already rows and columns. An Excel file is not: it is a structured binary format that may have merged cells, multiple sheets and hand-formatted headers. Plotwell pulls the data server-side, identifies the header row (which is not always the first row), and coerces date strings like Jan-24, 1/2024 and 2024-01 into the same month before any analysis runs. If the file cannot be parsed, it says so rather than running analysis on a misread table.
- 02Column understanding
What each column actually is
Before any chart is chosen, Plotwell builds a profile of each column: the type (date, category, integer, decimal), the cardinality, the range, and whether the column is sparse. A column that is 90% empty is noted as thin. A text column with 600 distinct values is noted as high-cardinality. These profiles are what goes to the model, not your raw rows, which keeps the step fast and keeps your data from being sent unnecessarily.
- 03Chart selection
Picking the views that answer a question
The model receives the column profiles and is asked to choose the charts that actually say something. A date column paired with a number column is a time series. A category column with a number column is a bar. A single number column with few rows is a table, not a chart. Choices that would just fill space are skipped. The model is told to prefer the view that answers a question a reasonable person would ask of this data.
- 04Aggregation
Every bar is a sum of your rows
The charts are not drawn by the model. Once the columns and chart types are chosen, the aggregation runs in code against your actual rows: group by the category column, sum the number column, sort descending. The model never draws bars from memory or produces numbers from training data. This is the step that makes the dashboard trustworthy rather than plausible.
- 05Takeaways
What changed, what looks off, what to look at
The model receives the aggregated output and writes two to four sentences per chart. Every figure it cites is checked against the computed output before you see it. A takeaway that contains a number Plotwell cannot find in the aggregated data is dropped, and the count of dropped figures is reported. A chart with nothing interesting to say gets no takeaway, because writing a neutral observation wastes the one line of text a reader actually reads.
Questions
Before you upload anything
See it on your own data
Four dashboards a month are free. Drop in the CSV you already have.