Which Insurance Jobs Disappear First? VCs Just Named Them.
Welcome to DX Brief - Retail, where every week, we interview practitioners and distill industry podcasts and conferences into what you need to know.
In today's issue:
Why 95% of GenAI Pilots Die in the Sandbox – And How to Join the Other 5%
Which Insurance Jobs Will Disappear First And How to Prepare
How Insurers Are Rewiring Operations Beyond Cost Reduction
1. Why 95% of GenAI Pilots Die in the Sandbox – And How to Join the Other 5%
Talk by Frederik Bisbjerg: Why generative AI fails in insurance – and how smart leaders turn pilots into real results (Oct. 25, 2025)
Frederik Bisbjerg works across the UAE and GCC insurance markets as a global transformation advisor, and he's watching the same pattern repeat: every insurer talks about AI, most run pilots, but only 5% actually deploy GenAI into production.
The revolution is happening in media – AI creating videos, images, social content – but not inside insurance companies where 30% productivity gains remain theoretical rather than realized.
Why? Because insurers don't understand the fundamental difference between deterministic AI (which has clear rules and predictable outputs) and probabilistic GenAI (which produces different results each time).
This matters: compliance requires determinism, GenAI offers probabilism. Insurers need a framework for identifying use cases where probabilistic outputs are acceptable, tools that work without IT integration, and leadership that treats "genius on demand" as infrastructure rather than experiment.
The framework many insurers miss:
Deterministic vs. probabilistic AI – one works for compliance, one doesn't. Traditional AI is deterministic: same input produces same output every time. You train a fraud detection model, it consistently flags the same risk patterns. Regulators and reinsurers understand this.
Generative AI is probabilistic: same prompt produces different outputs each time. Ask ChatGPT to summarize a policy document twice, you get two different summaries. Both might be good, but they're not identical. This isn't a bug, it's how GenAI fundamentally works.
Why does this matter? Because insurance runs on compliance, regulatory approval, and reinsurer acceptance, all of which require predictability. You can't file policy language with regulators if your AI generates different wordings each time. You can't price risk if your AI produces different premium calculations on identical risks. You can't defend claims decisions if your AI's reasoning varies between reviews.
The strategic implication: most use cases insurers initially target for GenAI – pricing models, policy generation, claims decisions – are actually terrible fits because they require determinism that GenAI cannot provide. The 95% of pilots that fail do so because insurers try applying probabilistic tools to deterministic problems, then discover regulators and reinsurers won't accept the approach.
The 5% that succeed use GenAI where probabilistic outputs are acceptable: summarizing documents (multiple valid summaries exist), identifying inconsistencies (flagging potential issues for human review), augmenting research (providing starting points humans validate).
"Genius on demand" beats enterprise AI transformation: start with adjacent tasks, not core systems. Harvard Business School research shows massive productivity gains when knowledge workers use GenAI for "adjacent tasks", i.e., the work surrounding their core responsibilities rather than the core work itself.
An underwriter's core task is making risk decisions. Adjacent tasks include: researching industry trends, summarizing lengthy reports for executive briefings, comparing policy wordings, drafting email communications.
This matters because adjacent tasks don't require IT integration, regulatory approval, or reinsurer acceptance. You can start using ChatGPT or Claude today for these tasks without any organizational change.
The mistake insurers make is that they try to deploy GenAI in core systems (underwriting engines, policy admin, claims platforms) where integration requirements, security concerns, and IT backlogs create 12-24 month timelines. By the time you deploy, the AI models have evolved and your implementation is already outdated.
The alternative: treat GenAI as "genius on demand" available to every employee for adjacent tasks. Train people to use it effectively, create guardrails around data security, but don't require IT integration.
What to do about this:
→ Audit your top 10 planned AI use cases against the deterministic vs. probabilistic test. For each use case, ask: does this require identical outputs every time (deterministic) or are multiple valid outputs acceptable with human validation (probabilistic)? If your use case requires determinism – pricing engines, policy generation, claims decisions – you probably shouldn't use GenAI or you need human-in-the-loop workflows. Redirect those resources to probabilistic use cases like document summarization, inconsistency detection, or research augmentation.
→ Launch a 90-day "genius on demand" literacy program targeting 100% employee adoption. Not an AI strategy, not a transformation initiative but a practical training program teaching every employee to use GenAI for adjacent tasks in their daily work. Create role-specific prompt libraries: underwriters get research prompts, claims managers get document review prompts, executives get report summarization prompts. Track adoption monthly (what percentage are actively using it?) and collect ROI stories. Bisbjerg suggests aiming to get from 30% to 100% active users across your organization.
