Introduction

EXL sees a major opportunity for insurers within the enterprise operational canvas. Whereas insurers have put effort and invested in using analytics to make better decisions and automating manual processing, we notice limited emphasis on the upstream challenge of converting unstructured data to structured data, and extracting usable insights for better decisioning. In other words, only half of the operational canvas is fully serviced.

There are two common roadblocks in automating and adding intelligence to this part of the value chain. The first is that approximately 80% of the data generated from customer inputs is unstructured, and needs to be converted to a structured, machine-readable form. Second is the considerable challenge of data residing in multiple legacy systems, and a lack of data lakes or single dataassets providing a single source of truth.

Insurance companies today are under unprecedented customer pressure to process real world inputs faster, cost-effectively, and intelligently. Customers expect insurers, like retailers, to remember their choices, use their data effectively and respond to queries quickly. Though many insurers have taken significant steps towards their enhancing operational efficiency, rising consumer expectations and continued cost pressures demand that they do even more. EXL can help.

Enterprise operational canvas

Insurers today are manually processing increasing amounts of bulk-generated data and documentation on a daily basis, engaging up to 5-10% of enterprise bandwidth, increasing related costs for manual processing, and further impacting the customer experience. In fact, we estimate that a large global insurer can spend around $125 -$175 million and 4-6 million hours annually on manual document handling across multiple operational processes.

Exhibit 1: Enterprise operational canvas

THE CASE FOR NATURAL LANGUAGE PROCESSING IN INSURANCE

Natural language processing (NLP), a field of artificial intelligence (AI), is used to transform manual processes and analyse large volumes of unstructured data. This technology may be a game changer for insurers. Applied by experts who understand the insurance domain, an AI-driven data ingestion frameworks can:

  • Reduce error rates
  • Drive better and faster customer experience
  • Improve compliance and reduce monotonous tasks for human operators, while also avoiding the prohibitive cost of legacy system changes or migrations.

Demand for AI-driven data ingestion frameworks is exploding. The market size is anticipated to grow exponentially from around $3B in 2016 to around $18B in 2025, clocking 24% annual growth on a year-on-year basis.

MOVING AWAY FROM MANUAL:

The case for AI-driven data ingestion

Manual data processing is mundane, time consuming, and error prone. Insurers and brokers worldwide face the unenviable task of efficiently processing large volumes of documents spanning the customer journey, from onboarding to servicing to claims—a deluge of structured, semi-structured, and unstructured data. Processes such as submission, booking and issuance, bordereaux management, FNOL, and investigation generate complexities due to the large number of documents in a multi-channel input with wide variation. Manually processing these documents takes significant time and effort away from high-value core activities such as underwriting, claims adjudication, and policy onboarding. With customers demanding a near real-time response on every request, insurers and brokers must accelerate processing and swiftly make accurate decisions.

Documents for manual processing across insurance value chain

Consider an example of a broker submissions process for a large insurer. As Exhibit 3 denotes, multiple pain points arise from non-standardised formats, missing information, manual application checks, duplication efforts, and no lead generation insights for increased profitability.

Manual data ingestion in broker submissions

Exhibit 3: Manual data ingestion in broker submissions

To achieve the levels of efficiency needed to compete in a digital world, insurers and brokers must not only think smarter, but also execute smarter, by embedding intelligence into every business process.

Introducing EXL Xtrakto.AI™ - Proprietary Intelligent Framework:

Move up the curve with AI-powered automated data

EXL Xtrakto.AI™ is a next-generation, AI-based framework which uses AI and NLP to bring significant efficiencies to historically manual processes such as extraction and document classification. Our framework provides insurers with intelligent data ingestion capability that understands what documents are about and the information they contain, extracts relevant information and sends it to the right place.

EXL’s framework fundamentally changes the way insurance operations are done today, moving from production-based to quality assurance-based ways of working.

