Automatic data cleaning

Automatic data cleaning means the problems in your data are found and fixed without anyone writing rules or scripts first. In Querri, the Data Audit runs more than 180 data quality checks on every table as it arrives, and the Janitor, an AI agent, cleans what it found with you. The result is a cleaned view that answers, dashboards and agents use from then on.

A data audit is a systematic check of a dataset for the problems that make its numbers wrong: missing values in disguise, numbers stored as text, duplicates, mixed formats, gaps. In Querri, the Data Audit is a page that runs more than 180 checks on every table as it arrives and lists what it found, most urgent first, with why each problem matters.

Data quality checks that run without anyone writing a rule or asking for them. Querri checks each table when it is uploaded or connected and each view when it is built, and records every problem it finds as a data quality note you can acknowledge, dismiss or clean.

An AI agent does the cleaning work with you instead of handing you a list. The Janitor reads the Data Audit's evidence, tries fixes on your rows, asks you the calls only you can make, shows its plan, and builds a cleaned view when you say so. Querri then measures that view again to prove which problems are gone.

Any table that reaches Querri, from a file or a connector. That includes Shopify, Stripe, Salesforce, HubSpot, Pipedrive, QuickBooks, NetSuite, Klaviyo, Mailchimp, Intercom, Zendesk, Google Ads, Meta Ads, Amazon Seller, Google Analytics, Mixpanel, Amplitude, warehouses and databases such as Snowflake, BigQuery, Redshift, Postgres, MySQL, SQL Server and Oracle, files in Google Drive, and spreadsheet and CSV uploads. Data cleaning for Shopify orders or Salesforce opportunities works the same way as for an exported CSV.

Yes. Each table gets its own audit and its own cleaned view, so you clean each source first and then join the cleaned views. Dates end up in one format, currencies are handled the way you chose, spellings are folded and IDs are repaired within each source, which is what lets the join line up. The audit also checks detected join keys across tables, such as child rows whose key has no match in the parent table.

Yes. Spreadsheet exports get checks of their own: totals and subtotal rows inside the data, header rows repeated mid-file, months laid out as columns, Excel serial numbers posing as dates, IDs turned into 1.23E+11, zip codes that lost their leading zero, and a file that stopped at Excel's last row. Upload the export as it is and the Data Audit lists what it found.

Within each table, yes, and that is what makes tables from different systems agree. The audit spots several date formats in one column, proves day-first or month-first from the values where it can, and flags mixed currencies or an amount whose currency disagrees with the currency column. The Janitor asks the calls only you can make, such as whether to convert currencies or keep them separate, and the cleaned view writes every date and amount one way.

It separates three cases: rows repeated exactly, keys that repeat with the same values, and keys that repeat but disagree, such as one order id with two totals. The first two can simply be collapsed. When repeats disagree, nothing in the data says which is right, so the Janitor asks you which row to keep, with its recommendation first.

Your raw data stays exactly as it arrived. The fix is a cleaned view, defined in SQL you can open and built from the source rather than written over it. The source stays beside it so you can check one against the other.

A cleaned view is a saved view that applies the Janitor's fixes to the source table, and the source then reads Cleaned as with the view's name. Because it is a saved definition rather than a one-off edit, it runs the same way on next month's data. When the view is rebuilt, Querri measures it again, and a problem that comes back reopens its note.

Many data quality tools watch pipelines for engineers, or show column statistics and wait for someone to build the fix, and a cleaning script only checks what it was written to check. Querri runs its checks on every table before anyone asks, ranks what it found by the questions and KPIs that read each column, asks you the calls that change a number, and builds the fix as a view it then measures again.

Notes follow the same access rules as the tables they describe, and where someone can only see some of the rows, the example values are left out. A fix is never applied on its own: the Janitor shows its plan on a review card and builds the cleaned view only when the person working with it says so.

Yes. Dismiss a note with a reason and it stays settled, so the next agent reads your decision instead of raising the problem again. Use Make a fact to save what you know about the data, and agents read that fact whenever they use it.