Why AI in insurance CRM matters for carriers in 2026
In my experience working with VP Distribution and IT architects at US P&C carriers between $500M and $5B GWP, the AI-in-insurance-CRM conversation in 2026 has shifted in a way that surprises some executives. Two years ago, the question was whether to invest in AI capability for the CRM. The carriers that asked that question and decided to wait have lost ground every quarter to competitors who moved faster. The question now is which AI use cases to prioritize, how to govern the models in production, and how to avoid the failure patterns that have shown up in the carriers who moved without a framework.
McKinsey’s July 2025 report on the future of AI for the insurance industry frames the stakes bluntly: AI leaders in the insurance industry produced a total shareholder return 6.1 times higher than laggards over five years, a spread wider than in most other sectors. McKinsey also estimates that generative AI alone could produce $50 to $70 billion in additional insurance industry revenue, with the largest impact across marketing, customer operations, and software engineering. That is the carrier-side context: AI is not a feature checklist item; it is a strategic capability that compounds outperformance over time.
I have spent close to ten years building producer-facing systems and the AI integration patterns underneath them, including the eAgent platform Decerto built for Warta (Talanx Group) serving 40,000 producers, the lead management platform that powers Warta’s customer engagement and cross-sell triggering, the IRON sales platform for InterRisk (Vienna Insurance Group), and the Higson product configurator that runs the product catalog at Allianz Poland. The AI use cases I describe in this article are not theoretical for me. They are the working capabilities I have seen succeed in production at carrier scale, alongside the AI deployments that did not produce the expected outcomes and the governance gaps I have watched cost carriers time and money to fix.
This guide covers the three AI capability waves shaping insurance CRM, the five highest-impact CRM-side AI use cases for mid-tier P&C carriers, the NAIC model governance overlay that turns AI from regulatory exposure into competitive advantage, the things AI in insurance CRM cannot do yet, and how to sequence the investment. It is written for VP Distribution, VP Marketing, Chief Data and Analytics Officers, IT architects, and product managers leading the AI strategy for customer engagement. It is opinionated.
I am Maciej Wir-Konas, Head of Agent Portal at Decerto.
The state of AI adoption in insurance CRM today
Before describing where AI in insurance CRM is going, it helps to know where it is. The benchmarks below are anchored to McKinsey’s July 2025 report, NAIC carrier surveys, and the Aite-Novarica producer engagement research.
Predictive analytics is essentially universal
By 2026, close to 80 percent of insurers rely on advanced rating and pricing models, with an additional 11 percent planning to implement them soon. Predictive analytics for fraud detection (currently 33 percent) and severity assessment (currently 29 percent) are expected to reach 65 to 70 percent within the next two years. These are the carrier-side analytical AI capabilities that have been maturing for a decade. They are baseline, not differentiating.
Generative AI adoption is accelerating
Over half of insurance carriers report already using large language models or generative AI in production, and another 29 percent plan to begin adopting these technologies within the next two years. The use cases that have shipped first: claims document summarization, customer service response drafting, marketing content generation, and submission ingestion automation. The CRM-specific use cases - conversational interfaces, embedded copilots, content personalization at scale - are accelerating in 2026.
Agentic AI is the emerging third wave
Agentic AI accounted for 21 percent of public AI deployments in the insurance sector in Q4 2025, with the majority focused on claims management. The CRM-specific agentic AI use cases - autonomous customer outreach, multi-step workflow execution, cross-system orchestration - are early but maturing fast. McKinsey describes the progression as “rule-based, machine learning, agentic” - the third wave is where competitive differentiation is moving.
The adoption gap is widening
The same McKinsey research shows a sharp divergence between AI leaders and AI laggards. Leaders are building modular, agent-enabled stacks with domain-centric rewiring. Laggards are running scattered pilots that produce demo-ready capabilities and few production outcomes. The 6.1× TSR spread is the financial signal of that divergence. Mid-tier P&C carriers in the $500M to $5B GWP range are mostly in the laggard cohort by McKinsey’s framing, with execution discipline rather than capability being the primary gap.
