How Name Matching Helps Match Insurance Companies and Consumers

AI and Machine Learning Are Transforming the Insurance Industry

AI and Machine Learning technologies are transforming many business sectors. In doing so, they have made it increasingly possible for vendors to better predict what kinds of customers are more likely to purchase the goods and services they offer.

An interesting example of this is provided by the insurance industry. Several start-ups are trying to revolutionize the purchasing of insurance. Based on a bootstrapping approach, they are able to improve recommendations to new potential customers as to which insurance companies are best suited to offer a satisfactory policy at the right price. (These start-up companies are middle men between the insurance companies and customers. The insurance companies are paying for the leads. The customers pay nothing.)

There’s a feedback loop operating here too. Once they have received the leads provided by the start-ups, the insurance companies can send back to them the names and other information on that subset of leads who actually go on to purchase a policy. The start-ups thus have both positive (the lead purchased a policy) and negative examples (the lead did not purchase a policy) for the recommendations they have made. This is a perfect situation for AI and Machine Learning techniques to improve their predictions of who is likely to buy. The insurance companies’ motivation in providing the feedback is to improve the quality of leads that come to them.

Name Matching Solves a Tough Problem in the Feedback Loop

There’s one particular AI technology, Name Matching, which solves a serious problem in the feedback loop: the insurance companies send a lot of inaccurate information back. Names are frequently misspelled, name elements such as middle initials may be absent, email addresses may be absent, etc. The companies have a lot of challenges in matching up the insurance companies’ data with their own database that contains the original data. At present the companies are limited to manual, human-review processes, and this is tiresome and slow.

Matching Data Records Is Hard

Matching up names and other biographical data is a difficult task. Aside from data that is simply missing or misspelled, there are other phenomena that could hinder a match. For example:

• Nicknames: Robert Bob
• Initials: John F. Smith John Smith vs. J. F. Smith
• Abbreviations in addresses: Avenue Ave.
• Differences in birth date formats: 10/09/71 October 9, 1971 vs. Oct. 9, 1971

Given that the U.S. is such a diverse country, there are also personal names from a wide range of cultures and ethnicities. These require their own specialized matching algorithms. Here are a few of the many examples:

• Missing elements: Haroun al-Rashid Haroun Rashid (Arabic personal names may exhibit or lack certain elements like “al”.)
• Transliteration variations: Bashir al-Assad Bachir al-Assad (Since Arabic names were originally written in Arabic script and since there are different and conflicting rules for spelling Arabic names in a Latin-script language like English, systematic variations in spelling occur.)
• Name order variations: Moon Jae-in vs. Jae-in Moon (Chinese and Korean names traditionally place the surname first, but sometimes they will occur in Western order.)
• Optional name component: Raoul Jiminez Ramos Raoul Jiminez (Spanish personal names include both the father’s and mother’s surnames, but sometimes the latter is dropped.)

In order to match at a very high rate of accuracy, the matching algorithms need to learn these characteristic variations (as well as many others).

AI-Based Name Matching Can Automate Customer Record Matching

Advanced Name Matching uses AI and Machine Learning to train its matching models using a very large sample of real-world name data. It has different matching models for each entity type such as person, address, etc., thus achieving the highest possible accuracy for each type. It is also tunable, since for every enterprise there are specific business rules that need to be followed: for example, one field in a record may by itself be enough to determine a match, but the match on that field has to be exact (a typical case for a business would be an email address). On the other hand, some fields will be considered not as crucial because their values may change over time (for example, home addresses).

To determine what is a good match when the field entries differ, businesses need to be able to set numerical weights on each field to specify precisely how important they are for overall matching. In addition, there must be a way to combine the field results to produce a combined score for an entire record. Finally, Name Matching must scale well to handle matching of large volumes of names.

In sum, smart AI-based Name Matching benefits both customers and insurance companies. It allows quick and accurate filtering so customers can find the best insurance for themselves, and companies can receive a higher volume of good leads. The insurance companies will be able to spend fewer resources on manual matching of names and more on closing sales.