Emerging Tech Impact Radar: Artificial Intelligence in Insurance
Published 9 August 2023 - ID G00786204 - 64 min read By Analyst(s): James Ingham, Kimberly Harris-Ferrante, Moutusi Sau Initiatives: Industry Product Planning and Strategy
AI innovations are no longer solely the domain of specialists as a range of different vendors incorporate novel techniques into their offerings. Product leaders targeting insurers must understand the timing and impact of AI advancements as they build dedicated vertical specific roadmaps.
Overview Key Findings
Recommendations Product leaders developing AI-enabled insurance solutions as part of an industry product planning and strategy activity must: The massive data footprint within insurance companies, including the large number of disparate systems, will drive the need for data fabric in the future as insurers improve data availability for decision making, automation and unlocking new insights. ■ Generative AI has potential across the value chain for both internal and customer- facing use cases to improve document processing, customer self-service, marketing, data science and operations such as claims, underwriting and product filings. ■ Basic chatbots, as well as computer vision (CV), intelligent applications and natural language processing techniques are the AI techniques closest to early majority adoption as insurers strive to improve customer service and operational efficiency. ■ Internal-facing use cases for advanced virtual assistants and decision intelligence will take longer to achieve early majority adoption, as insurers focus on the use of AI in customer engagements, rather than employee empowerment. ■
Gartner, Inc. | G00786204
Page 1 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Analysis This research document is contextualization of Emerging Tech Impact Radar: Artificial Intelligence published in October 2022. The entirety of this work is applicable only for vendors offering solutions to insurance companies. The use of artificial intelligence (AI) continues to spread throughout the insurance industry, with many companies starting their AI journeys while others are growing their use of AI. In the short term, AI is helping insurers solve immediate problems around automation of manual tasks, supporting digital channel interactions and faster identification of claims fraud. Longer term, AI can help drive strategic transformation such as new product innovation, dynamic customer engagement and panoptic personalization. The Impact Radar The emerging technology impact radar is an analysis of the maturity, market momentum, and influence of emerging technologies and trends. This report evaluates 14 emerging AI technologies with significant near- and short-term impact to the insurance industry. Capture improvements in customer-service-related metrics as part of your advocacy program from early adopter insurers, to help support new business development with fast followers ■ Create persona-specific advanced virtual assistants (VAs) and decision intelligence that help augment the roles of underwriters, claims managers and customer service representatives, instead of more general enterprisewide VAs. ■ Look for areas where generative AI advancements improve and accelerate the accuracy and efficiency of existing AI deployments — such as real-time summarization of customer call issues or extracting data from complex policy documents — before turning to net new use cases. ■ Start by demonstrating the value that data fabric brings in optimizing existing insurance processes in underwriting, claims and actuarial functions, before expanding relationships with insurers to more transformational data fabric use cases, such as personalization and dynamic pricing. ■
In scrutinizing these 14 technologies, we have identified four overarching themes:
Gartner, Inc. | G00786204
Page 2 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
The impact of AI techniques on insurance is immense. Many products and services either have embedded AI-based products or services or are going in that direction in the near future. This report on emerging technologies and trends in insurance analyzes 14 technologies in depth that have a profound influence on AI-based products and services (see Figure 1). AI-enabled applications and new use cases will enable innovative solutions and even create disruption in the way customers and employees engage with systems and with each other. Several technologies fall under this space, including CV, advanced VAs, intelligent applications and decision intelligence. ■ Responsible and human-centric AI highlights how AI can benefit people and society, manage and mitigate AI risks, and enable vendors to be ethical and responsible, while complementing AI with a human touch and with common sense. ■ Advancements in model-centric AI technologies are improving model accuracy, functionality and potential applications. Techniques like generative AI will enable use cases for ingesting, understanding and even generating various content, languages and speech. ■ Data-centric AI explores various data and analytics techniques to make sense of business data, as well as pushing AI-enabled decisions to be not only more accurate, but also explainable and ethical. AI technologies like tabular synthetic data and graph technologies fall into this category. ■
Gartner, Inc. | G00786204
Page 3 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Figure 1: Emerging Tech Impact Radar — Artificial Intelligence in Insurance
The objective of this research is to guide product leaders on how emerging technologies and trends are evolving and impacting areas of interest. Providers can leverage this knowledge to determine which technologies or trends are most important to the success of their business and when it makes sense to advance their products and services by investing in them. Refer to the How to Use the Impact Radar section for more information. Emerging Technologies or Trend Profiles Table 1 presents emerging technologies in AI based on time to early majority adoption. Use the table to jump to specific profiles. Each profile name is linked to the full technology profile to enable easy navigation.