The CRM-specific implication
Insurance CRM sits at the intersection of customer engagement, distribution operations, and data infrastructure - the three domains where McKinsey identifies the highest AI productivity impact. Carriers that build AI into the CRM as a foundational capability rather than a phase-two enhancement compound advantage over time. We covered the underlying CRM architecture in our insurance CRM definitional guide and the feature-level evaluation framework in our insurance CRM features evaluation guide.
The 3 AI capability waves - predictive, generative, agentic
I find the three-wave framework useful for thinking through CRM AI strategy. The framework is mine, informed by the deployments referenced later, and it maps cleanly to where AI value compounds for mid-tier P&C carriers.
Wave 1 - Predictive and analytical AI
The first wave includes the machine learning techniques carriers have used for over a decade: classification (this customer is at risk of non-renewal), regression (this risk has a 7 percent expected loss ratio), clustering (these customers behave similarly), recommendation (this customer is likely to accept an umbrella policy offer). The models are typically trained on historical data, run as scoring services, and feed downstream workflow decisions.
For insurance CRM, Wave 1 powers customer segmentation, churn propensity scoring, cross-sell propensity scoring, lifetime value estimation, and lead scoring. Mid-tier carriers should treat Wave 1 capabilities as table stakes by 2026; a CRM without them is materially behind the market.
Wave 2 - Generative AI
The second wave adds large language models and the ability to work with unstructured data - documents, emails, call transcripts, customer notes, free-text descriptions. The models can summarize, classify intent, generate draft content, extract entities, and answer questions against domain knowledge.
For insurance CRM, Wave 2 powers conversational interfaces (chatbots, virtual assistants), customer service response drafting, marketing content personalization at scale, document understanding (summarizing claim files, parsing ACORD forms), and the embedded copilots that help service reps and producers do their daily work. We covered the embedded virtual assistant feature in Section 6.3 of our insurance CRM features evaluation guide.
Wave 3 - Agentic AI
The third wave moves from AI as a tool that responds to prompts to AI as an autonomous agent that executes multi-step workflows across systems. An agentic AI for a customer service rep might receive an inbound call inquiry, pull the customer profile, identify the relevant policy, check claims and billing for context, draft a resolution, take the action in the connected system, and document the outcome - all autonomously with human oversight at decision points.
For insurance CRM, Wave 3 is early but accelerating. The 2026 use cases that are shipping: autonomous follow-up on cross-sell offers, multi-step renewal workflows with human escalation rules, cross-system orchestration where the CRM coordinates actions in the PAS, claims, billing, and the agent portal automatically. We covered the orchestration capability in Section 6.5 of our insurance CRM features evaluation guide.
The sequencing principle
In my experience working with mid-tier carriers, I recommend approaching the three waves in order. Get Wave 1 predictive capabilities working at production scale, with proper governance, before moving to Wave 2 generative use cases. Get Wave 2 generative use cases working with human-in-the-loop oversight before moving to Wave 3 agentic deployments. Carriers that try to skip Wave 1 to launch agentic AI demos usually end up with capabilities that look impressive at the executive review and underperform in production because the underlying analytical foundation is weak.
The rest of this guide walks through five specific CRM-side AI use cases mapped to the three waves, with implementation notes from production deployments.
AI use case 1 - Customer intelligence and segmentation
The first AI use case where mid-tier carriers should invest. It is mature, well-understood, and produces measurable retention and revenue lift within the first year of deployment.
What the capability does
The CRM uses AI to score every customer on dimensions that matter operationally: churn propensity (probability of non-renewal in the next 90, 180, 365 days), cross-sell propensity (probability of accepting an offer for each product line), claim propensity (probability of filing a claim in the next 12 months), lifetime value (expected discounted future premium minus expected losses and servicing cost), and risk profile evolution (whether the customer’s risk is increasing or decreasing).
