Insurance
Webinar
How Bajaj General Insurance Settled Motor Claims in 20 Minutes And Why They Never Built an Innovation Lab
Date
June 11, 2026
Author
Ayush Mourya
INTRO
A motor accident. A stressed customer. A claim that traditionally took days of surveyor visits, back-and-forth paperwork, and manual assessments. Bajaj General Insurance collapsed that journey to 20 minutes not by hiring a dedicated AI innovation team in a glass office, but by injecting AI directly into operations, the P&L, and the Monday morning review. That gap between an innovation experiment and an operational reality is exactly where most Indian insurers are stuck today.
For nine years, Deepu KV Head of Operations and Customer Experience at Bajaj General Insurance ran 1,400 people, 200+ locations, and 150 million+ customers. Over that period, Bajaj General Insurance became the most awarded digital insurer in India, holding the highest NPS in the industry and the lowest complaint ratio for more than a decade. In a recent InsurTech Insiders session, he broke down exactly how they did it and the lessons are relevant for every insurer thinking about AI claims processing automation in India today.
The Problem: Insurance is a Low-Touch Product with High-Stakes Moments
The single insight that frames everything about insurance operations is this: your customer doesn't use you every day.
Unlike a banking app used 365 days a year for transactions an insurance customer may interact with their insurer on only five out of 365 days. And when they do, it's almost always because something has gone wrong. A car crash. A hospitalisation. A disputed claim. These are high-anxiety, high-stakes moments where every minute of friction compounds the damage to trust.
Traditional motor claims in India made this worse. A customer with a damaged vehicle had to wait for a physical surveyor. The surveyor assessed the damage manually, filed a report, and the payment cycle began. The entire process could take days. For a customer who just had a crash, that's not a claims process it's a test of patience they didn't sign up for.
The same pattern held in contact centers. Bajaj General Insurance found that customers calling in were routed to IVRs that could handle a fraction of their needs, forcing escalations to human agents for the most basic queries. The system was designed around what was easy for the insurer, not what resolved the customer's problem.
What AI Does Differently: Domain + Tool, Not Tool Alone
Here is the most important operational principle from the Bajaj General Insurance story: AI without domain expertise is useless.
When they built their AI-powered motor claims assessment what they call "on-the-spot" settlement they didn't simply license a computer vision tool and point it at photographs. They spent months bringing their own motor claims experts into the training process. These experts codified hundreds of claims categories: Is this damage repairable or a write-off? Is the dent surface-level or structural? Is this a genuine claim or a fraudulent one?
That domain knowledge built over decades of physical surveyor assessments was transferred into the AI model. The result: a customer with a damaged vehicle needs to do exactly two things. Click photographs. Upload them to the app. The AI engine processes the images, estimates the damage value, presents it to the customer for consent, and initiates payment. The whole process takes 20 minutes.
This is what distinguishes effective AI claims processing from an innovation experiment that never reaches production: the tool inherits the domain, not the other way around.
The same principle drove their AI agent deployment for customer service. Rather than building a generic bot, they analysed the top seven reasons customers called their contact centre, then expanded to 23, then 42. The AI agent was trained specifically on those use cases. At launch, it resolved 70% of queries autonomously. Today, that number is 99.5% with only 0.5% of interactions requiring human intervention. That's not a demo statistic. That's an operational reality running at scale across 150 million customers.
This is exactly the architecture that SubVerseAI's ConVerse and AgentVerse products are built around AI agents that handle real, high-volume, high-stakes insurance interactions end-to-end, with stateful orchestration and human escalation built in for the edge cases that genuinely require it.
Real Proof The Bajaj General Insurance Operational Blueprint
Before: A motor claims customer with a damaged vehicle called the insurer. A surveyor was dispatched. Physical assessment took hours to days. Payment followed after manual report processing.
After: Customer clicks two sets of photos in the mobile app. AI engine assesses damage against a model trained on thousands of historical claims. Customer receives estimated settlement value, gives consent, payment is initiated. Total time: 20 minutes.