Table 1: Emerging Technologies in Artificial Intelligence Based on Time to Adoption
Gartner, Inc. | G00786204
Page 4 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Now
1 to 3 Years
3 to 6 Years
6 to 8 Years
Chatbots
Computer Vision
Advanced Virtual Assistants
Autonomous Processing
Intelligent Applications
Decision Intelligence
Data Fabric
Natural Language Processing
Generative AI
Fully Autonomous Driving
Graph Technologies Image and Video Synthetic Data
Tabular Synthetic Data
Responsible AI
Source: Gartner
Gartner, Inc. | G00786204
Page 5 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Now Range Chatbots Back to Top
Analysis by: Kimberly Harris Ferrante and James Ingham
Description: A chatbot is a domain-specific conversational interface that uses an app, messaging platform, social network or chat solution for its conversations. Basic chatbots use simple, decision-tree-based implementations that are always narrow in scope. A chatbot can be text-based or voice-based, or a combination of both.
Sample Providers: Amazon; boost.ai; Hi Marley; Pand.ai; Rulai
Range: Now (0 to 1 Year)
Gartner estimates that insurance adoption of chatbots is approaching 60% of the way to the early majority target. Chatbots in insurance have been predominantly deployed in property and casualty (P&C) insurance, with more limited use across individual and group life insurance. It is worth noting that there are also rule-based chatbots available on the market that scan for keywords, which should not be confused with AI-based chatbots. These first gen solutions will also be superseded by Large Language Models (LLMs). Effective use of a chatbot enables increased revenue through expanded support of online sales of complex products (especially relevant to life insurers) and lower costs by driving sales and customer interactions to a less expensive self-service channel. Chatbots help insurers respond to increasing expectations from their customers when engaging with their provider — which translates to higher financial empowerment and revenue positive actions for policyholders.
Mass: Medium
Product leaders can select from a wide range of conversational platforms to create chatbots for insurance. Some platforms use a complex mix of semantic and machine learning (ML) technology, and most are rigid, menu-based systems with preconfigured responses.
Gartner, Inc. | G00786204
Page 6 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Chatbots are rarely deployed stand-alone by insurers and are most commonly used alongside ML and VAs. Chatbots are primarily used in sales/distribution and customer service functions (including no-touch claims processing), with some implementations spanning multiple functional areas as insurers strive to improve the customer service experience. Many of the front-office use cases also extend out to the agency channel and can be used on a variety of platforms, including voice, web, mobile and Facebook Messenger. Specific use cases in the insurance industry include support for administrative tasks. These can include how to obtain a copy of policy or find content on a webpage, point of quote and bind, automated billing chat capability, handling property first notice of loss (FNOL) claims and claims intake, and requesting photos and video, reducing the time to inspect. Life insurers are also using chatbots to support personal health and wellness enrollment and coaching, as insurers are now progressing from sales-oriented use cases through to additional self-service use cases for policyholders. Strong understanding of insurance terminology is essential to meet engagement needs.
Recommended Actions:
Focus on domain expertise, prebuilt terminology/learning and insurance-specific use cases when targeting the insurance industry, including insurance-related terminology and specific interactions that occur through the insurance value chain, such as FNOL notification or midterm adjustments. ■ Identify critical points in customer processes where the conversation needs to be switched from chatbot to human to maintain a positive customer experience. Incorporate queue management capabilities to expedite switching between channels and rapidly transfer the customer to a customer service representative or call center agent. ■
Recommended Reading:
Prioritize Chatbot Application to Meet Current and Future Self-Service Demand ■
Gartner Fast Answer: What should I know about Chatbots?
■
Gartner, Inc. | G00786204
Page 7 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
1 to 3 Years Computer Vision Back to Top
Analysis by: Kimberly Harris-Ferrante, James Ingham and Nick Ingelbrecht
Description: Computer vision (CV) is a process and set of technologies that involve capturing, processing and analyzing real-world images and videos to allow machines to extract meaningful, contextual information from the physical world.
Sample Providers: Amazon Web Services (AWS); Cape Analytics; Clarifai; Genpact; IBM; Microsoft Azure; Tractable
Range: Short (1 to 3 Years)
CV adoption is being driven by the price/performance of vision systems, heightened demand for monitoring and surveillance, and the need to automate image and video analysis to cope with increased volumes of unstructured image data. CV growth is less advanced in insurance compared with other verticals but taking off quickly as insurers seek new ways to improve claims, underwriting and customer service processes. The choice of where CV capabilities reside is primarily dictated by business considerations, availability of data/images to support the process, and the need for real- time intelligence, as well as potential data privacy and security concerns. CV adoption will be driven by increasingly sophisticated video and image search capabilities, fine-grained object and behavior recognition, improved metadata extraction, and advanced analytics. This data can be supplied by the consumer (e.g., submitting a picture of the damaged vehicle) or through a third-party data broker (e.g., video imagery of a city block). It can be combined with sophisticated rule engines and expert systems to deliver predictive and prescriptive analytics. Common challenges hindering CV providers include lack of suitable training data, ecosystem (channel, service and technology) dependencies, as well as lack of skills in adopter organizations. However, low-code/no-code platforms are rising to meet this challenge and are driving end-user adoption.
