Text Analytics for AML & Due Diligence

Reduce Human Labor with Artificial Intelligence

AI works to reduce the number of erroneous articles compliance officers must review. Investigators are compelled to follow up on every red flag, so the only way to reduce the number of cases is to present fewer irrelevant results, by increasing the precision of the search.

Watch List Screening

Names are vital data points in financial compliance, anti-fraud, government intelligence, law enforcement, and identity verification. Yet it can be challenging to match names.

Search for: Yoko Jean Hawkinberry

Does your system find:
Y. Jeanne Hockenbury
Yoko J. Hawkinbury
Hawkenberry, Y. J.


Your data includes variables such as misspellings, aliases or nicknames, initials, and non-Latin scripts that are not always aligned properly in your database.

Trusted by Mission-Critical Systems

Rosette provides enhanced name matching to mission-critical systems, including the watchlist system used by United States Customs and Border Protection as well as major commercial AML systems. Rosette saves money by reducing false positives (erroneous matches) and reduces risk by catching false negatives (missed matches).

Due Diligence Monitoring

In searching on people and companies, traditional due diligence struggles with too many articles that reference the wrong person or company. Or, they miss articles where names appear spelled different ways.

Search for: Somerville-based Acme Bank executive Alan Rose Gardener

Bad match: The gardener created a bank of roses on the path for the home of Alan Smith.

Good fuzzy match: Executive Allen R. Gardner was convicted of embezzling $1.6 million from Bank of Summerville.

The Approach

  1. Entity extraction finds people, organizations, and locations in each article, so that searches on people names only looks at people names, and not words that coincidentally match parts of a name (e.g., “gardener,” “rose”)
  2. Entity linking uses the surrounding context of a name and compares it to your knowledge base of known bad actors to link names in articles with the correct person (and not erroneously link different people with the same name)
  3. Fuzzy name matching overcomes more than 13 types of variations in 20+ languages

Benefits of Intelligent Fuzzy Name Matching

Unlike legacy solutions that generate static lists of name variations, hoping to match the name being queried, Rosette employs a hybrid of intelligent algorithms.

  • Compares names based on linguistic, orthographic, and phonological algorithms that understand how names are formed and how they vary
  • Handles disorganized and messy data
  • Matches name components within a single field or in the wrong fields
  • Handles inverted names and names that are incorrectly glued together
  • Speedily processes names with many components

Fuzzy name matching overcomes

Problem: Language barriers
Solution: Cross-lingual name search to find all mentions of a person or company regardless of the language it is written in

Problem: Word/name ambiguity
Solution: Uses the context of name in an article and compares it to known data about a bad actor in a knowledge base to correctly match name to real-life actor

Problem: Name variety
Solution: Higher precision reduces load on investigators

Rosette Advantages

  • Cross-lingual search in 20+ languages including Arabic, Chinese, Korean, and Russian
  • Simultaneous consideration of 13+ types of name variations in each search
  • Confidence-scored results for relevancy-ranked search results
  • Configurable to your data and needs
  • Easily integrates with existing systems
  • Intelligent algorithms that understand cultural and linguistic influences on how names are formed
  • Traceable match score reports
  • Scalable & lightweight enough to run on a laptop

Have a question for us? Or simply seeking some advice on NLP? Speak to one of our experts.

Select Customers

Red Flag Group

Frequently Asked Questions

Is Rosette customizable to my needs?

We understand that different customers have different requirements. Rather than deciding what degree of similarity equates a match ourselves, we've made the match threshold tunable so you can look at only the most likely matches, or all possible matches.

Does Rosette support transaction screening in Russian, Arabic, and Chinese?

Yes. Global identity and financial investigations now depend on cross-lingual matching and linguistic expertise across non-Latin script language, including "difficult" languages such as Russian, Arabic, Japanese, Chinese, Korean.

Why can’t we rely on our current approach of generating variations to match misspelled names?

A three-component name translated into English can have hundreds of variations. In a climate of escalating international regulation and penalties and when even a single failure to match a single name could jeopardize your customer, it is too risky to rely on outdated methods that do not accommodate all world languages and linguistic mastery.

Can Rosette adhere to watch list requirements originating from foreign countries?

Yes, we have customers all over the globe. Every country where you do business sets it's own KYC standards. Your product needs to reliably screen names against each and every list.

Will my current solution scale affordably as name matching demands grow?

Probably not! As watchlists grow, the number of name matches grows exponentially for name-generation solutions, but not for knowledge-based solutions like Rosette. Your customers need a solution that absorbs change, with minimal additional hardware or cost.

Is there a way to increase accuracy and cross-lingual matching functionality in the product we have now?

There are name matching products designed to seamlessly layer on top of existing functionality without adding a lot of additional hardware. Rosette easily plugs into many of the most popular search tools in the industry.

Can Rosette match names even when components are entered into the wrong fields?

Yes, and your solution needs to as well. When your customers are held accountable for accurately matching names, it is risky to rely on, for example, a bank employee in Chicago knowing that a Mexican customer’s surname is often two words, where one may be mistakenly entered as a middle name.