The scores update continuously as new data arrives from the PAS, claims, billing, agent portal, and marketing automation systems. They feed downstream workflow decisions - who to call this week, which renewals to flag for early attention, which cross-sell offers to surface to producers, which customers to suppress from outbound campaigns.
The data foundation requirement
Customer intelligence AI works only as well as the underlying data foundation. A CRM that cannot resolve the same customer across PAS, claims, billing, and marketing systems produces scores that disagree with operational reality. We covered the customer data foundation specifically in our Agent 360 architecture guide - identity resolution is the prerequisite for everything that follows.
The implementation pattern that works
In my experience deploying customer intelligence AI at carrier scale, start with one score that drives one downstream decision. The most common starting point: churn propensity feeding the renewal review queue. The model scores all upcoming renewals, the renewal team prioritizes outreach to the highest-risk customers, and the carrier tracks retention lift over 6 to 12 months. Once that loop is working, add the next score and the next decision.
The implementation pattern that does not work: build a customer intelligence platform with twelve scores, hand it to the operations team, and assume they will use them. The scores become a dashboard that nobody acts on, the AI investment is justified by potential rather than realized impact, and the next executive review questions the program.
The retention business case
McKinsey reports that AI-driven digital lead generation and targeting have reduced customer churn by up to 50 percent in some cases by engaging clients with the right message at the right point in their journey. That is the high end of what is achievable; the median carrier sees retention improvements in the 5 to 15 percent range from a working customer intelligence program. Either way, the business case is compelling for a $500M to $5B GWP carrier where each retention percentage point is meaningful.
The compliance overlay
Customer intelligence models that drive customer-impacting decisions fall under NAIC model governance principles. The model has to be documented, the training data has to be auditable, the model performance has to be monitored, and the fairness implications have to be reviewed. We cover the governance framework specifically in Section 9.
AI use case 2 - Cross-sell propensity and retention scoring
The cross-sell-and-retention use case sits closely alongside customer intelligence but deserves its own treatment because the operational program differs and the business case math is different.
The cross-sell propensity model stack
Three models work together for a cross-sell program: (a) life event detection that identifies the triggering moment for an offer - the eight life events we covered in our cross-sell playbook for carriers; (b) product propensity scoring that estimates the probability the customer would accept each available cross-sell offer given their profile and the trigger context; (c) channel preference scoring that estimates the optimal channel (producer call, mobile push, email, direct mail) for each customer for each offer type.
The retention scoring model
The retention model scores every active customer on probability of non-renewal in the next renewal cycle. The features that matter most in practice: premium increase relative to last renewal, claim experience in the prior 24 months, payment behavior (autopay status, late payments), prior cancellation attempts, recent service interactions (complaint signals), and competitive market signals where available.
The combined operational workflow
The cross-sell and retention models feed a unified operational workflow in the CRM. Each week, the workflow generates: (a) the renewal review queue prioritized by churn risk and retention value; (b) the producer outreach list with the next-best-offer for each customer; (c) the marketing campaign exclusion list that suppresses outreach to customers who are already in producer-led conversations or recent claim resolutions; (d) the management dashboard tracking the program’s impact on retention rate, cross-sell attach rate, and producer adoption.
The implementation gotcha that derails most programs
The most common failure mode I have watched: the carrier deploys cross-sell propensity models and starts sending offers without proper suppression rules. Customers receive 3 to 5 offers from different campaigns and producers within a 30-day window. Customer complaints rise. Producers complain that their offers are being undercut by marketing campaigns. The program gets suspended pending governance review. Months are lost. The fix is straightforward (centralized suppression rules with a clear hierarchy: producer-initiated offers suppress marketing campaigns, claim activity suppresses cross-sell for 60 days, recent complaint suppresses all outbound for 90 days) but it has to be in place from day one, not added after the program has built customer-experience debt.