Mass: High
Gartner, Inc. | G00786204
Page 8 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
CV delivers high impact as a critical business disruptor through 2025. CV is an essential precursor to automation through the analysis of unstructured image/video data, enabling new business models and new forms of engagement. Moreover, it will serve as the principal mechanism by which enterprises elicit real-time operational insights and business intelligence from large volumes of dark data assets (stored image and video).
Current and emerging use cases for CV include insurance specific use cases as well as industry use cases that improve risk management, such as:
Image analysis to determine age and type of damage for fraud detection ■
Claims estimation and vehicle damage assessment
■
Wildfire property and damage assessment
■
Property inspection, for example, roof for underwriting, risk selection or claims assessment. Video surveillance for security, worker safety, pandemic mitigation and
■
Insurers, like other industries, can leverage CV to improve operational efficiency, avoiding or reduce costs and improve product or service quality. Insurers can also deliver data insights on business operations and frontline personnel such as claims adjusters or underwriters. Specifically, insurers can improve risk assessment to drive underwriting profitability and efficiencies, reduce loss as it relates to fraud identification, and deliver improved CX as core processes around claims handling and policy issuance are expedited. regulatory compliance across all industries. Examples include in-cab driver monitoring, in-store security threat identification, worker-machine interaction monitoring and health standards enforcement in food services. Adoption of these applications is accelerating, in line with the evolving threat environment.
Recommended Actions:
Map out the process for risk assessment, underwriting and claims to determine tasks where CV can help with business outcomes ■
Partner with core system (e.g., claims management solutions for P&C insurance) and adjacent vendors such as underwriting workstations to support an integrated, frictionless process. ■
Gartner, Inc. | G00786204
Page 9 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Look for third-party data providers that have images such as aerial imagery or governmental flood map data to assist in implementation and prove solutions fit in different lines of business. ■ Assess emerging niche insurtechs that offer various types of CV solutions aimed at specific lines of business to determine their competitiveness. ■
Recommended Reading:
Emerging Tech: Revenue Opportunity Projection of Computer Vision ■
Emerging Technologies: Emergence Cycle for Computer Vision ■
Emerging Technologies Tool: Video Analytics Functionality Matrix ■
Hype Cycle for Digital Life and P&C Insurance, 2023, specifically the profiles for Holistic Fraud Management Solutions, and AI Remote Inspection Solutions. ■
Intelligent Applications Back to Top
Analysis by: James Ingham, Alys Woodward and Eric Goodness
Description: Intelligent applications are enterprise applications with embedded or integrated AI technologies, such as intelligent automation, data-driven insights and guided recommendations. These deliver a more personalized interface, improve productivity and support “process-embedded” decision making. Injecting optimization, advisory and decision-support capabilities into process-centric workflows delivers significant enhancements to traditional, highly procedural enterprise business applications. Intelligent business applications will be enabled by the principles of composability, allowing for components to be added and recombined as required.
Sample Vendors: CCC Safekeep; Guidewire; Google Docs; IBM; JAGGAER; Microsoft 365; Oracle Cloud Applications; Salesforce Einstein GPT; ServiceNow; Vymo
Range: Short (1 to 3 Years)
Gartner, Inc. | G00786204
Page 10 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Intelligent business applications will “cross the chasm” for early majority adoption within the next one to three years in insurance. This is due to widespread embedding of AI technologies within both business applications (ERP, CRM) and, more prominently, insurance-specific solutions such as P&C core platforms. These offerings drive customer loyalty and application dependence, rather than specific revenue lines for vendors, with AI enhancing the usefulness of the whole application. Intelligence can make the difference between a compelling business application and something that is “nice to have” but could be supported by manual processes or handcrafted in productivity tools like spreadsheets and documents. For example, core platform providers are building events-handling constructs and automation into claims management modules to expedite claims processing. Insurance-specific use cases for intelligent applications typically sit on a spectrum due to the complexity of risk. Simple underwriting cases for commoditized personal lines can be written on an STP basis, taking an exception-based underwriting approach for cases on the edge of the risk appetite. Simple property damage claims can be automated using CV and analytics for damage estimation, while serving up insights to support claims adjusters with more complex multiparty claims that require knowledge and high touch.
Mass: High
The vast majority of applications will ultimately become intelligent in some way via the use of AI technologies. Applications will incorporate one or more of the capabilities listed in the following bullet points, making the mass for intelligent applications very high. Intelligent applications that deliver better performance (namely, accuracy), user productivity and faster inferencing will provide a strong business case for users across business functions, geographies and industries.
Intelligent applications use AI in insurance in the following ways:
Data capture and response: AI technologies such as NLP, text analytics, deep neural networks and image recognition can be used for extraction of terms and conditions from policy wordings, or analysis of images in support of risk improvements or claims estimation. ■
Process augmentation: AI technologies like machine learning, decision intelligence, knowledge graphs and explainable AI can provide more intelligent actions for an application. In the future, process augmentation can be extended further to orchestrate interactions in the supply chain, automatically dispatching risk engineers, loss adjusters or other experts.