Our framework allows insurers and brokers to build a near-touchless data ingestion capability, ultimately unlocking efficiencies and reducing costs. It drives repeatability, scalability and speed to value across the enterprise, reducing effort by up to 70%. In addition, it drives significant auxiliary, growth-enabling benefits, including:

  • Enhanced customer experience enabled by lower turnaround times and improved quality
  • Data and process standardisation, as well as new data for decision analysis
  • Insurer ownership
  • Better accuracy with reduced error rates

TRACKING ADVANCES IN NLP TECHNIQUES

NLP is rapidly evolving, with new breakthroughs identified almost every week. Powerful machine learning algorithms incorporating concepts such as transfer learning are being developed using pre-training to build on existing algorithms. In order to evaluate these NLP models, the industry-wide benchmark General Language Understanding Evaluation (GLUE) has been established. It is based on a set of complex NLP tasks such as text classification, question answer pairing, and other capabilities. Within one year, more than ten NLP algorithms have been developed which performed better than humans on all tasks.

Does this mean we will be able to replace humans in various document processing processes? No. While we are making rapid progress on the AI/NLP front, we are still far from general AI algorithms which can directly replace humans in certain tasks. Even the most powerful NLP algorithms developed to date are an example of narrow AI, which only works for a specific set of tasks and an on the data which it has already been exposed to. Hence, the most advanced NLP solutions cannot work as plug-and-play software —instead, they contain several modules which must be stitched together or custom-developed, as needed.

It incorporates some of the latest NLP algorithms, and customises them further by exposing them to real-world data and tasks. For any new NLP use case, our framework consists of various customised pre-trained models which accelerate the solution development process. Each use case will have a varying development timeline and estimated benefit, depending on nuances such as data availability, languages involved, handwriting utilisation, and other areas.

EXL Xtrakto.AI™ is where three key frameworks levers meet to move beyond just automation and deliver the right technology solution for enterprises at scale:

  • Lever One - Operating Model: Triage incoming work by complexity and client into tiers; move the work to the right location
  • Lever Two - AI-Driven Extraction: Automatically extract and classify data using combination of technologies, such as OCR, traditional machine learning, and deep learning
  • Lever Three - Operationalising the frameworks: End-to-end process re-engineering; embed AI-models into the operational workflow and drive change management within teams

These levers, when applied across the current state of document processing, lead to an evolution of the current heavy human, or highly manual, state, to an NLP and AIenhanced thin human, or highly automated, state. Thin human models are digitised and machine-dependent, with extremely limited need for human intervention.

Consider the previous example of broker submissions process. End-to-end transformation to the target state is possible when the above three levers are applied at various levels of process: (See Exhibit 4)

Transformed target state for broker submissions

Exhibit 4: Transformed target state for broker submissions

Our framework also comprises several modules which promote faster deployment and scaling of the framework. Implementing an intelligent automation solution with AI at scale will re-invent the data management cycle, drive human-machine collaboration, and achieve at least a fourfold return on investment in the technology.

OPERATIONALISING EXL Xtrakto.AI™:

Building towards an intelligent insurer

The key difference between insurers investigating and industrialising intelligent frameworks lies in how they approach it, moving beyond considering, conceptualising and piloting towards deployment at scale. EXL’s twopronged approach helps clients set up a factory model to support swift delivery.

1. Prioritisation Roadmap

EXL’s market-tested segmentation approach for manually processed content builds on four broad categories based on content characteristics and feasibility of transformation: (See Exhibit 5)

Prioritisation roadmap for content categories

Exhibit 5: Prioritisation roadmap for content categories

Opportunities are then prioritised for framework implementation based on two filters:

  • Business value: The nature of work (classification, extraction, search), high-level process flow, cost of error, average handling time for manual effort, volumes
  • Execution complexity: Availability of AI training data, document types, variety of document formats, handwritten docs, language, current workflow, readyto-use algorithm availability

2. Execution roadmap

The use of intelligent automation at scale remains relatively immature for many insurers and brokers, inhibiting complete transformation. Some apply transformation levers in a siloed manner, limiting progress beyond pilot phases. EXL’s market-tested, threephase approach deploys the automated data ingestion framework on the above-mentioned content categories, with different blends of human and NLP/AI technologies:

