Venu Mathukumalli · 813.600.8927 · Venu.Mathukumalli@gmail.com · linkedin.com/in/venumathukumalli
Case study 03 Deep dive
B2C, direct to customer Individuals and businesses buy insurance online
A price from one sentence, not five pages. An AI-first way to buy auto, home, life and business insurance, where the application fills itself and the customer stays in control.
It started from an industry signal. In February 2026 insurers began offering quotes inside ChatGPT, and insurance stocks fell on the news. I asked what buying insurance should feel like when the first touchpoint is a conversation, not a form.
At a glance
01 Problem
Buying insurance online still works like a paper form moved onto a screen. A typical personal auto quote asks for about 24 fields across five pages, including a driver license number and part of a Social Security number, before the customer sees a single price.
Commercial insurance is heavier. A real healthcare association renewal application I worked from runs to seven pages: general information, liability and employees, services and licensed beds, medical services and abuse protocols, directors and officers, crime, and a statement of values for up to ten locations. Most of it is “any changes?” questions the organization already answered last year.
The customer does the insurer’s data entry, and only then learns the price.
Meanwhile, the front door is moving. In February 2026 a digital insurer and an insurance comparison platform launched quoting apps inside ChatGPT, and insurer and broker stocks fell in Europe, the US and Australia on fears of disruption. Industry analysts described it as a reset of insurance distribution: people getting personalized quotes inside AI assistants they already use.
02 Impact
This is a concept, so the business impact is modeled from the prototype and the real application it is based on, not measured. I have marked which is which.
What I delivered
Modeled, not measured
What I took from it
This is a concept with a clickable prototype, not yet tested with customers. In the prototype, simple rules stand in for a language model and for carrier rating systems. Carriers, prices, records and organizations are sample data, and the healthcare renewal is modeled on a real application with all identifying details removed.
03 Product
Carriers still need a complete application, so the form doesn’t disappear. It moves behind the conversation. The customer says what they need in one sentence, and the AI does the data entry, visibly, field by field.
The AI reads one sentence, works out the line of insurance, and fills every field it can: from the customer’s own words, from records, and from sensible defaults it labels as assumptions.
Only questions that change the price, one at a time, with answer chips or free text. Usually two or three. Everything else waits.
Carriers’ rating engines price the filled application. The AI recommends a plan, says why, and shows what moves the price.
Identity, payment and consent come only now, after a price. The AI never buys on the customer’s behalf.
Across my three case studies, the AI’s freedom follows who bears the cost of a mistake. In Claims Adjuster Copilot a licensed adjuster approves every AI proposal. In Enterprise Tool Governance an agent acts inside limits admins set. Here the AI does all of the paperwork, but the customer makes the decision, and for commercial renewals a licensed broker reviews everything before insurers see it.
The journey
From a single sentence to a policy, plus the same idea applied to a commercial renewal. Select a screen to jump to it.
Describe what you need in your own words, or pick a line of insurance.
01 HomeThe AI fills the application and asks only what changes the price.
02 Conversation10 Full formThree plans, a recommendation with a reason, and the price explained.
03 Quotes04 ReviewIdentity and payment only now, then the policy is issued.
05 Decide06 Issued09 AgentFor businesses, last year’s answers carry over and a broker reviews the changes.
07 Renewal08 BrokerThe screens
Select any screen to view it full size. Carriers, prices, records and organizations are all sample data.
The centerpiece. From “Auto insurance for my electric SUV in LA. I commute to work.” the AI picks up location, vehicle and use, fills 15 fields and asks only two questions: the driver’s age and driving record. On the right, the carrier’s five-page application fills itself, color-coded by where each answer came from, so the customer can see exactly what the AI did.
Ask only what changes the priceThree plans from sample carriers, with an AI pick and a plain reason, such as “battery repairs are expensive, so collision cover matters.” Deductible and limits re-price instantly. “What moves your price” shows each factor’s effect in dollars, based only on the customer’s own answers.
Explain the priceEvery answer by page, with its source and a way to change it. Assumptions are the first thing to check, because if an assumption is wrong, the price is wrong. Nothing goes to a carrier until the customer buys.
Show every assumptionThe first time the site asks who you are. Name, license, SSN and payment appear only here, after a price, and the customer confirms the AI’s answers are accurate before “Buy” is enabled.
Sensitive data lastBusiness insurance
The same idea, applied to the seven-page healthcare renewal I started from. Most answers don’t change from one year to the next, so the AI carries them over and asks only what might have changed. A licensed broker reviews anything flagged before insurers see it.
| For the customer | Traditional renewal | Renewal by exception |
|---|---|---|
| Pages to complete | 7 | None. Review is optional |
| Answers to give | 64, most of them unchanged | 6 change checks, plus 2 signatures |
| What the broker receives | A full form to re-read line by line | The same full form, with changes flagged |
| Who decides | The customer and the broker | Still the customer and the broker. The AI prepares |
Counts come from the prototype’s sample renewal, modeled on the real application. They are not measured results.
Last year’s 64 answers are carried over across seven pages. The AI checks six things that might have changed: services, staff and volunteers, funding, incidents, planned changes and the liability limit. Each page turns green as it is confirmed.
Carry over, then confirmThe renewal lands with the broker, not directly with insurers. What changed is flagged, an indicative premium range is shown with its assumptions, and a timeline makes the human steps clear: broker review, insurers, terms, signature.
A person reviews before insurers see it04 Experience
Type your own request, or pick a scenario. Answer with the chips or in your own words, change the coverage, review the answers and buy. The bottom bar jumps to any screen, and “Reset demo” starts over. All data is sample data.
The prototype is a desktop-sized app, so it opens best in its own tab.
Open the prototype ↗How I got here
This page shows the problem, the impact, the product and the experience. The thinking behind them is in a 23-slide presentation: discovery, research materials, personas, journey maps, ideas, user flows and a validation plan.
The clickable prototype and the 2-minute quick peek are on my portfolio site.
I'm happy to walk through the thinking behind this, or my enterprise projects, in more detail.