■
Gartner, Inc. | G00786204
Page 11 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
User experience: Conversational UI platforms and NLP, facial recognition and other AI applications can be used for understanding user emotions, context or intent and predicting user needs. This is essential in the claims channel to respond to the emotional intensity that comes with more severe personal lines claims, such as a house fire. ■ Process-embedded analytics: AI technologies like augmented analytics are plugged into a process to assist users or drive autonomous action. They will create more predictive and prescriptive analytics that can then be presented to users in the form of insights, or as guided recommendations for further evaluation. Guiding underwriting or claims technicians to process steps helps navigate the patchwork state-based legislation that insurers must adhere to. ■
Recommended Actions:
Take an insurance-specific, persona-based approach to adding AI capabilities to improve decision making, user experience and efficiency. Include features that provide value into underwriter or claims adjuster workflows that augment their approach to using current solution capabilities in an “evolution not revolution” concept. ■ Ensure the development of innovative and differentiated offerings by pursuing a broad continuum of near and adjacent technologies that influence the success of your intelligent applications. ■ Target line of business (LOB) prospects by hiring sales and marketing professionals who offer deep expertise and insights relating to the sector or core processes served by intelligent applications. ■
Recommended Reading:
Strategic Roadmap for the Composable Future of Applications ■
Top 10 Technology Trends Driving Change for Life Insurance CIOs in 2023, specifically the profile on “AI Adoption Focuses on Tools That Embed AI” ■
Natural Language Processing Back to Top
Analysis by Kimberly Harris Ferrante and James Ingham
Gartner, Inc. | G00786204
Page 12 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Description: Natural language processing (NLP) or natural language technologies (NLT) enable an intuitive form of communications between humans and systems. Specifically, NLP includes computational linguistic techniques aimed at parsing and interpreting (and sometimes generating) human languages. NLP techniques deal with the pragmatics (contextual), semantics (meanings), grammatical (syntax) and lexical (words) aspects of natural languages. The phonetic part of speech recognition is often left to speech- processing technologies, like speech to text.
Sample Providers: Amazon Comprehend; Google Cloud Dialogflow; IBM Watson Assistant; Microsoft Language Understanding (LUIS); Nuance; SAS
Range: Short (1 to 3 Years)
Gartner estimates that NLP use is also at 20% to 60% of the early majority target in insurance (see Note 1). Although a significant number of deployments have previously been observed, these are often for chatbots and speech recognition, and many of these deployments are likely to be narrow scope, decision-tree systems and not capable of true dialogue capabilities. NLP has been utilized and deployed inside P&C personal or commercial business units to extract content from large documents, while adoption in group life business, in particular, appears to be lagging. Content discovery for documents — including inspection reports or large policy documentation — is a rising use case that is gaining momentum in markets where AI adoption has typically been low. NLP has many emerging applications, too, in the contact center, for example, to assist in sentiment analysis. Specific use cases in the insurance industry include policyholder authentication, delivering policyholder assistance services, content extraction/discovery, and automating policy customization and personalization. In P&C commercial lines, NLP can be used to automate the process of reading property inspection surveys, which can run into hundreds of pages for complex specialty risks and global program business. This is a time- consuming aspect of the underwriting and rating process. Common challenges faced when using NLP include extracting semantic meanings in conversations, building vocabulary and generating dependency graphs from tagged parts of speech.
Mass: High
Gartner, Inc. | G00786204
Page 13 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
The use cases for NLP can include conversational user interfaces to improve personal and business efficiency, enhancing customer and employee interactions and satisfaction, transcription and translation, enablement of smart office/smart home, and even robotics. NLP will also support the transformation of human-machine interaction by enabling more natural voice communications. This can lead to significant disruption for existing products and technology providers that decline to incorporate NLP into their software, devices and robot solutions, since they will be at a competitive disadvantage or may even become obsolete. Insurers have frequently deployed NLP alongside chatbots and VAs for customer service. They’ve also used them alongside technologies such as ML and robotic process automation (RPA) in support of claims and fraud detection functions as sentiment analysis techniques are used to augment analysis of transactional data. NLP techniques are primarily deployed in policy issuance, customer service and sales/distribution, but can also be used in agent support, sales/distribution and billing/payment processing functions. Although some point solution deployments have been observed, particularly in individual life, more complex deployments have also been observed in property and casualty. These span multiple functional areas as insurers look to simultaneously scale and increase the efficiency of the business.
