How Lakehouse powers NLP for Buyer Service Analytics in Insurance coverage

Obtain the Databricks Insurance coverage NLP Resolution Accelerator


The present financial and social local weather has redefined buyer expectations and preferences. Society has been pressured to develop into extra digital, and this extends to customer support in insurance coverage firms.

Nevertheless, there’s a important problem in approaching this concern with a data-driven mindset. Traditionally, structured information has been the principle ingredient that permits firms to research the previous with the intention to perceive and predict the long run. By leveraging Pure Language Processing (NLP), firms can analyze completely different sources of unstructured information generated by clients, reminiscent of audio from telephone calls and free textual content from chat messages.

Pure Language Processing (NLP) refers to a set of applied sciences that allow computer systems to know, interpret, and generate human language. Within the insurance coverage business, NLP can be utilized to automate duties that require understanding and processing massive quantities of written or spoken language, notably in customer support, claims processing, and underwriting.

At present, clients have completely different expectations concerning product customization and worth. They like insurance coverage built-in into their every day lives relatively than one thing they renew annually. A seamless buyer expertise is now anticipated, with 58% of shoppers stating that their customer support expectations are greater than they have been a yr in the past, in keeping with TalkDesk. One other report from Bain & Firm reveals that globally, 59% of 28,765 shoppers in 14 international locations need life insurers to reward them for wholesome residing.

From the angle of insurance coverage suppliers, maintaining with buyer calls for could be difficult. Forrester estimates that 53% of assist groups have seen a rise in assist queries for the reason that pandemic began. Offering a digital, self-service expertise turns into paramount with the intention to scale back stress on insurance coverage contact facilities.

Insurance coverage firms have been utilizing chatbots and IVRs for years to answer buyer queries about widespread insurance coverage matters reminiscent of checking the standing of a declare, reporting claims, and understanding insurance coverage protection. Nevertheless, the problem is to make sure that IVRs present a fascinating consumer expertise with out overwhelming customer support brokers with an excessive amount of complexity and overhead. In response to Oliver Wyman, many of the ache factors leading to a less-than-ideal buyer expertise are as follows:

  1. Screening: Buyer chatbots typically wrestle to interpret extra advanced buyer requests. These chatbots could have restricted comprehension capabilities, making it troublesome for them to know the explanation behind a buyer’s name. Giant Language Fashions (LLMs) have proven improved capabilities in comparison with conventional Pure Language Processing (NLP) on this space.
  2. Routing: There may be restricted routing of consumers, because the chatbot could not absolutely perceive the character of the shopper request, ensuing within the buyer being routed to an individual, and most certainly ready on maintain once more.
  3. Decision: Customer support brokers lack important instruments to resolve inquiries rapidly. Chatbots could not precisely summarize buyer requests, retrieve insurance coverage paperwork to confirm protection, or present a listing of related options for the agent. Because of this, brokers could must ask the shopper to repeat the explanation for calling, inflicting additional delays as they retrieve the shopper’s coverage and related particulars, to not point out degradation of the customer support expertise.

To enhance the shopper expertise, insurance coverage firms ought to contemplate leveraging extra superior applied sciences, reminiscent of LLMs, to boost chatbot comprehension and routing capabilities. Offering brokers with complete instruments that permit fast entry to buyer info and related assets may assist streamline the decision course of.

It may be daunting for insurers to outline one of the best technique for offering higher buyer expertise and care, whereas coping with the challenges of scaling and coaching their customer support workforce. Learn how to correctly steadiness the 2 sides of this coin? And most significantly, the place to begin?

Potential Outcomes of Making use of NLP to Buyer Service

Investing in a holistic digital transformation technique might help insurance coverage firms seamlessly scale their operations whereas shifting budgets and human assets from operational processes to precise product and worth creation. Within the context of customer support, one of many key issues when formulating such a method is name deflection. By understanding widespread buyer ache factors and empowering them with self-service channels, insurance coverage firms can scale extra simply whereas offering a sooner and extra related buyer expertise.

