Using Artificial Intelligence, a proprietary phonetic algorithm, lexicons, and a contextual scoring engine, matchit defeats the errors, inconsistencies, and challenges commonly found in contact and business data.
Relying solely on match keys to find matching records presents two major problems:
Sound familiar? It’s the same process humans use to judge similarity.
To make matchit truly powerful, we knew pre-processing had to be removed and accessibility put front and center. To achieve this, we tapped into Artificial Intelligence.
Natural Language Processing (NLP) refers to AI methods concerned with understanding human language as it might be spoken or heard. Using NLP techniques like lexical semantics, matchit develops an understanding of your data based on what it is and not where it resides in a table.
matchit understands that Anthony and Tony can be used in place of each other; that IBM is an acronym for International Business Machines, and that “Lyft, Inc.” should be treated as a business name despite being found in an address field.
The journey begins as your data enters matchit where it is first normalized. Using AI to understand individual data elements, matchit breaks down complex and concatenated fields into their constituent parts. The quality of fields like name, address, and email are calculated and the resulting qualitative scores inform matchit on the trustworthiness of specific attributes.
This process of normalization plays directly into phonetic tokenization. Now that we’ve isolated values like first name, company name, street name, and town name, we can generate phonetic tokens on these fields to help circumvent errors in your data.
Instead of using off-the-shelf algorithms like Soundex or Metaphone that were never designed for the nuances of contact and business data, matchit uses a proprietary phonetic algorithm that has an enhanced understanding of stress syllables and pronunciation.
Once matchit has made sense of the data, the normalized and tokenized values are used to seek out similar records. It’s important to note that we aren’t finding matches yet, we’re simply identifying groups of records that are good candidates for further comparison.
Unlike conventional solutions, this clustering process doesn’t depend on a single datapoint being accurate, consistent, and present. Leveraging the generated values from the previous steps, matchit is able to locate two matching records that have nothing exactly in common.
Why candidate groups? The short answer is scale. matchit can run efficiently on over a billion records and perform real-time lookups on massive datasets. Without candidate grouping, this wouldn’t be possible.
As candidate groups are created, they’re funneled into the contextual scoring engine where records are compared two at a time. All the available data that make up your records are graded for similarity and assigned a composite score.
Where conventional approaches rely on the comparison and scoring of match keys, matchit is comparing and scoring the contact name, company name, address, zip, telephone, email, website, and even custom fields individually.
The composite scores are brought together into a single match score which establishes the overall similarity between two records.
Built for a variety of deployments and use cases, matchit easily scales with the growing needs of a business.