Recommended Actions:
Align with business stakeholders on accuracy and efficiency baselines for the process, which may vary across LOBs. Evaluate the entire front- and back- office elements of the service model in tandem to understand where and how NLP solutions can be integrated. ■ Treat NLP as a component that integrates with other platforms/applications, such as intelligent document processing solutions, and prioritize loose coupling based on workflow or event triggers. ■
Recommended Reading:
Hype Cycle for Natural Language Technologies, 2023
■
Gartner Fast Answer: What Should I Know About Natural Language Processing? ■
Gartner, Inc. | G00786204
Page 14 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
3 to 6 Years Advanced Virtual Assistants Back to Top
Analysis by: James Ingham and Annette Jump
Description: Advanced virtual assistants (VAs) execute complex tasks (or assist people), deliver predictions and accelerate decisions via deep contextual and domain-specific intelligent capabilities.
They are powered by a combination of:
More advanced user interfaces (like 3D and multimodal) ■
Domain-specific virtual assistants (VxAs) are a type of advanced VA designed to perform skilled, domain-specific tasks (like in healthcare, banking, retail or legal). VxAs incorporate customizable, pretrained language models (by task and industry) and integrate with enterprise applications and domain-specific systems. This enables VxAs to automate more complex high-value tasks and proactively engage with skilled professionals by offering some advisory capabilities — an expert system. Natural language processing (NLP) — multi-intent recognition, syntactic- and semantic-based methods, neural real-time machine translation, and synthetic voices ■ Semantic and deep learning techniques (such as DNNs), enabling decision support and personalization ■
Sample Vendors:
Virtual Enterprise Assistants: Amelia; boost.ai; IBM Watson Assistant; Kore.ai; Omilia; Oracle Digital Assistant ■
Virtual Customer Assistants: Amelia; boost.ai; Omilia; Soul Machines; Yellow.ai ■
Range: Midrange (3 to 6 Years)
Gartner, Inc. | G00786204
Page 15 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Advanced VAs are still three to six years away from early majority adoption in insurance. Gartner estimates that Advanced VA use lags other industries at 5% to 20% of the early majority target in insurance (see Note 1) with deployments observed in both life insurance and property and casualty insurance. However, in the last 12 months, we have seen noticeable advancements around prebuild enterprise software integration, advancing domain knowledge and enablement of multimodal functionalities. This helps to drive the expanding adoption of emerging use cases for specialized, domain-specific VxAs. VxAs possess a higher level of semantic intelligence and business process automation and higher containment rates, as well as provide proactive outreach and some end-user advisory capabilities. Specific use cases in the insurance industry include helping policyholders understand complex insurance jargon and engage with safe driving, usage-based auto insurance programs. Insurers have also used VAs to improve sales productivity and effectiveness of partner channels by increasing engagement with brokers. Virtual assistants have also been used in claims servicing to automate supply chain management, creating automated WhatsApp groups to introduce the loss adjuster and other group members to the customer.
Common challenges faced by organizations in adoption of advanced VA solutions include:
Lack of domain knowledge capabilities
■
Integration issues with relevant enterprise applications and data stores ■
Issues with organization’s acceptance
■
Insurers in particular have also suffered with overhyped/disappointing results with earlier VA pilots or proofs of concept (POCs) ■
Mass: High
Gartner, Inc. | G00786204
Page 16 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Insurers are using VAs alongside a combination of AI technologies, including NLP, chatbots and speech recognition in order to assist both policyholders and internal users. Virtual assistants are primarily used to support sales/distribution and customer service functions. Implementations can range from point solutions for customer service to more complex deployments in larger carriers supporting an increased range of functional areas, improving talent management as insurers scale the business. Advanced VAs can transform how employees interact with enterprise applications via conversational front ends to improve employee productivity, enhance consumer experience, and increase engagement with IoT and devices. Advanced VAs also have the potential to transform how employees interact with enterprise applications via conversational front ends and with advanced VAs identifying patterns in relevant business data, providing analytic insights and alerts/notifications based on real-time changes. Virtual customer assistants could be embedded in D2C customer portals and financial advisors. VxAs — expert systems — could be at the forefront of use augmentation for underwriters, claims adjusters and call center agents and also progress to simple task automation, as well. Various types include: quotation assistant, onboarding assistant, claims assistant, policy assistant and renewal assistant. A VxA could also serve as a powerful tool for lead generation — engaging prospective customers and offering personalized policy recommendations on the basis of their profile and requirements. However, at this time, these persona or value-chain-based offerings are in their infancy, and Gartner has not observed production deployments by insurers of these types of VxAs.