Operationalisation roadmap for content categories

Exhibit 6: Operationalisation roadmap for content categories

Implementation timeline for a sample 'On deck' category

Exhibit 7: Implementation timeline for a sample “On deck” category

Continuing with our example of automated data ingestion for the certificate of insurance (COI) process, in order to support swift delivery of capability at scale, the three levers could be applied over a period of six to eight months. However, value in terms of utilising employees more efficiently for lower costs can be realised in as little as two months. EXL follows a phased approach that continuously refines the benefits case as we build out NLPdriven models on a much larger data sample and executes an implementation and workflow integration plan over the timeline. The execution timeline contracts with each use case, as the capability matures. Once set up, multiple use cases can be executed in parallel.

Typical approach for operationalisation

Exhibit 8: Typical approach for operationalisation

CASE STUDY:

How a large global insurer expedited value from EXL Xtrakto.AI™

EXL can point to multiple use cases across the value chain where insurance and financial services clients have benefited from our framework. A global insurer currently implementing EXL’s AI based data-ingestion framework partnered with EXL to reimagine the CoI process workflow. The three levers are being applied in a phased approach over a six-to-eight month period, and will enable the client to realise cost savings in a short space of time to intelligently automate almost half a million requests per year from more than 8,000 clients.

EXL’s framework is enabling the extraction of meaningful insights from documents, emails and customer service data. These insights allow quicker delivery of accurate customer outcomes, improvement in operational resilience, and increases capacity for the client’s operations teams to focus on more value-adding tasks.

As a result, the client will see a >50% improvement in efficiency.

What Differentiates EXL’s Proposition

EXL’s hybrid framework is differentiated from available market solutions due to its AI/ML driven engine and ability to generate domain specific insights. It addresses all critical dimensions pertaining to solution categories in the market –variety in input documents, OCR capability, NLP/NLU capabilities, domain expertise, ability to customise extraction engine, speed of deployment, and human + digital integration.

Why EXL?

EXL’s insurance domain expertise and digital capabilities are globally recognised. We were named as The Leader in the Everest Group’s 2020 P&C Insurance PEAK™ Matrix. EXL demonstrates significant breadth of proprietary AI/ NLP models, frameworks, and toolkits. Our solution brings enhanced NLP, machine learning, and artificial intelligence capabilities for accelerating document processes to meet business objectives for our clients. We believe that our solution is differentiated in four key dimensions:

Intelligent Framework-Market Landscape

Exhibit 9: Intelligent Framework-Market Landscape

  • Domain expertise that allows us to underwrite benefits: NLP framework development leverages our vast domain expertise. Most importantly, we ensure that our framework output plugs into the operational workflow that allows human representatives to manage exceptions and allows EXL to underwrite the benefits
  • Advanced NLP capabilities: Our proprietary stack of accelerators and frameworks built by our large team of data science professionals are fine-tuned for specific use cases; our stack speeds up the AI model development and integration process by up to 40%
  • Customisation: We understand that every insurer and broker has different processes, and do not follow a ‘one size fits all’ approach. EXL caters to multiple document types, including structured, semi-structured, and unstructured, and formats including jpeg, tiff, emails, and html
  • Flexible deployment with modular components: We can deploy our framework on premise or on the cloud, and allow our clients to choose solution components. For example, many clients choose to take our user interface and validation screens, while others have requested that we build customer document portals on their website

If you’d like to explore how leading insurers and brokers are leveraging our solution and the value it could deliver for you, we’d welcome the opportunity to talk and arrange a demo for you.

To see what Insurance clients are saying about EXL’s AI and NLP capabilities, please visit this link: https://www.exlservice.com/AIatScale

Acknowledgements/Contributors:

Raghav Jaggi

Mohit Manchanda

Roopak Chadha

Wayne Reed

Gaurav Iyer

Sumit Taneja

Chaithanya Manda

 

Written by

Rohan Regis
Vice President Insurance, UK & Europe

Rahul Singh
Senior Consultant, Insurance Industry Solutions

Contact US