CleanDims

Join any datasets, even without common keys.

Sales and census population. Suppliers and company registries. One department’s figures and another’s. They rarely share an ID. CleanDims builds the missing key from the text columns they do share: places, industries, vendors.

example values
  • New York Countycounty_population.county_name

    County, rolls up to the state

  • NYSTORE_SALES.REGION

    Alias of the state

  • New YorkSTORE_SALES.REGION

    State or city? Sent to review

New York

id
PLC-0036
level
state
parent
PLC-0001
aliases
NY, N.Y., New York State

Store sales by state and population by county now join at state level.

Matches a canonical valueKnown aliasAmbiguous, goes to a person

The problem

The only link is a text column, and it breaks in three ways.

When there is no shared ID, analysts rebuild lookup tables by hand for every project. The work is slow, error-prone and never reused.

Spelling

AWSawsAmazon Web Services

One vendor, three strings. An exact-match join drops two of them.

Hierarchy

NYNew YorkNew York County

“New York” is a state and a city. One source reports by state, another by county, and nothing says which.

Vocabulary

Computer SoftwareSaaS / CloudHi-Tech

One industry, named three ways by three systems.

How it works

From connected sources to one canonical ID.

CleanDims reads the schemas and distinct values of your sources, never whole rows. Every step a machine scores, a person approves, and each decision is kept and reused.

Your databases and filesOpen datapre-mapped by CleanDimsSchemas anddistinct valuesno row data leaves the sourceColumns groupedinto dimensionsfive signals, one scoreyouYou approve each mappingscores from 0.60 to 0.90 wait for youUnique values collatedevery value, with record countsValues cleanedand groupedone grouping per canonical valueyouYou review the restat or above the threshold:auto-approvedCanonical IDs publishedwith hierarchy levelsDownload or syncthe outputsjoin any sources on the IDMonitornew values, new columns, driftnew valuesre-collated

The dashed boxes are the two human gates: a mapping takes effect only when someone approves it, and a grouping below the threshold waits for a reviewer. Everything else runs on its own.

The product

Connect sources, approve the mappings, download the IDs.

The demo is the product with a sample workspace loaded: your own databases, open datasets attached, one dimension already published and two waiting for review.

Open data

Public datasets, already cleaned.

CleanDims maps public datasets to the same canonical dimensions your data uses. Two examples from the open data catalogue:

Federal procurement portal · monthly

Federal contract awards

Contract awards to suppliers, with place of performance and industry code. Recipient names are typed by contracting offices and vary widely.

As published
Dell Federal Systems L.P.
Oracle America, Inc.
Amazon Web Services, Inc.
SALESFORCE.COM INC
Cleaned
VEN-0080 Dell
VEN-0047 Oracle Database
VEN-0033 Amazon Web Services
VEN-0042 Salesforce
80 raw values became 51 canonical valuesExplore in the demo
National statistics office · annual

Business establishments by county and industry

Establishments and employees by county and industry. County and industry names are written several ways across state submissions, often with codes attached.

As published
NEW YORK COUNTY, NY
COOK COUNTY, IL
Los Angeles County, CA
Fulton County, Georgia
Cleaned
PLC-0055 New York County
PLC-0084 Cook County
PLC-0062 Los Angeles County
PLC-0087 Fulton County
253 raw values became 89 canonical valuesExplore in the demo

Where it fits

CleanDims matches dimension values. Other tools match records or fields.

A text join needs the values inside a column to agree. That is a different job from deduplicating customers or wiring one system’s fields to another’s.

ApproachWhat it matchesWhat that means for a join
Entity resolutionRecords about the same customer or company, in data you holdLinks rows, but gives two datasets no shared dimension
Master data managementInternal master dataCostly, and takes months to deploy
Integration mappingFields between systemsDoes not touch the values inside those fields
SQL, scripts and LLM promptsWhatever this project needsRestarts on every project, with no shared canonical layer
CleanDimsDimension values across open and internal datasetsOne canonical layer with hierarchy levels, reused by every later join

Pricing

Half the effort a join would take by hand.

We estimate the hours an analyst team would spend on the join and charge half of that. Joins on dimensions you have already mapped take less work, so they cost less.

  • Open data, mapped once

    Public datasets are mapped ahead of time. You see which ones join with yours, and at which level.

  • Your own data

    Upload files or connect systems. Your columns map to the same dimensions.

  • A person for the hard cases

    Every mapping carries a confidence score. Uncertain ones wait for review, and each decision is reused.

Have two datasets that should join and don’t?

Tell us what they are. We will tell you whether they join, at which level, and what the join would cost.