The business case math
For a mid-tier carrier in the $500M to $5B GWP range, a working cross-sell-and-retention AI program typically produces 2 to 5 percent retention lift and 1 to 3 percent net new premium from cross-sell over the first 12 to 18 months. The math compounds over time as the models learn from production outcomes and the operational program refines.
AI use case 3 - Conversational AI and embedded copilots
The Wave 2 generative AI use case that produces the most visible productivity lift for service reps and producers. The implementation patterns are still maturing, and the failure modes are real.
Customer-facing chatbot for service inquiries
The chatbot answers customer questions about coverage, billing, claim status, payment plans, and routine policy service. For routine questions (what is my premium, when is my next payment, what is my deductible), the chatbot resolves the inquiry without human escalation. For complex or sensitive questions, it routes to a human service rep with the conversation context preserved.
The evaluation criterion: ask how the chatbot handles questions it cannot answer well. A chatbot that confidently produces wrong answers is a liability. A chatbot that acknowledges uncertainty and routes cleanly to a human is useful.
Service rep copilot for inbound calls
In my experience watching production deployments, the service rep copilot is the highest-productivity copilot use case I have seen ship. The copilot listens to the customer call (with appropriate consent capture for TCPA and state recording rules), surfaces the customer’s profile and recent context in real time, suggests responses to questions, drafts after-call notes for review, and queues the follow-up actions for the rep to confirm. Rep handle time decreases meaningfully, after-call work decreases substantially, and customer satisfaction improves because the rep can focus on the customer rather than on system navigation.
Producer copilot for daily work
The producer copilot answers questions about the carrier’s appetite (“can we write a builder’s risk for a project in California with these characteristics?”), surfaces product information, drafts customer communications, and helps the producer complete submissions faster. For independent producers writing across multiple carriers, the copilot’s job is to make the carrier easier to do business with than the competitor’s portal.
Marketing content copilot
The marketing team uses generative AI to draft campaign content, personalize at scale across customer segments, and adapt content for state-specific regulatory requirements (different states have different disclosure requirements, opt-out language, and product approval status). The compliance review workflow has to remain human-supervised - the AI drafts, a human approves before sending.
The hallucination problem and the mitigation pattern
Generative AI systems will produce confidently wrong answers if their training data is incomplete or the question falls outside their domain. For insurance specifically, the mitigation pattern that works in production: retrieval-augmented generation (RAG) where the AI grounds its responses in the carrier’s actual product documentation, NAIC suitability guidance, and customer data, with fallback to “I cannot answer that confidently, let me connect you with a human” when the confidence is low.
The implementation gotcha: vendors who claim their conversational AI never hallucinates are either selling a heavily constrained interface (which produces low engagement) or are not measuring hallucination rates honestly. Ask for the hallucination rate measurement methodology and the production benchmarks. Reasonable answer: a documented methodology with measurable benchmarks. Unreasonable answer: confident reassurance.
The change management overhead
Conversational AI and embedded copilots fail more often from change management than from technology. Service reps and producers need training, confidence in the tool, and time to develop the working pattern of using it effectively. Carriers that ship the tool without the change management investment see adoption stall at 20 to 30 percent of the workforce. Carriers that invest in the change management see adoption climb past 70 percent within 90 days.
AI use case 4 - Process automation in CRM workflows
The process automation use case combines traditional robotic process automation (RPA) with generative AI for the unstructured-data steps that RPA could not handle alone. This is where the operational productivity gains for back-office CRM functions concentrate.
ACORD form ingestion and data extraction
Producers submit ACORD forms in PDF, image, or scanned format. The AI extracts the structured data, validates it against the CRM’s product taxonomy, flags exceptions for human review, and populates the customer record and quote pipeline. This eliminates the data rekeying that traditionally absorbs a meaningful fraction of new business operations time.
