Product

Data quality

With our proprietary K.Refine algorithm, we audit, clean and reconcile data before it reaches your dashboards. That is what we call a data refinery: raw sources go in, clean and traceable data comes out.

Book a meeting with an expert Pricing

Reliable data, whatever the source

K.Refine

Data cleaning

K.Refine suggests the corrections, and you review them before applying them.

  • Spaces, capitals, variants and unusual numbers
  • Every correction you keep becomes a rule for new rows
  • No language model, no AI credit

Unique value

Deduplication

A unique-value column drops duplicates as soon as you import.

  • Case, accents and spaces ignored
  • Two spellings of the same value caught: “Lyon Est” and “LYON-EST”

Linked tables

Data matching

Link two tables like a VLOOKUP, on one to five shared columns.

  • Lenient matching: case, accents and spaces ignored
  • A trial run shows how many rows match before you link

Rules

Data quality rules that apply themselves

Write a rule once: it corrects every new row, whether typed in, imported or written by a workflow.

  • Tested on every row before it runs
  • Every run can be undone from the history

How K.IOP cleans your data, in 5 steps

  1. Import

    Excel, CSV or a workflow

  2. Analyse

    K.Refine spots what is off

  3. Review

    Keep or drop each proposal

  4. Apply

    Each correction becomes a rule

  5. Use

    Dashboards, alerts and AI read the clean table

K.Refine compared with other data cleaning tools

We ran the same files through every tool, with their clean version cell by cell. K.Refine fixes almost half of the errors, damages almost nothing, in under a second.

Criterion K.RefineK.IOP OpenRefineOpen source Raha + BaranAcademic research rapidfuzzPython library
Errors fixed 48%6,269 of 13,0366%77359%7,7343%371
Correct cells damaged 58out of 96,496Noneevery group reviewed by hand1,0592,254
Time per file Under 1 s1 to 17 splus the review40 s to 2 minAbout 1 s
What you have to give it Linked columnsoptional, declared in K.IOPA reviewof every group of values20 rows corrected by handThe columns to processand a similarity threshold
No code YesYesNoPython scriptNoPython script
Fixes new rows On every rownothing to rerunBy handby replaying the historyBy rerunningon the whole fileBy rerunningthe script

K.IOP benchmark of 28 and 29 October 2026: 8 files and 13,036 errors, including four public research datasets (hospital, flights, beers, rayyan). K.Refine with the column links declared in K.IOP; some of these files were used to tune it. Brands mentioned belong to their owners.

Deterministic by design, with no language model, and your data never leaves our servers.

Every rule run is recorded and can be undone. Your figures rest on rules, not on guesswork.

Language model
Nonezero AI credit used
Per analysis
200,000rows and 30 columns
Reversible
50,000rows per rule run

Common use cases

  • ERP vs files reconciliation

    Reconcile ERP data and field files to eliminate discrepancies and duplicates.

  • Supplier data cleansing

    Standardize and verify multi-format supplier references before any decision.

  • Ready for an operations desk

    Feed an operations desk with data that is already cleaned and reconciled.

Frequently asked questions about data quality

What guarantees that a displayed number is right?

Every rule run is recorded and counted, from the source file to the figure on screen. The rules are written down, and the test shows them before anything is applied.

How are duplicates detected?

Mark a column as a unique value: on import the first row is kept and those repeating its value are dropped. Case, accents and surrounding spaces are ignored. Without a unique column, every row is imported.

Can a transformation be tried without applying it?

Yes. “Test” shows the rule on 50 rows, before and after, with the rows it touches. Nothing is written until you run it.

What is a cleaning rule?

A rule selects rows by criteria, then writes a value or the result of a formula into one or more columns. Nothing changes until it runs. In automatic mode it applies to every row added or modified, and every run can still be undone from the history.

Does data cleaning use AI?

No. K.Refine is an algorithm with no language model, and it uses no AI credit. AI only steps in when you ask it to: writing a formula, proposing a rule, analysing a table.

Can I link two tables as with a VLOOKUP?

Yes. A linked column brings in a value from another table on one to five criteria, ignoring case, accents and spaces, and stays up to date on every write. After an import, K.IOP suggests the link when a shared key emerges.

How much data can K.Refine analyse?

Up to 200,000 rows and 30 columns per analysis. The number of tables and rows per table depends on your plan.

See pricing

Before and after fixing

  1. Connect

    Integrations

    Sources arrive as they are: ERP, APIs, files, legacy systems, with no migration.

    See every connector

  2. Make reliable

    Data quality

    K.Refine cleans, deduplicates and reconciles the data in between.

  3. Operate

    Dashboards

    Reliable data feeds the indicators, the alerts and the AI assistant.

    Explore dashboards

Traceability and reversibility. Every rule run stays in the table’s history and can be undone up to 50,000 rows.

See data quality on your own files

Book a live demo: we start from your real data.

Book a meeting with an expert Create my free account
  • Dry run
  • Runs you can undo
  • No language model