Check data quality with AI
Explore a new table before analysis and find out which missing, repeated or unusual values could change the answer to your question.
To check data quality with AI, establish what one row represents before counting duplicates. Three rows with the same order number might represent three different items purchased together. The explore-data guide helps inspect the table’s structure, candidate keys, distributions and coverage before the analyst decides whether a record needs attention.
A date column can look complete at first glance yet end several weeks before the period you want to report. Ask for the earliest and latest observations as well as the missing count. Keep any suspicious examples alongside the business question they could affect.
What it does
- Inspect column types, ranges and common values
- Estimate the proportion of missing observations
- Check suspected duplicates against the table grain
- Find gaps in the dates covered
- Identify useful grouping fields and possible joins
Install and use
- Install the skill and supply an authorised dataset or a permitted warehouse connection. Calculations require a host that can inspect the file or execute a read query.
- Ask “This export contains order lines. Describe what each row means, suggest a line identifier and count rows without an order date.”
- Check suspicious examples with the data owner before removing records or treating an unusual value as an error.
FAQ
Source
Made by Anthropic. Original skill: explore-data, published on GitHub under the Apache-2.0 licence.
View on GitHub