2. Which Insurance Jobs Will Disappear First And How to Prepare
India Insurtech Association annual event, Panel: The Future of Insurance is AI - Funding Vertical AI Companies with Albert Shyy (Eurazeo), Lars Gehrmann (Qatar Insurance Group), Adarsh Chokhani (Assurekit), and Ashish Fafadia (Blume Ventures) (Nov. 14, 2025)
Corgi in the US (building the first full-stack AI insurance company) just raised $70 million with a simple thesis: take large P&L line items that are human-led in insurance companies and reduce them by 80% through AI. Meanwhile, VCs from Eurazeo, QIC Digital Ventures, and Blume are seeing insurance companies adopt GenAI in production faster than any other financial services sector.
Investors predict that claims processing, underwriting, and customer service roles will see the most dramatic transformation. Here's what leading VCs say about which AI-first insurance companies will win, and which traditional insurance functions are heading toward obsolescence.
TLDR:
Value chain ownership is critical for AI success in insurance: horizontal AI tools fail because insurance products are too complex. Each product (motor vs. property vs. health) requires different AI approaches and delegated authority structures to work effectively.
Three job categories face elimination by 2030: VCs unanimously identify claims processing, customer service, and traditional underwriting as most vulnerable, with the caveat that human insight still matters for proposition development and complex claims scenarios.
The framework many insurance executives miss:
AI adoption requires vertical integration, not horizontal tools. Chokhani from Assurekit laid out a critical insight: "When we look at the BFSI spectrum, a lot of products in the banking space are transactional in nature. But insurance is fairly more complex. Every product has to consume and adopt AI differently."
This matters because high-frequency products like parametric insurance require on-spot decisioning powered by AI. But property products – where you might get one claim per 100,000 policies – require a completely different approach combining manual review with AI support. The investors on this panel were clear: if you don't have value chain ownership, AI adoption is extremely hard. There are too many stakeholders to coordinate for the system to work cohesively.
Think about what this means for your AI strategy. Most insurers are buying point solutions that promise to "add AI" to existing processes. But Assurekit gets delegated authority from partners precisely because their AI-led models own enough of the value chain to make autonomous decisions. When your AI system is just one tool among many in a fragmented process, it can't actually deliver on its promise.
Sales becomes obsolete; claims and underwriting transform dramatically. When asked which insurance jobs will disappear due to AI – like the whaling industry in the 1850s when chemical processes replaced whale oil – the panel gave remarkably consistent answers with nuanced differences.
Albert Shyy from Eurazeo: "Claims processing certainly is an area undergoing a lot of change right now. In the longer term, underwriting is going to see a lot of change." Lars from QIC disagreed slightly: "I don't believe it will go totally to zero in claims because you can learn a lot about the proposition from claims. For me, the job category most impacted is customer service – and this is not necessarily insurance, it's everywhere."
Adarsh offered a contrarian view focused on growth rather than efficiency: "I want there to be a world where sales is obsolete because I want the product to be so intuitive that users can come and buy it instead of a push-led model, B2B or B2C." He argued the human element still helps in crisis management throughout the value chain, but making products self-service eliminates the need for traditional distribution.
"Anything which is customer-facing is going to require much less people for the same scale, or maybe this scale is good enough for 20x the growth."
What to do about this:
→ Audit your AI investments for value chain ownership. Map which AI tools you're deploying and identify whether they control enough of the end-to-end process to make autonomous decisions. If your AI tools are just advisory systems that still require three human handoffs, they won't deliver the efficiency gains you're expecting. Prioritize AI implementations that can own outcomes, like auto-adjudication of simple claims or straight-through processing of standard risks.
→ Plan your workforce transition around the three-year horizon. With VCs predicting claims processing, customer service, and underwriting will see dramatic headcount reductions, start planning now for workforce redeployment. Identify which staff can be retrained for higher-value work (like proposition development, complex case handling, or customer relationship management) and which functions should stop hiring immediately.