📊 RESULT: 20-minute motor claim settlement · 99.5% AI-autonomous contact centre resolution · Highest NPS in the Indian insurance industry · Lowest complaint ratio in the industry maintained for 10+ consecutive quarters
What makes this particularly instructive is not just the outcome it's the architecture. Bajaj Allianz never built an innovation lab. They didn't create a separate digital team insulated from operations. The AI was deployed where the customer problem actually lived: inside the claims workflow, inside the contact centre, inside the agent training process.
Deepu's framing: "You have to start with the problem in mind and solve for that using a tool." Innovation labs work backwards they find tools and look for problems. Bajaj Genral Insurance worked forwards.
Before & After What This Looks Like in Practice
BEFORE: An insurance contact centre receives a query from a customer asking about policy inclusions and exclusions. The agent searches an FAQ repository manually while the customer waits. Average handle time: 8–12 minutes. If the query is outside the agent's knowledge, it escalates further.
AFTER: An AI agent listens to the customer's query in real-time and surfaces the exact FAQ from the repository to guide the agent or handles the query end-to-end autonomously. Query resolved before the call ends. 99.5% autonomous resolution rate. First-contact resolution increases measurably without adding headcount.
The same principle extends to fraud detection. Bajaj Genral Insurance claims fraud AI was trained by their own fraud investigation team the same human experts who had spent years spotting fraudulent patterns. The model now flags suspicious claims at intake, before any payment is initiated, protecting genuine customers' settlement speed while filtering out fraud. One architectural detail that's publicly known: the system requests a 360-degree image set, not just the damaged component precisely because that broader context is what helps distinguish a genuine claim from a staged one.
Comparison Table: Traditional Insurance Operations vs. AI-Powered Operations
Metric | Traditional IVR / Manual Process | SubVerseAI ConVerse + AgentVerse |
|---|---|---|
Motor claim settlement time | 2–5 days (surveyor-dependent) | 20 minutes (AI photogrammetry + instant processing) |
Contact centre query resolution | 70% deflection, 30% to humans | 99.5% autonomous resolution |
Peak load handling (festival seasons) | Missed leads, queue overflow, agent burnout | Scales automatically — 0% missed interactions |
Fraud detection speed | Post-claim manual review | Real-time AI flag at intake |
Language coverage | English + Hindi only (most insurers) | 30+ languages natively supported |
First contact resolution rate | ~40–50% industry average | 60%+ with DataVerse-backed agent intelligence |
Monthly conversation volume | Limited by headcount | 1M+ monthly conversations handled autonomously |
The Multilingual Imperative
One lesson from the Bajaj General Insurance session that tends to get underestimated: multilingual AI is not a feature. It's the difference between a tool that serves your English-speaking urban customers and one that serves India.
Bajaj General Insurance built their AI agents voice bots, chat interfaces, their agricultural app Farmra in multiple Indian languages from the start. The result: they didn't just improve NPS in metros. They maintained high NPS scores in smaller cities and towns as well. Their Sarvatra Bima initiative a digitally equipped van that took insurance to villages across India ran entirely in local languages, with AI-powered transactions including policy issuance and claim settlement happening in areas without a single insurance office.
For InsurTech operators thinking about India at scale, this is the benchmark. The question is not "does our AI work?" it's "does our AI work in Kannada, in Tamil, in Marathi, in Bengali, at the same quality level it works in English?"
SubVerseAI's ConVerse layer handles 30+ languages natively, with no translation middleware adding latency or degrading accuracy. Code-switching mid-conversation a customer switching between Hindi and English in the same call is handled without dropping context.
The Architecture Principle: Modular Core, Dynamic Interface
One strategic insight from the session that applies directly to every insurer evaluating AI platforms today: keep your core stable, make your interfaces modular.
Bajaj General Insurance approach was to treat their data core the system of record as the stable foundation, and the customer-facing interface as a dynamic layer that evolves with technology. When call centres were the norm, the call centre was the interface. When chat AI matured, the chat layer became the interface. When voice AI matured, voice became the primary layer. Each shift happened without rearchitecting the core.
