Blog, NLP

Why is Natural Language Processing Relevant for the insurance industry?

So far, digitalisation has made life more complex for insurers. In many aspects, the developments in past decade have put customers in the driver seat. They demand transparency, tailor made solution at competitive prices, and a variety of ways of interaction. At the same time, the work environment for people in the front line of fulfilling these requests has not become easier: more systems and tools to navigate and feed, more guidelines and manuals to understand and apply, more stakeholders internally and externally to satisfy.

All of this collides with the illusion, that digitalisation means information is available for anybody anytime. Real life experience however resembles more the bon mots: You want to hide any information? Put it on the company’s intranet, and surely, nobody will ever find it again!

So, how can people delivering insurance be helped tackling the growing complexity?

why is Natural Language Processing relevant for the insurance industry

What looks easy to the customer has become a nightmare for employees. Graph from Argo & Partner AG

The one thing most information sources required to take good underwriting or claims decisions have in common is that the information is stored in some form of text. Be it in pdf, word, e-mail, or intranet-sites. Having a technology that allows to “understand” all those texts from various sources and serve it to the experts when needed in a condensed form is very powerful. That is where Natural Language Processing, NLP comes in. Alias, as all good things, they are not easy to reach.

Search for “stock” in Google and you will get many pages about equities as well as about soup. Or consider the following sentences:

“The jaguar eats meat.”

“The jaguar eats gas.”

In the first sentence, we can understand that the jaguar in question is an animal as it eats meat. In the second sentence, the context changes completely with the word “gas”: Jaguar is a car that consumes gasoline.

What this shows is that in language, ambiguity does not just lead to fuzziness of meaning but potentially to complete misunderstanding. Picking the most likely meaning of a word does not do the trick. And pure machine learning is too ineffective.

In fact, it can be proven mathematically, that full understanding is impossible (see box).

why is Natural Language Processing relevant for the insurance industry

However, understanding can be approximated sufficiently by performing several layers of analysis, and by underpinning the analysis with as much language knowledge as possible. It means, NLP cannot just be programmed, it requires a solid linguistic basis, i.e. a knowledge graph.

why is Natural Language Processing relevant for the insurance industry

The knowledge graph is a representation of knowledge that includes the meanings of words and the relationships between concepts.

Cogito ergo sum[1]

A very advanced solution reaching a 90% “understanding” level is Cogito®.

Cogito, Latin for “I think” is built on a knowledge graph developed over more than 20 years for 14 languages, capturing more than 2 million words and 4 million relationships, it conducts lexical, grammatical, syntactical and semantic analysis, i.e. a “deep semantic analysis” to mimic human understanding of text. In addition to this “understanding”, the process results in a structured representation of a previously unstructured text, which now can be processed in digital formats in many ways.

why is Natural Language Processing relevant for the insurance industry

Conclusion

Natural Language Processing will play a critical role for insurer on the road to digitalisation. If applied wisely, it helps the experts take better decisions in an environment of increasing internal and external complexity. It also can help improve the customer experience without driving costs through the roof.

There are countless use cases imaginable, e.g.:

  • Make customer service people in call centres more effective in answering questions from customers, incl. automated answering of simple e-mail requests and usage of chat bots
  • Help avoiding costly mistakes by pointing underwriters to inconsistencies in tailor made wordings
  • Match reported claims against similar closed claims to speed up decision making process and potentially reduce claims leakage
  • Semantic analysis of claims reports to support fraud detection
  • Support desk top review of third party risk reports by suggesting rating incl. reasoning
  • Allow for more in-depth due diligence process by automatically screening information from internet on specific topics, or names

 

Lukas Stricker & Andre Guyer

 

The authors work for Argo & Partner AG, a company specialist in delivering digital innovation in the insurance industry and partnering with the Artificial Intelligence company Expert System, the owner of Cogito®.

[1] René Descartes, 1596-1650


Share On

Menu