Recommended Actions:
Develop persona-based VAs that assist insurance-specific roles, workflows and everyday tasks, over general insurance VAs for knowledge and analytics. ■
Expand advisory and proactive capabilities of your software by leveraging advanced VAs and pursuing deployment alongside P&C or life insurance core platforms. Establish the API coverage and ease with which integrations can be scheduled, which may vary dramatically between established players and new entrants in this market. ■
Recommended Reading:
Emerging Tech Roundup: ChatGPT Hype Fuels Urgency for Advancing Conversational AI and Generative AI ■
Gartner, Inc. | G00786204
Page 17 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Emerging Technologies: Top Business Value Patterns in Advanced Virtual Assistant Adoption ■
Emerging Technologies: Vendor Differentiation Patterns in Virtual Assistant Technologies ■
Decision Intelligence Back to Top
Analysis by: Moutusi Sau, Kimberly Harris-Ferrante, Alys Woodward, Erick Brethenoux
Description: Decision intelligence (DI) is a practical discipline used to improve decision making by explicitly understanding and engineering how decisions are made and how outcomes are evaluated, managed and improved via feedback. In the current dynamic and complex business environment, with the increasing pace of business, DI provides a way to focus the need for information, collaboration and feedback. Common use cases for decision intelligence include the use of predictive modeling and AI to support underwriting and claims next best action or to guide human decision makers through the process, including CSRs delivering customer service. AI is an important component of DI in the form of ML techniques. Legacy decision management tools used rule-based optimization, but to address greater complexity of decisions, decision intelligence must incorporate AI, as well as must be real time to shorten the time to make decisions. This is especially true for employees that are actively engaged with customers such as in the contact center or as a claims adjuster.
Sample Vendors: ACTICO; Exponentia.ai; FICO; Fractal Analytics; Kalepa; Quantexa
Range: Midrange (3 to 6 Years)
Decision intelligence is three to six years from early majority adoption. This is due to lack of proper coordination between business units and the inability to impartially reconsider critical decision flows within and across departments that diminish the effectiveness of early decision intelligence efforts. In insurance, there is an overall lack of understanding about what decision intelligence is and how it can help organizations in driving business results. Most companies are focusing on the use of AI in other areas, especially around customer engagements, rather than employee empowerment, which will also slow down adoption.
Gartner, Inc. | G00786204
Page 18 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
General AI immaturity can mean decision intelligence seems too ambitious and distant. However, the more advanced organizations from which less mature organizations learn best practices are embracing decision intelligence more fully. Common challenges faced when using DI include the unpredictability of the sustainability of organizations’ decision models, which are based on the quality of transparency. Autonomous decision models require greater scrutiny at design time to avoid negative outcomes and higher expenses. Organizations need decision intelligence — however, their propensity to invest depends on the level of regulation, risk control and analytical maturity. Not until insurers place a greater focus on employee engagement and decision accuracy will the investment in decision intelligence rise.
Mass: High
The combination of AI techniques and the confluence of several technology clusters around composite AI, smart business process, decision management and advanced personalization platforms are creating a new market around decision systems platforms supporting the DI discipline. The technology that underpins DI includes many data and analytics technologies. However, the discipline focuses specifically on driving business value from supporting and improving results from strategic (fewer, high risk and broad in scope) and operational (numerous, individually low risk and narrow in scope) decisions. The mass is very high because, although these capabilities will be brought to market by multiple different providers, the decision intelligence trend is very broad in impact. Advancements, however, in industry niche solutions may accelerate use of decision intelligence in insurance. Instead of opting for horizontal solutions, niche industry vendors such as those offering underwriting workstations or claims management are slowly enhancing their solutions to offer more predictive analytics and some AI. For example, fraud vendors offer fraud scoring with combined litigation management, which offers enhanced decision making for fraud special investigators. CRM vendors are adding AI to their customer servicing solutions that will script the CSR through discussions using a variety of customer segmentation and next best action models. The more that industry solutions are enhanced with decision intelligence, the easier it will be to bring it to buyers and mid-market buyers.
Gartner, Inc. | G00786204
Page 19 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Long-term decisions such as strategic investments will require a different set of technology components supporting operational decisions via continuous real-time intelligence. Hence, the need for a wide range of technology components to support DI across both tactical and strategic decisions. The impact of DI will be to advance the use of data and analytics and AI-related components (such as ML) by providing a broad discipline for organizations to use as a framework.
Recommended Actions:
Target specific decisions and build solutions to support those decisions to convince prospects early in the sales cycle of the measurable value that the solutions will offer. Work with the business to identify best application areas based on acceptance to change, use of advanced data techniques, availability of data and desired business outcomes (e.g., managing talent and skill gaps, driving underwriting profitability or enhancing the customer experience in the call center).
■
Avoid the “nice to have” trap of providing general technology platforms. Focus on removing costly manual effort and increasing automation in low-value and low-risk decisions as a steppingstone to more advanced decision intelligence. ■
Recommended Reading:
Innovation Insight for Decision Intelligence
■
Insurance Scenarios: Sense and Plan for Alternative Futures in an Era of Constant Change ■
When to Automate or Augment Decision Making
■
Generative AI Back to Top
Analysis by: Kimberly Harris-Ferrante and Svetlana Sicular
Description: Generative AI refers to AI techniques that learn a representation of artifacts from data, and use it to generate brand-new, unique artifacts that resemble but don’t repeat the original data. These artifacts can serve benign or nefarious purposes. Generative AI can produce totally novel content (including text, images, video, audio, structures), computer code, synthetic data, workflows and models of physical objects. Generative AI also can be used in art, drug discovery or material design.