This is precisely the design philosophy behind SubVerseAI's three-layer stack. DataVerse is the intelligence core the unified customer data layer, stable and persistent. ConVerse is the interaction layer handling voice, WhatsApp, email, and video, modular by channel. AgentVerse is the execution layer the workflow orchestration that acts on what ConVerse hears and what DataVerse knows. When the next interaction paradigm emerges, the core doesn't change. Only the interface adapts.
FAQ
Q1: What is AI claims processing automation and how does it work for insurance companies in India? A: AI claims processing automation replaces manual, surveyor-dependent claim assessment with AI agents that can process photographic evidence, assess damage or verify claim validity in real-time, and initiate payouts without human intervention. For Indian insurers, this typically means deploying a combination of computer vision (for motor claims), natural language AI agents (for query handling), and workflow automation (for backend processing). The key requirement is training the model on domain-specific claims data not generic AI models.
Q2: Why do most Indian insurance contact centres still struggle to resolve customer queries autonomously? A: The core issue is that most AI deployments in insurance contact centres are built around generic use cases FAQ bots that handle simple queries rather than the actual top reasons customers call in. When Bajaj General Insurance designed their AI agent stack, they started by auditing every inbound call reason and training specifically against those scenarios. Insurers that skip this domain-grounding step end up with AI that handles 30–40% of queries and deflects the rest to human agents, defeating the purpose of automation entirely.
Q3: Is AI-powered claims assessment more accurate than manual surveyor assessment? A: For standard motor claims dents, surface damage, component replacement decisions AI photogrammetry has reached a point where accuracy is comparable to or exceeds manual surveyor assessment, primarily because it eliminates human variability and applies consistent classification logic at scale. Bajaj General Insurance reports that customer disputes against AI assessments have become extremely rare as model accuracy has improved. The key differentiator is training methodology: models trained on thousands of domain-specific claims cases significantly outperform generic computer vision tools.
Q4: How long does it take to deploy AI agents for insurance operations in India? A: With pre-built integrations and domain-specific configuration, AI agent deployments for insurance contact centres and claims workflows can go live within 30 days. SubVerseAI's standard deployment model starts with a 30-day pilot connecting to existing telephony, CRM, and policy data systems and produces measurable first-contact resolution improvements within the first billing cycle. Full enterprise-scale deployments with multiple agent types typically complete within 60–90 days.
Q5: What ROI can Indian insurers expect from AI contact centre automation? A: The ROI equation for insurance AI automation has two components: direct cost reduction and revenue recovery. On cost: eliminating manual handling for 99%+ of routine interactions reduces cost per interaction by 35% or more (as seen in SBI Payments' DataVerse deployment). On revenue: AI agents that handle 100% of inbound volume during peak season Diwali renewals, monsoon health surges eliminate missed leads entirely. Acko Insurance scaled from 6,000 to 15,000 daily calls handled autonomously, with 2–3× improvement in lead-to-sale conversion. Combined, these outcomes typically produce payback periods of well under 12 months for enterprise-scale deployments.
The Next Step
The Bajaj Genral Insurance story is the clearest available proof that AI claims processing automation in Indian insurance is not a future aspiration it's a current operational reality. The companies that get there fastest share a common approach: they start with the specific operational problem, they bring domain expertise into the training process, and they deploy into production rather than piloting in isolation.
If your claims workflow still depends on physical surveyors for standard motor damage. If your contact centre is still routing 30%+ of queries to human agents. If your AI deployment is still running as a protected experiment rather than a live P&L contributor those are solvable problems, with demonstrable timelines.
Book a 30-day pilot at subverseai.com first results in weeks, not months.
More Blogs
Stay ahead with the newest advancements in AI automation. Discover productimprovements, feature releases,
Insurance
What are the Biggest Security Risks for AI Agents, And How Can Enterprises Prevent It?
Feb 12, 2026
Insurance
Webinar
What are the Biggest Security Risks for AI Agents, And How Can Enterprises Prevent It?
Feb 12, 2026
AI in action
What are the Biggest Security Risks for AI Agents, And How Can Enterprises Prevent It?
Feb 12, 2026