Industry
Insurance
Reinventing the Insurance Industry with a Digital-first Strategy
Products & Platforms
TCS BFSI Platforms
TCS BFSI Platform: Transforming Financial Services for Better CX
Highlights
- The rise of GenAI and large language models (LLMs) has heralded a new era for system architecture, enabling a unified conversational user interface for various user personas.
- AI agents have begun to support automated backend processes. Agents can read, retrieve, and reason over dense wordings, endorsements, and jurisdictional rules, thereby reimagining the traditional insurance platform.
- This paper analyzes the shift from traditional insurance system design to a modern approach that uses GenAI, large language models, agentic systems, and self‑adapting user interfaces. It also includes a future readiness roadmap to build a user-focused platform.
Overview
With the rise of GenAI and LLMs, industries are reimagining their IT platforms and realigning their business models to achieve key transformational goals in their operating models.
Typically, such realignment leads towards key shifts in the system architecture, signaling the advent of a unified conversational user interface for various user personas; and, AI agents supporting automated backend processes orchestrated in a business-aware manner. The key goals for insurance CXOs are:
- Adopt a paradigm shift from operational workflows to outcomes: Today’s customer experience needs operational processes to move beyond static quote/claims flows into user intent and plans of action, through a system that understands policy language, orchestrates intelligent automation tools and thus delivers outcomes with guardrails.
- Provide an experience to stakeholders through insurance domain and context- driven auto-adaptive layers: Decision-oriented guided journeys for internal and external users are critical. Self-adaptive UI for underwriters, adjusters, brokers, and policyholders − coupled with agentic orchestration grounded in policy wordings and underwriting rules, and retrieval-centric data planes with permissions− are technology enablers that can achieve an easy and assisted experience.
- Implement built-in governance as protection to autonomous behavior of machine-centric intelligence: While machine intelligence is bringing in efficiency and improved experience, checks and controls are paramount − to moderate machine-driven decisions, and ensure transparency and non-discrimination. Finally a human-in-the-loop is necessary to ensure bind, payments, and large loss reserves to protect the interests of the carrier’s business outcomes and ethical image.
- Create a holistic business impact: The focus is gradually shifting for carriers towards higher straight-through processing (STP) rates, faster first notice of loss (FNOL) to settlement, reduced leakage and fraud, and experiences that earn trust and address customer vulnerability. This is specifically in demand to ensure top line and bottom-line goals, manage customer expectations, and minimize churn.
Industry challenges
Insurance is ripe for agentic, adaptive systems for many reasons, ranging from complex policy language to fragmented tools.
Let’s have a closer look at these:
- Policy complexity: Coverage is encoded in dense wordings, endorsements, and jurisdictional rules. Agents can read, retrieve, reason, and summarize them.
- High stakes decisions: Underwriting, pricing, and claims require oversight, lineage, and auditability—ideal for architectures with explicit guardrails.
- Fragmented tooling: Core insurance administration systems, rating, billing, customer relationship management (CRM), document generation, e-signature, know your customer / anti-money laundering (AML), and data bureaus are all prime for orchestration of intelligent and safe tools with a defined scope of boundaries and agentic flow.
- Human moments: Claims are emotional. A UI that adapts and explains the “why” in the user journey builds trust when it matters most.
Solution and benefits
Insurers' vision now requires reimagining the insurance platform architecture.
Figure 1, shows the shift from traditional reference architecture to next-generation reference architecture, embedding the contextual benefits of GenAI, LLM, agentic systems, and self-adaptive user interface. Notable changes in this next-generation reference architecture are described below:
Figure 1: Traditional to next-generation reference architecture
Experience layer: Self-adaptive conversational UI demands the following key outcomes from the agentic system:
- Persona-aware surfaces: User interfaces like underwriter workstation, adjuster console, broker/partner portal, contact center desktop, and policyholder app should now be able to adapt not just based on persona but also on the business process context and applicable historical insights-driven guided assistance within the user journey.
- Intent capture in user interface: Users are now more interested in leveraging natural language (“Add flood coverage to my home”), structured forms, device context, and policy history more in the context of interaction with the self-adaptive UI.