Gartner, Inc. | G00786204
Page 20 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Sample Vendors: Amazon; Anthropic; Adobe Sensei; Google DeepMind; Grammarly; IBM; Microsoft; MOSTLY AI; NVIDIA; OpenAI; Phrasee
Range: Midrange (3 to 6 Years)
Range is three to six years because, while generative AI is becoming accessible, many generative techniques are new and more are coming to the market. Reproducibility of generative AI results will be challenging in the near term. A fragmented, specialized and rapidly evolving technology provider landscape (many small tech startups) and offerings (such as generating only images or only text) currently require a combination of tools rather than a single solution. Compute resources for training large generative models are high and are not affordable to most vendors. Generative adversarial networks (GANs), variational autoencoders, autoregressive models, diffusion AI models and zero/one/few- shot learning have been rapidly improving generative modeling while reducing training data requirements. Insurers have begun experimenting with generative AI already, with potential across the value chain for internal and customer-facing use cases. While concerns over validating of results and governance are high, the long-term outlook for use in the industry is positive. Generative AI has the potential to improve document processing, customer self-service, marketing, data science and operations such as claims, underwriting and product filings, for example. Safety concerns and negative use of generative AI, such as deepfakes, might slow adoption in some industries and slow down the use in customer-facing applications. As human validation is required in many cases, it is likely to be used more for human augmentation to drive knowledge worker productivity and decision making in areas such as risk selection, customer servicing or claims handling. Technologies that provide AI trust and transparency will become an important complement to the generative AI solutions.
Mass: High
The mass is high, because in insurance, there are many business impacts from helping with competitive intelligence gathering in marketing, personalization or products/services, improved cross-sell/upsell with agents, and improved catastrophe and underwriting analysis. Adoption of AI is already high, with growth anticipated over the next few years. The use of generative AI will accelerate AI’s use overall in the industry but also drive maturity as insurers adopt new technology types and build out new use cases (including heightened use of unstructured content).
Gartner, Inc. | G00786204
Page 21 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
In the end, it will assist employees with decision making and productivity, as well as helping improve customer satisfaction and online closure for digital channels. Insurance is a traditional business with many opportunities still for enhanced automation, such as the use of intelligence document processing (IDP). Generative AI will be leveraged by vendors in this market, introducing it to insurers as they purchase new automation tools. The use of digital channels, especially chatbots, is increasing as insurers seek to improve customer interaction and promote self-service. The incorporation of generative AI into chatbots to improve usability and outcomes will drive greater use of generative AI within the industry. Customer experience solutions will also leverage generative AI to enhance call center interaction. Synthetic data that is produced using generative AI techniques supports the accuracy and speed of AI delivery. Synthetic data draws customer and partner attention by helping them augment scarce data, mitigate bias or preserve data privacy. Gartner expects synthetic data to be available as part of most AI platforms. We predict that by 2025, synthetic data will reduce personal customer data collection, avoiding 70% of privacy violation sanctions. However, generative AI has limitations — ensure you do not overuse synthetic data, for example, when you need a real “ground truth.” Generative AI will disrupt software coding. When combined with existing development automation techniques, it has the potential to automate up to 70% of the work done by programmers. ML and NLP platforms are introducing generative AI capabilities, along with transfer learning for reusability of generative models, making them accessible to customers.
Recommended Actions:
Assess automation strategies to identify human-based work that can be augmented through the use of generative AI, including roles such as underwriting, claims, compliance, distribution or the contact center. Identify how generative AI can help with talent issues and productivity enhancement needed for each role. ■ Providers for insurance core systems, CX solutions, chatbots and IDP/automation should prioritize opportunities where generated AI data could benefit your existing product offerings. For example, it might be used to develop new digital channel capabilities that are tightly integrated with systems of record and harmonized data. ■ Examine and quantify the advantages and limitations of generative AI by analyzing where generative AI could bring breakthroughs, as it requires skills, funds and caution, then weigh technical capabilities with ethical factors. ■
Gartner, Inc. | G00786204
Page 22 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Build teams to continue to assess generative AI’s impact on security and compliance, and build governance and oversight to ensure ethical and regulatory risks are addressed. ■
Recommended Reading:
Innovation Insight for Generative AI
■
Client Question Video: What Do I Need to Know About Generative AI? ■
Top Strategic Technology Trends for 2022: Generative AI ■
Gartner’s Top Strategic Predictions for 2023 and Beyond — Seizing Uncertainty ■
Quick Answer: Will Machine-Learning-Generated Code Replace Developers? ■
Emerging Tech Roundup: ChatGPT Hype Fuels Urgency for Advancing Conversational AI and Generative AI ■
Emerging Tech: Venture Capital Growth Insights for Generative AI ■
Emerging Tech: Generative AI Needs Focus on Accuracy and Veracity to Ensure Widespread B2B Adoption ■
ChatGPT Research Highlights
■
Graph Technologies Back to Top
Analysis by: Moutusi Sau, Alys Woodward, Robin Schumacher, Sharat Menon, Jim Hare
Description: The term “graph technologies” refers to graph data management and analytics techniques, which enable the exploration of highly connected data, specifically, the relationships between entities such as organizations, people or transactions. Analyzing relationship data can require a large volume of heterogeneous data, storage and analysis — all of which is not well-suited to relational databases. Graph analytics consist of models that determine the “connectedness” across data points. These range from simple node, edge traversal and triple pattern matching for transactional uses, to complex multihop queries, reasoning and inference, and algorithms for analytical workloads.