Customer onboarding automation
When a customer binds a new policy, a sequence of CRM activities needs to fire: welcome communication, autopay setup if applicable, document delivery, agent portal access provisioning, marketing automation enrollment, suitability documentation if applicable. The automation handles the sequence with human oversight on the decisions that need human judgment.
Renewal workflow automation
For straight-through renewal eligibility (no significant change in risk, no claim experience, in-appetite), the AI executes the renewal workflow autonomously: generate the renewal package, issue the renewal documents, update the policy in the PAS, communicate to the customer. For non-straight-through renewals, the AI prepares the case for human review with the relevant context already assembled.
Claims-related CRM workflows
When a claim is opened, the CRM needs to fire customer communications, suppress cross-sell campaigns for the customer for 60 to 90 days, update the customer’s risk profile, and prepare the retention conversation for the renewal that may follow. When a claim closes, the CRM fires the satisfaction follow-up, the retention re-engagement workflow, and the eventual cross-sell re-eligibility. The automation coordinates this without the operational team having to assemble it manually.
Marketing campaign execution
Trigger-based marketing campaigns (renewal window, life event, policy anniversary, claim closure satisfaction) execute autonomously with the suppression rules from Section 5 applied centrally. Campaign content adapts to the customer’s segment and channel preference. Outcomes feed back into the propensity models for future targeting.
The automation evaluation framework
For each candidate automation, ask three questions: what is the human time currently spent on this workflow at production volume; what is the error rate or rework rate from manual execution; what happens when the automation produces an unexpected outcome. The workflows where the automation produces clear time savings, error reduction, and recoverable failure modes are the right candidates. Workflows that need significant human judgment in every instance are not ready for automation.
AI use case 5 - Predictive analytics for pricing and underwriting context
The CRM is not the system of record for pricing or underwriting decisions - those live in the rating engine and the underwriting workbench respectively. But the CRM is where the customer-facing context for those decisions surfaces, and the predictive analytics that inform them affect what the CRM displays.
Risk profile evolution scoring
As a customer’s risk profile evolves over time (claim experience, policy changes, behavioral signals), the CRM surfaces the evolution to service reps and producers. A customer whose risk is improving may be eligible for premium reductions at renewal; a customer whose risk is deteriorating may need coverage adjustments or pricing changes. The predictive scoring drives the conversation guidance the CRM surfaces.
Quoting context and competitive intelligence
When a producer or service rep is quoting a new piece of business for an existing customer (an additional policy), the CRM exposes the predicted quote outcome, the competitive context (what comparable customers paid at competitive carriers based on the carrier’s market data), and the bind probability. This is where the CRM connects to the quoting layer - we covered the quoting architecture specifically in our insurance quoting software guide.
Cross-sell quoting prefill
For cross-sell offers, the CRM uses predictive analytics to prefill the quote inputs from existing customer data, third-party data (VIN lookup for new vehicles, property data for new homes), and behavioral signals. The producer’s job becomes verifying the prefill and binding rather than collecting data from scratch. Time-to-quote drops dramatically, conversion rates improve, and producer productivity rises.
Appetite guidance for new submissions
For commercial lines especially, the CRM exposes the carrier’s appetite intelligence to producers in real time - whether a specific risk will likely be written, what indicative price range applies, what additional information is needed. This reduces wasted effort on both sides of the producer relationship. We covered the appetite guidance feature specifically in Section 6.4 of our insurance CRM features evaluation guide.
The integration boundary that matters
The CRM exposes pricing and underwriting predictive context to users; it does not replace the rating engine or the underwriting workbench. The integration boundary is clean: the CRM surfaces the context, the rating engine produces the price, the underwriting workbench owns the bind decision for non-STP risks. Carriers that try to embed pricing and underwriting decision authority in the CRM end up with a system that competes architecturally with the systems of record and produces operational conflicts.
AI governance - NAIC, fairness, explainability for CRM models
The AI governance overlay is where carriers separate from each other in 2026. The carriers building governance into AI deployment from day one are taking strategic advantage; the carriers treating governance as a phase-two retrofit are accumulating regulatory exposure.