- Built-in trust in user experience: Facilities like inline citations to policy clauses, underwriting guidelines, and claim manuals; “why we asked” explanations; clear risk/impact indicators, are now a must to build user trust and empathy along the user journey.
Business components as domain agents with agentic orchestration: Automated orchestration AI agents of various types will form the new view of the business component layer.
- Planner-executor for business process agents: Agentic architecture mandates deconstruction of user journey goals into automated tooling for minimal user interaction through voice, natural language, and documents – with a wait only required for human-centric approvals.
- Specialist automation agents: Automation agents enable the system to leverage the power of LLMs and GenAI to undertake an intelligent process entirely by the system. Examples of such agents are:
- Risk underwriting and appetite checks and risk-data enrichment based on automated interface calls or information extraction from documents.
- Rating/premium calculation agent, requirement generation and processing during underwriting or claims triage/coverage determination, subrogation, and fraud detection.
The business component layer will also enable key architectural patterns to adopt cognitive agents to leverage the LLM and communicate across systems. Key architectural aspects that would be undertaken to leverage GenAI features mapped to specific use cases are listed below:
- Model gateway: Route to LLMs, re-rankers, OCR for documents, vision computing for damage estimation, and speech for call summaries and synthesization.
- Structured reasoning: Function calling into agents like rating engines, geocoding/peril modelling, fraud scoring, and document generation.
- Model context protocol (MCP): Contextual data ingestion, enrichment, and validation within the agents through interactions with third-party systems enabled through a unified gateway, thus alleviating the well-known NxM problem, where N represents the LLM-powered business AI agents, and M represents the various third-party systems.
Data platforms in new age architecture will have various AI-specific data planes in conjunction with modernized data planes for the real-time generation and processing of data insights and the embedding of such insights into user journeys for contextual predictive guidance for the user.
The focus in an agentic, self-adaptive system will shift from a traditional enterprise data platform with staged data processing and a disjointed data event streaming for specific use cases. The need for self-adaptive behavior in the user interface now necessitates real-time event and data streaming, real-time curation of enterprise data assets with data quality assured through the Medallion architecture model, which ensures high data quality through multi-step refinement,and finally, real-time generation of machine learning insights on the data features engineered from the ingested/ streamed data into the enterprise data platform.
The following data planes will now combine the transactional databases alongside the vector databases and data structures used for AI agentic processing:
- Vector store: Embeddings for policies, endorsements, claim notes, call transcripts, and knowledge articles with permissions metadata.
- Knowledge graph: Entities and links—customer, policy, risk object, coverage, peril, claim, provider, broker, reinsurance treaty, and lineage to sources.
- Operational stores: Core insurance administration system for truth, billing, claims, CRM; feature store for pricing/propensity; event streams for telemetry and feedback loops.
With the above considerations, a mapping of the shift of architectural principles in traditional architecture to next-gen architecture is presented in Table 1:
Principle dimension |
Traditional architecture |
Next-gen architecture (Intent/plan/act) |
User interaction |
Static forms, fixed journeys |
Conversational, goal-oriented, adaptive journeys |
Processing logic |
Deterministic process services |
Domain agents with tools and humans in the loop approvals |
Process knowledge |
Scattered PDFs/wiki |
Retrievable, permissioned, and cited cognition chunked into vector DBs |
User interface |
Contextual persona apps |
Persona-aware, confidence-aware, accuracy-aware, and self-adaptive conversational chat through voice or text |
Governance |
Afterthought audits |
First-class policy, process and data lineage and automation-driven controls |
Outcomes |
Low STP, high handling time |
Higher STP, higher agentic automation, faster cycles, fewer handoffs |
Table 1: Comparison of architectural principles from traditional to next-gen agentic architecture
Some of the core solution patterns which would be applied to the insurance domain are listed below:
Retrieval augmented generation (RAG) over insurance knowledge.
Various types of information retrieval and generation steps will be undertaken to build the insurance intelligence within the system:
- Various heterogeneous retrieval data types like vectors, keywords, and graph hop will be used across business entity information like policies, endorsements, jurisdictional/ business rules, and claims manuals.
- Include structured data snippets like clauses, effective dates, coverage limits, and authorization scopes to contextualize the data with respect to the business process.
