DataPrep cleans a spreadsheet in a fixed sequence of steps, scores the result for quality, and records each step so you can review exactly what changed before you export. It is the room most teams start in, because almost every export needs the same five repairs before it can be used.
When to use it
- A bank, ERP or billing export has dates in three formats and amounts with currency symbols and lakh/crore grouping.
- The same rows appear twice after a re-export, and blank rows break every formula downstream.
- Column headers differ between months, so last month's workings do not run on this month's file.
- An auditor or a colleague needs to see what was changed, not just the end result.
What it does
- Standardises column headers (lowercase, underscores instead of spaces, special characters removed).
- Drops rows that are completely empty.
- Removes exact duplicate rows.
- Parses date columns into one consistent date format.
- Normalises number columns, including currency symbols, thousands separators and lakh/crore grouping.
- Scores the cleaned data for completeness, uniqueness and validity, and gives an overall grade from A to F.
- Lets you explore the cleaned rows, search a column, compare before and after, and read the audit trail.
Inputs and outputs
- Accepted files
- .csv, .xlsx — one file at a time
- Where you review
- Overview, Data explorer, Data quality, Audit trail and Export tabs
- Options
- Each step can be switched on or off; date, number and duplicate-key columns can be named
- Exports
- Cleaned CSV, XLSX, Parquet or JSON; the cleaning report and audit trail as JSON
- Tailored to you
- The steps, the column mappings and the export formats are set up for your files
Example
Example — fictitious data
- An invoice export with four date columns in mixed formats (07/01/2026, 2026-01-07, 07-01-2026).
- Result: 8 duplicate rows removed, 4 blank rows dropped, all 4 date columns standardised, quality grade A.
- The DataPrep sample on this site shows this run step by step.
What this does not do
- Dates written with a month name, such as 06-Feb-2026, are not recognised by the date step and come out empty. Check the date step's 'values parsed' count in the audit trail.
- Duplicates are matched exactly, value for value; near-duplicates are kept.
- The quality score measures structure (empty cells, duplicates, parse failures), not whether the values are correct.