Getting started
“Our data is a mess and nobody has time to clean it.” Good — that is the normal case.
Almost every conversation stalls here, and it should not. Cleaning is part of the product rather than something you have to finish before you start. Nothing below requires a data engineer, and none of it changes your original file.
First, the formats
The awkward file is usually the important one.
The exports a public institution actually has are rarely tidy CSVs. These four are the ones that normally stop a project on day one.
The stacked export
Finance and analytics systems export several tables into one sheet, one after another, with a title row between each. Most tools read that as a single broken table. The sectioned importer finds the boundaries and splits it into separate datasets.
PDF reports
Ratings books, foundation reports, state budget documents. Tables are extracted from the PDF rather than retyped, which is how most of this data currently gets into a spreadsheet.
Microsoft Access
The membership database somebody built in 2011 and nobody has replaced. Read directly — you do not have to migrate it first.
Excel and CSV
Including the ones with merged header cells, a blank first row, and a total row at the bottom that is not data.
Then, the profile
Every file is inspected the moment it lands.
Before you ask a single question, you get an honest account of what you have just uploaded. Most people find something they did not know about their own data in the first five minutes.
- Which columns have missing values, and how many
- Which look numeric but contain text
- Dates written four different ways in one column
- Duplicate rows, and near-duplicates that differ by whitespace
- Categories that are the same thing spelled differently
- Values far outside the rest of the column
- What each column actually is — a measure, a rate, a level, a count, a category or a time axis
Then, the fixes
Suggested, previewed, and only applied if you agree.
It proposes a recipe
Standardise these dates, trim this whitespace, merge these three spellings of the same county, drop this total row. A list of specific operations on specific columns — not a button labelled “clean”.
You see the result before it happens
The preview shows the data as it would be, side by side with how it is. You approve the operations you want and ignore the rest.
Applying it creates a new version
Your original is never overwritten. Cleaning produces a new version of the dataset that records which operations produced it, from which source. If a decision turns out to be wrong you go back to the version before it.
The reasoning stays attached
Six months later, when a funder asks why a figure changed, the lineage answers it. That is a harder question to face than the cleaning itself.
Send us the file you are embarrassed by.
Genuinely — the messy one. We will profile it, tell you what is actually in it, and build something from it. Free, and you keep the result whatever you decide.
Wondering where the file goes? Read how governance works.