- Reranking and freshness of the embedded data will yield reduced hallucinations and also be adaptive in nature to respect business changes such as midterm endorsements, moratoriums, etc.
Self-adaptive UI mechanics.
- Confidence-aware flows: Conversational flows will now need to consider the accuracy and confidence around the contextual outputs produced by the AI agent to the chat window. Outputs with high confidence will trigger fewer questions and fewer iterations to reach the end business outcome, while outputs with low confidence must show conditions, clauses, and options to enable the user to take actions, as well as make the AI agent learn further from the user response.
- Role-aware composition: Contextual contents will be required to be generated and provided in a persona-driven manner. For example, reserve history and special investigation unit flags will need to be provided to the adjusters to enable them to assess the claim correctly, while brokers will be required to be supplied with an explainer on the customer’s insurance appetite, along with citations to relevant documents. Similarly, a customer/policyholder will now see plain language guidance on the ‘why, what, and how’ of the insurance policy that is being purchased and/ or serviced by the insurance company.
Data architecture: The backbone for this agentic solution will rely heavily on various types of data stores based on the business context, purpose, and controls, and due considerations will need to be implemented for each of these specific types of data stores, as shown in Table 2.
Store |
Insurance-specific purposes |
Notes |
Vector DB |
Semantic search over wordings, claims, transcripts |
Enforce row level permissions; attach clause IDs |
Knowledge graph |
Entities/relationships and lineage |
Supports provenance and regulatory audits |
Document store |
Source of truth for artifacts |
Enables, versioning and jurisdiction tags |
Feature store |
Pricing, fraud, propensity features |
Allows privacy controls and drift monitoring |
Event stream |
Telemetry and feedback |
Drives evals, SLA alerts, and retraining signals |
Table 2: Various data store types in next-gen agentic architecture, their applicability in insurance business, and contextual controls.
Data design and access need to be planned end-to-end to ensure retrieval respects the line of business, role, and tenancy.
Future readiness
A typical architecture adoption roadmap for the carrier will follow a gradual shift to maturity and adoption path to an agentic world, as discussed below:
- Start with grounded copilot/assist.
- Claims copilot/assist in adjuster console: Summarize, retrieve clauses with citations, and draft communications.
- Underwriting assistant: Appetite checks and submission summarization.
- Add automation tooling.
- Read-only business transactions first (quote comparisons, reserve suggestions), then write business transactions with approvals (endorsements, payments) in a controlled manner.
- Planning for contextual business agents.
- Re-imagine claims journey step by step starting with automatic FNOL using computer vision, followed by agents for automated claims triage, automated estimation and reserve creation, and finally, upon human checkpoints within authority boundaries, and agents for payment disbursal and settlement of the claims.
- Renewal planning with ‘what-if’ scenarios and broker negotiation support.
- Adopt self-adaptive UI
- Introduce conversational- and confidence-aware chat forms, dynamic required docs, and citation-based explanations across all user segments.
Key KPIs and use case-specific outcome improvements that would matter in the transformed business model are:
- STP rate increase: Quotes-bound and claims settled without human touch for low-risk flows.
- Reduction in cycle time: FNOL to payment; quote to bind; reserve adjustment latency.
- Leakage reduction: Claims overpayment variance and missed subrogation recoveries.
- Improvement in net promoter score (NPS)/customer satisfaction: Especially during claims, measure trust via “explanation helpfulness.”
- Fairness: Disparate impact metrics across protected classes where applicable.
- Improvement in overall cost of operations: Reduction in handling time, rework rate, and cost per transaction.
Insurance supports people when they need it the most. Design and decision systems for adaptable agents, self-adaptive user interfaces, and transparent decisions in an agentic world of next-gen insurance solution architecture not only streamline processes but also build a reliable, user-focused software platform.
TCS is already realizing the Agentic AI insurance reference architecture through targeted investments in its insurance products and solutions. TCS BaNCS™ for Insurance features adaptive conversational UI agents and AI-driven business process agents that replace human-intensive tasks like quotations, underwriting, claims intake, and claims assessment.
With TCS BaNCS™ for Insurance, a fully agentic platform complete with self-adaptive user interfaces, orchestrated AI agents, and MCP-based integrations is within reach, empowering global insurers in their transformation journey.