3. How Insurers Are Rewiring Operations Beyond Cost Reduction
Insurtech Story podcast, Episode: Rewiring the Insurance Operating Model: From Offshoring to Orchestrating Value with Jitin Sharma (Infosys) (Nov. 11, 2025)
For decades, insurers treated offshoring as pure labor arbitrage: ship work overseas, cut costs, achieve scale. But Jitin Sharma, who heads insurance consulting for Asia-Pacific at Infosys and has advised AIG, Allianz, and AXA across 18 years, says we've moved far beyond the “factory” model.
Today's leading insurers are building value orchestration networks that blend global talent, technology platforms, and ecosystem partners into seamless capability engines. The goal isn't just efficiency anymore – it's agility, innovation, and customer impact.
TLDR:
Transform Global Capability Centers from cost centers into innovation hubs by moving beyond transactional work to strategic capabilities like data analytics, AI development, and customer experience design – leading insurers now use Global Capability Centers (GCCs) for competitive advantage, not just savings.
Deploy GenAI strategically through the full operating model (claims processing, underwriting, customer service, fraud detection) rather than point solutions, creating connected intelligence that learns across the entire value chain.
The framework many insurers miss:
Move from labor arbitrage to capability orchestration: your GCC is a competitive weapon, not a cost center. "I started my career where offshoring was all about labor arbitrage." That era is over. Today's smart insurers aren't running factory models where offshore teams execute predefined tasks. They're building "smarter, more connected global teams that can actually drive innovation."
Think of it like this:
The old model was a one-way street: HQ defines processes, offshore executes.
The new model is a value network: global teams contribute strategic capabilities, bring market insights, and drive innovation back to the enterprise.
What does this look like in practice? Leading insurers are moving GCCs up the value chain. Instead of just processing claims and handling customer service tickets, these centers now handle data analytics, AI model development, customer experience design, and even new product development.
When your Singapore or Manila or Bangalore center is building your next-gen underwriting engine or designing your claims automation strategy, you're orchestrating value, not managing costs.
The shift requires a mindset change. You're "no longer running a factory model." You're building a distributed capability engine where talent, regardless of location, contributes to strategic priorities.
Integrate GenAI and intelligent automation across the operating model, not in silos. Most insurers approach GenAI tactically: a chatbot here, a document scanner there. Sharma sees the bigger picture: "How are emerging technologies like GenAI, data platforms, and intelligent automation reshaping the insurance operating model from process efficiency to experience-led transformation?"
The strategic principle: GenAI isn't a point solution. It's infrastructure that should connect across your entire operating model. Deploy it systematically through claims processing (automated first notice of loss, damage assessment), underwriting (risk evaluation, pricing optimization), customer service (intelligent routing, sentiment analysis), and fraud detection (pattern recognition, anomaly detection).
When you deploy GenAI as connected infrastructure rather than isolated applications, you create network effects. Your claims AI learns from underwriting patterns. Your customer service AI feeds insights to product development. Your fraud detection AI improves risk assessment. The whole system gets smarter together.
This matters because incremental AI deployment doesn't create competitive advantage. Systematic deployment across the value chain does. Insurers need to think about "data-led decisioning" as the foundation, not just efficiency tools.
What to do about this:
→ Audit your GCC mission and capabilities. Schedule a strategic review of what work your Global Capability Centers actually do. Map current activities (processing, support, execution) against strategic capabilities (analytics, innovation, experience design). If 80%+ is transactional, you're still in the old model. Build a 12-month roadmap to shift 30-40% of GCC work to strategic value-creation activities.
→ Deploy GenAI with a connected architecture, not point solutions. Don't let business units buy individual AI tools in isolation. Create a cross-functional task force (underwriting, claims, customer service, IT) to map where GenAI can create value across the full customer and policy lifecycle. Build a unified data platform first, then layer AI capabilities that share learning and insights across functions.
→ Build capability partnerships to close the talent gap. Don't try to hire your way out of the skills shortage – the market can't supply enough talent. Instead, partner with systems integrators, technology providers, and capability centers that can provide domain-specific expertise (insurance + AI, insurance + cloud architecture). Structure these as true partnerships with knowledge transfer requirements, not just staff augmentation.
Disclaimer
This newsletter is for informational purposes only and summarizes public sources and podcast discussions at a high level. It is not legal, financial, tax, security, or implementation advice, and it does not endorse any product, vendor, or approach. Insurance environments, laws, and technologies change quickly; details may be incomplete or out of date. Always validate requirements, security, data protection, regulatory compliance, and risk implications for your organization, and consult qualified advisors before making decisions or changes. All trademarks and brands are the property of their respective owners.