Gartner, Inc. | G00786204
Page 23 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
The rise of Graph Structured Query Language (GSQL), Apache TinkerPop Gremlin and SPARQL Protocol and RDF Query Language (SPARQL) as a Structured Query Language (SQL) interface that queries graph databases will be pivotal to increasing the adoption of graph technologies. This is particularly true among the business/data analyst community that’s already familiar with using SQL.
Sample Vendors: Amazon; Cambridge Semantics; DataStax; Progress (MarkLogic); Microsoft Azure Cosmos DB; Neo4j; TIBCO Graph Database
Range: Midrange (3 to 6 Years)
The range for graph technologies is three to six years from early majority adoption across the total addressable market, due to the wide range of possible applications for graphs and the complexity of addressing them. Graph technology can help in many areas such as improving internal claims handling. The innovation can enable faster, broader searches for data to identify potentially fraudulent activity on a claim. Graph enables data-driven decisions and consistency in claims handling. One primary outcome of this effort is a projected reduction in claim cycle time and improved claim payment accuracy. Graph enables an investigator to deeply explore relationships surrounding a claim: every person, every vehicle, every incident report, every policy. For example, there could be a legacy claim, which is fraudulent — and a connection from the person associated with that claim to a new claimant — the relationship might be two or three times removed, but a graph can uncover this nonobvious insight. Some startups are now starting to productize solutions to apply graph technologies to solve for shortest path, pattern identification, next best action and compliance use cases to support insurance. Despite the rise in graph analytics solutions that make it possible to query graph solutions using SQL, there is still demand for new skills related to graph- specific knowledge, which currently restricts growth in adoption. Insurers generally lack AI and cloud skills, which will hamper use of tools in-house. The new skills required include knowledge and experience with the Resource Description Framework (RDF), property graphs, the Gremlin graph query language, SPARQL Protocol and RDF Query Language (SPARQL), as well as executing graph analysis in Python and R.
Mass: High
Gartner, Inc. | G00786204
Page 24 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Graph technologies continue to show increased demand globally, albeit focused on specific industries such as finance or retail. Established AI techniques (such as Bayesian networks) are increasing the power of knowledge graphs and the usefulness of graph analytics through further nuance in representational power. Graph databases are ideal for storing, manipulating and analyzing the widely varied perspectives in the graph model due to their graph-specific processing languages and capabilities, scalability and computational power. Graph technologies help insurers to solve problems that would otherwise require manual interrogation and analysis, such as investigating complex fraud networks involving multiple parties, combating money laundering and reserve reporting. The additional scalability and computational power of graph databases also help enable next best action and product recommendations based on life events, and will eventually underpin future business models such as panoptic personalization. However, insurance companies must avoid misuse of customer information, digital creepiness, misselling, and breaking ethics and privacy laws (see Panoptic Personalization: An Insurance Trend for 2022).
Recommended Actions:
Prioritize insurance customer use cases that improve short-term claims processing, shorten SLAs and improve customer satisfaction by focusing on analytical query support for process optimization and straight-through processing. ■ Focus in the long term on applying graph technologies to future insurance business models where customer product and channel interactions feel personalized to a customer’s individual needs, taking into account the relationship with the closest entities. ■
Recommended Reading:
Understanding When Graph Analytics Are Best for Your Business Use Case ■
Tabular Synthetic Data Back to Top
Analysis by: James Ingham, Alys Woodward, Vibha Chitkara, Benjamin Jury
Gartner, Inc. | G00786204
Page 25 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Page 1 Page 2 Page 3 Page 4 Page 5 Page 6 Page 7 Page 8 Page 9 Page 10 Page 11 Page 12 Page 13 Page 14 Page 15 Page 16 Page 17 Page 18 Page 19 Page 20 Page 21 Page 22 Page 23 Page 24 Page 25 Page 26 Page 27 Page 28 Page 29 Page 30 Page 31 Page 32 Page 33 Page 34 Page 35 Page 36 Page 37 Page 38 Page 39 Page 40 Page 41 Page 42 Page 43 Page 44 Page 45 Page 46 Page 47 Page 48 Page 49 Page 50-51Made with FlippingBook - PDF hosting