The Best Way To Resolve Every Record
AI matching trained privately on your own data, deterministic rules for the identifiers you trust, and built-in handling for nicknames, mappings, stopwords and any language.
Any Entity With Any Schema
Resolve customers, households, citizens, suppliers, products, parts or any other entity, on the fields you already have.
Scale To Millions Of Records
Zingg learns from your labels not to compare every record with every other record, so matching stays fast as data grows.
Label Minimal Records Only
The Zingg interactive labeler picks a few edge cases for you to answer, so human effort stays small.
Add Rules For Known Identifiers
When your data has identifiers you trust, like emails, SSNs or passport IDs, add deterministic rules alongside the model trained on your labels.
Learned From Your Labels
Finds matches through typos, nicknames, abbreviations, missing fields and changed addresses, using the model you trained on 40–50 pairs.
Rules On Sure Identifiers
Records that agree exactly on every field in a rule match with the top score of 1. Known identifiers improve both accuracy and speed.
Probabilistic and Deterministic Matching run in one flow, so every record that has a match finds one: probabilistically, deterministically or both.
Every Variation Your Records Throw At You
Real data has nicknames, codes, filler words and many languages. Zingg handles them inside the match, so you spend less time preparing data first.
Cleaned Automatically
Punctuation, extra spaces and upper or lower case never split a match. Zingg takes care of them on every field, so you don’t have to clean them up first.
- Punctuation
- Trimming and extra spaces
- Upper and lower case
NicknamesEnterprise
Recognises name variants like Bob for Robert, so records aren’t missed on a technicality. Use the prebuilt nickname mappings or bring your own.
MappingsEnterprise
Each system codes categories its own way. Give Zingg a mapping file and it matches on M, Male and 1 alike, with no transformation before matching.
StandardiseEnterprise
Write one canonical value to your matched output for fields like job titles, company names or product codes. Standardising runs after matching, before output is written.
Stopwords
Common words like Mr, Pvt Ltd or Street add no signal and skew match scores. Zingg recommends stopwords for each field from your own data, and ignores them while matching.
Multi-Language
Resolve records in Chinese, Japanese, Thai, Hindi, German, French, Greek and more, with no language-specific setup.
Company NamesEnterprise
Abbreviations and legal suffixes like Corp versus Corporation stop splitting one company into many. Use the prebuilt company mappings or your own.
Company name matchingMatching Questions
How many pairs do I need to label?
Typically 40–50. Zingg picks the pairs to show you, so a short session is usually enough to train a model for your data.
Does my data leave my platform to train Zingg?
No. Labeling, training and matching run as a native workload on Databricks, Snowflake, Microsoft Fabric, GCP or AWS. Your records and labels stay in your own account.
What if I can't tell whether two records match?
Answer Can't Say. Those pairs are skipped, so only confident labels train the model.
Do I have to write matching rules?
No. Zingg learns matching from your labels. Rules are optional: in Zingg Enterprise you can add deterministic rules on identifiers you trust, alongside the learned model.
When should I add deterministic matching?
When your data has sure identifiers such as emails, SSNs or passport IDs. Each rule lists fields that must agree exactly, for example first name plus email, and records that meet any rule match with a score of 1. Zingg still matches everything else with the model trained on your labels.
Does Zingg handle nicknames and abbreviations?
Yes. Zingg Enterprise includes prebuilt mappings for nicknames and company names, like Bob for Robert or Corp for Corporation, and you can bring your own mapping files for your domain.
Which languages does Zingg match?
English, Chinese, Japanese, Thai, Hindi, German, French, Greek and other languages, with no language-specific setup.
What are stopwords and why remove them?
Stopwords are common words like Mr, Pvt Ltd or Street that appear in many records and add no signal. Zingg recommends stopwords for each field from your data, and ignoring them makes matching more accurate.
Which fields does Zingg learn from?
The ones you choose. You pick the attributes that matter for your entity, such as name, address, email or product code, and how each should be compared.
Resolve Every Record Where Your Data Lives
Start with open source today, or talk to us about running Zingg Enterprise on your platform.