The NAIC Model Bulletin on AI Systems Use
The NAIC adopted the Model Bulletin on Use of Artificial Intelligence Systems by Insurers in December 2023, and state adoption has been progressing through 2024 to 2026. The bulletin establishes expectations for AI governance, model risk management, third-party AI vendor oversight, and documentation. States that have adopted the bulletin have varying enforcement timelines, but the direction is clear: AI systems that affect consumer outcomes need governance equivalent to traditional model risk management.
The governance framework that works in production
The governance framework I recommend for insurance CRM AI has five components, drawing on NIST AI Risk Management Framework principles applied to the insurance-specific context: (a) model inventory documenting every AI system in use, its purpose, its training data, and its consumer-impact assessment; (b) bias and fairness review for every model that affects consumer outcomes, with documented testing methodology and remediation plans; (c) model performance monitoring with documented thresholds for retraining or model retirement; (d) third-party AI vendor due diligence covering the vendor’s data practices, training methodology, and ongoing model maintenance; (e) explainability documentation for any model that affects consumer outcomes, sufficient to support adverse action notices and regulator inquiries.
The data governance prerequisites
AI governance depends on data governance. The carrier needs documented data lineage (where does customer data come from, who has accessed it, what has it been used for), consent capture and tracking aligned to GLBA and state insurance regulations, PHI protection for any health insurance lines under HIPAA, and PII protection across the customer data lifecycle.
The third-party AI vendor question
Most carriers will run AI systems built by third-party vendors. The governance overlay for vendor AI is harder than for in-house AI because the carrier has less visibility into the model internals. The minimum vendor due diligence: documented model methodology, training data composition, bias testing approach, model performance benchmarks, model update cadence, and a clear support arrangement for regulatory inquiries.
The fairness and explainability practical reality
NAIC and several state insurance departments have signaled increasing scrutiny of AI models that affect consumer outcomes. Fairness testing for insurance AI is technically complex (insurance has legitimate risk-based pricing factors that correlate with protected characteristics, requiring careful analysis to distinguish discriminatory practices from legitimate risk segmentation). Carriers that have invested in well-designed fairness testing methodology can defend their models in regulatory review; carriers that have not are accepting regulatory risk.
The build vs buy implication
The governance overlay is a meaningful cost - it requires data science, legal, compliance, and audit capabilities that many mid-tier carriers do not have at scale. This is one of the reasons configuring a horizontal CRM with insurance-native AI overlays is harder than it looks - the governance burden falls on the carrier in either case, but the buy-from-insurance-native-vendor option includes vendor-provided governance documentation and model performance benchmarks. We covered the build-vs-buy decision specifically in Section 8 of our insurance CRM definitional guide.
What AI in insurance CRM cannot do yet
I am going to spend a section on the honest limits because most vendor content on AI in insurance CRM will not. If you are evaluating an AI investment, knowing what does not work in 2026 is at least as useful as knowing what does.
It does not replace producer judgment for complex commercial submissions
For straight-through personal lines and small commercial, AI can autonomously execute the workflow with high accuracy. For complex commercial submissions with multi-faceted risk profiles, AI augments the underwriter and producer rather than replacing them. Carriers that pitch their AI as replacing producer judgment for commercial business are overpromising; carriers that pitch their AI as making producers more productive on commercial business are positioning honestly.
It does not solve the data quality problem by itself
AI models trained on inconsistent customer data produce inconsistent scores. The carriers that get the most value from CRM AI invest in identity resolution, data cleansing, and master data management as foundational work. The AI gets better as the data gets better; the AI does not fix the data automatically.
It does not eliminate regulatory complexity
State-by-state insurance regulations, NAIC model governance requirements, HIPAA for health lines, GLBA for life lines, TCPA for outbound communications - these remain. AI accelerates the workflow within the regulatory framework; it does not simplify the framework. Carriers that ship AI without the compliance overlay create regulatory exposure faster, not slower.