To use digital transformation to customer support within the insurance coverage area, we have to perceive why clients are getting in contact. For example, we would wish to know:

  • What are the highest 10 causes for buyer calls?
  • What number of clients are calling for help with auto insurance coverage versus different varieties of merchandise, reminiscent of life insurance coverage or medical insurance?
  • How are these distributions altering month-over-month or year-over-year?

These insights will permit us to formulate a correct technique for customer support, in addition to different areas reminiscent of advertising and usefulness. Moreover, we are able to analyze which merchandise or matters are doubtlessly problematic, reminiscent of when there are too many contact re-occurrences from the identical buyer. Lastly, we have to examine whether or not our customer support workforce is well-prepared to serve our clients.

Doing this sort of evaluation requires organizations to transform uncooked textual content into well-understood statements (labeled textual content) and depends on structured information. Within the context of pure language processing, transformer fashions –reminiscent of BERT, GPT, and ChatGPT– have made it doable for firms to extract worthwhile, structured insights from one of these information at an unprecedented scale. These fashions permit for simple classification of buyer utterances primarily based on particular buyer intents, in addition to gauging buyer sentiment.

Implementation Challenges & Motivation

As soon as an organization understands its clients, it might probably transfer to the fitting aspect of the Knowledge and AI Maturity Curve. Pure Language Processing (NLP) and Transformer fashions can help in automating buyer experiences, reminiscent of utilizing chatbots for engagement, and personalization, reminiscent of predicting buyer intents and their subsequent interactions primarily based on prior historical past. Regardless of the potential of NLP and Transformers, the present adoption panorama in most enterprises exhibits that there’s a important untapped alternative. In response to McKinsey, as of the tip of 2022, solely 11% of enterprises have or plan to incorporate Transformer fashions as a part of their AI merchandise.


We’re happy to announce the launch of a brand new Resolution Accelerator specifically designed to set the technical finest practices and reusable capabilities across the creation and upkeep of NLP and Transformers fashions for insurance coverage name heart analytics. The answer accelerator is a set of artifacts (information, notebooks, code, visualizations) that provides firms a head begin on creating and deploying a machine studying resolution. Within the following sections, we’ll have a look at the completely different elements of this resolution accelerator.

Insurance NLP Solution Accelerator

The aim of this resolution accelerator is twofold:

  1. Detecting buyer intents at scale primarily based on textual information from an Interactive Voice Response (IVR) stream or preliminary interactions between a buyer and a service agent.
  2. Classifying opening sentences from clients to a customer support chatbot in real-time.


In conclusion, the capabilities of NLP, Transformers, and Giant Language Fashions (LLMs) will proceed to evolve. Nevertheless, it is vital to notice that no insurance coverage firm possesses good information. Latest adjustments within the macroeconomic setting, reminiscent of inflation, provide chain disruptions, evolving loss tendencies, and the impression of local weather change on catastrophes, together with workforce adjustments and updates in underwriting guidelines and protection eligibility standards, have drastically altered insurers’ enterprise combine. Because of this, historic information could not successfully generalize to future eventualities.

Insurers have two choices: they’ll both purchase exterior or third-party information to complement their structured information derived from operational techniques (reminiscent of insurance policies, exposures, premiums, protection, and claims), or they’ll improve their inner structured information with inner unstructured information, together with voice calls/audio, photos, textual content, and movies.


Our aim with this resolution accelerator is to show how straightforward it may be to attain digitization capabilities by using buyer unstructured information and leveraging NLP, Transformers, and a Lakehouse platform.

To get began on creating and deploying a machine studying resolution for detecting buyer intent primarily based on items of textual content from an Interactive Voice Response (IVR) stream or from a digital agent, obtain the Insurance coverage NLP Resolution Accelerator.

Obtain the Insurance coverage NLP Resolution Accelerator

Related Articles


Please enter your comment!
Please enter your name here

Latest Articles