It does not run reliably without governance
In my experience watching production AI models across multiple carrier deployments, models drift over time as the underlying data distribution changes. Models that worked well at deployment degrade silently. Without active monitoring, retraining cadence, and performance benchmarks, the AI’s effectiveness erodes and the carrier does not know until a customer complaint or a regulatory inquiry surfaces the problem. The governance overhead is not optional.
It is not a 90-day implementation at any meaningful scale
Vendors who promise CRM-wide AI deployment in 90 days are selling either a configured demo or scope so narrow that it does not include real data integration, governance overlay, or change management. Real AI deployment with the governance overhead, the data foundation work, and the operational change management takes 6 to 18 months depending on which use cases ship first and at what production scale.
Reference deployments and AI in production
The deployments closest to the architecture described in this article are public Decerto case studies. I lead the Agent Portal product and the AI capability runs as part of the CRM stack, so these are the references I personally trust most.
Lead management for Warta - AI-powered cross-sell triggering
The Decerto lead management platform built for Warta (Talanx Group) handles real-time customer engagement, lead routing, and cross-sell triggering across the 40,000-producer eAgent platform. The customer intelligence and cross-sell propensity capabilities described in Sections 4 and 5 of this article are the working architecture of this deployment.
Full case study: Enhancing Lead Management for Warta (HDI/Talanx Group).
eAgent for Warta - producer copilot at scale
The eAgent platform serves 40,000 producers, and the producer-facing AI features described in Section 6.3 - appetite guidance, copilot assistance, submission acceleration - operate at this scale daily. The deployment scale stress-tests the conversational AI mitigation patterns from Section 6.5 in ways that smaller deployments cannot.
Full case study: The eAgent system for Warta (HDI/Talanx Group).
IRON Sales Platform for InterRisk - productivity automation
IRON, the modern sales platform InterRisk (Vienna Insurance Group) built with Decerto, focuses on producer productivity and customer engagement automation. The process automation use cases described in Section 7 - ACORD ingestion, onboarding automation, renewal workflow automation - are the operational baseline of this deployment.
Full case study: Modern Sales Platform IRON for InterRisk.
Higson at Allianz Poland - the product intelligence layer
Allianz Poland uses Higson, the Decerto product configurator, as the product intelligence layer that the CRM AI taps into for cross-sell propensity, appetite guidance, and quoting prefill. The integration pattern - product configurator as the source of truth for product rules, CRM as the customer-facing surface - is the architecture that lets the AI work coherently across products and lines.
Full case study: Insurance product configuration for Allianz.
Warta SME - cross-sell triggering in production
The Warta SME deployment focuses on identifying personal lines customers who launch small businesses and become commercial lines prospects. The trigger detection and propensity scoring described in Section 5 are the working capability of this deployment - real-world cross-sell AI at carrier scale.
Full case study: Simplifying the insurance sales process for SMEs.
Sources and citations
- McKinsey & Company - “The future of AI for the insurance industry” (July 2025).
- McKinsey & Company - “Insurance productivity 2030.”
- McKinsey & Company - “How data and analytics are redefining excellence in P&C underwriting.”
- Bain & Company - Customer Loyalty in P&C Insurance.
- Bain & Company - “Reinvigorate Cross-Selling.”
- NAIC - Model Bulletin on Use of Artificial Intelligence Systems by Insurers (December 2023).
- NAIC - Producer Licensing Model Act.
- NIST - AI Risk Management Framework (AI RMF 1.0).
- NIST Cybersecurity Framework.
- Aite-Novarica Group / Datos Insights - producer engagement and AI adoption research.
- HHS - HIPAA Privacy and Security Rules.
- FTC - Gramm-Leach-Bliley Act overview.
- ACORD Standards.
- Big I (IIABA) / Agents Council for Technology (ACT).
Subscribe to receive the latest blog posts to your inbox every week.
