AI + InsurTech: The Fastest Path to Better Insurance

A stubborn stat hasn’t moved much in five years: around 70% of consumers will get an insurance quote online, but fewer than 30% actually buy online. That gap is where the real work is—and it’s also where AI in insurance is starting to pay off.

In a recent InsurTech-focused podcast conversation, Guidewire’s Chief Evangelist Laura Drabik laid out what’s changed (and what hasn’t) as startups and traditional insurers learn to build together. The most useful takeaway wasn’t “InsurTech is the future.” It was more specific: collaboration only works when the tech reduces complexity, improves trust, and plugs into the core operating model.

This post is part of our AI in Insurance series, and I’ll translate the interview themes into practical guidance for insurance leaders trying to drive growth: where AI fits in underwriting and claims, why personalization is a revenue story (not a UX story), and how to avoid integration traps that turn promising pilots into expensive shelfware.

Collaboration wins because insurers and startups bring different strengths

Traditional carriers don’t lose because they’re “slow.” They lose when innovation stays isolated from real operations. Startups tend to move faster, test bolder ideas, and build products around specific pain points (quote friction, FNOL delays, document chaos, fraud signals). Carriers, on the other hand, bring the hard stuff: regulatory expertise, distribution relationships, risk capital discipline, and decades of loss history.

Drabik’s point is worth stating plainly: InsurTech isn’t replacing insurers; it’s pressuring them to evolve. The most productive partnerships combine:

• InsurTech speed + product focus (solve one problem extremely well)
• Carrier scale + trust (make it safe, compliant, and repeatable)

A revenue gap that’s hard to ignore

The interview referenced an Accenture projection that innovation leaders in P&C could outpace laggards by up to 37% in revenue, with substantial revenue driven by new products and services through 2025. Whether your organization agrees with the exact number or not, the direction is consistent across the market: carriers that can ship new experiences and new risk products faster are separating from those that can’t.

If your 2026 planning cycle is happening right now (and for many carriers it is), this is the moment to decide: are you funding “innovation theater,” or are you funding operating capabilities that actually ship?

AI in underwriting is mostly about better inputs—and fewer dead ends

AI underwriting works when it improves decision quality and reduces cycle time without introducing new compliance risk. That’s the bar.

The podcast emphasized what many underwriting teams already feel: underwriting is becoming more ecosystem-driven. Usage-based insurance (UBI) is the easiest example. It requires a chain of partners—telematics providers, device manufacturers, connectivity, and data platforms. That’s not “nice to have.” It’s table stakes if you want pricing to reflect real behavior.

Where AI delivers tangible underwriting value

Here are four underwriting areas where AI consistently performs—because the work is repetitive, data-heavy, and measurable:

1. Pre-fill and triage: Pulling data from third-party sources and applicant documents to reduce questions and speed quote.
2. Risk segmentation and pricing support: Using predictive models to refine rating factors, especially where traditional variables are blunt.
3. Submission ingestion: For commercial lines, AI can extract key fields from loss runs, schedules, ACORD forms, and attachments.
4. Underwriter workflow: Routing, prioritization, and appetite checks so humans spend time on the risks that actually need judgment.

The “personalization bargain” is real

Drabik noted consumer research showing a meaningful majority (in the 60% range) are willing to share data if it results in personalized pricing. That matches what I’ve seen: customers will trade data for value, but only when the value is obvious and the usage is transparent.

One-liner worth remembering: Personalization isn’t a feature; it’s a pricing and trust agreement.

If you’re building AI models to use new data sources (telematics, IoT, behavioral signals), your underwriting strategy has to include two non-negotiables:

• Explainability: Not a research project—an operational requirement for regulators, agents, and customers.
• Consent and governance: If your data rights are fuzzy, your model value is temporary.

Claims automation is the quickest way to prove ROI—if you pick the right moments

Claims is where AI can create immediate operational leverage, because claims teams deal with heavy volume, high documentation load, and lots of status-chasing.

The interview highlighted how new data sources can:

• Automatically trigger claims (or alerts) based on events
• Speed up service scheduling
• Improve reserve accuracy
• Automate parts of payout processing

That’s a strong blueprint, but the best claims automation programs don’t try to “automate claims.” They automate specific claim moments.

High-impact claim moments to automate first

If you’re prioritizing an AI claims roadmap, these are the areas that typically deliver early wins:

• FNOL intake and summarization: Convert conversations, emails, and forms into structured claim notes.
• Document classification and extraction: Police reports, medical bills, repair estimates, photos.
• Straight-through processing (STP) for simple claims: Rules + models to identify low-risk, low-severity claims.
• Fraud detection triage: Flag patterns and inconsistencies so SIU focuses where it matters.

A practical stance: Don’t start with “AI fraud detection.” Start with “AI fraud triage.” The goal is to route smarter, not to accuse faster.

Reserve accuracy is an AI story (and a finance story)

Reserving is often treated as an actuarial and governance domain—which it is—but it’s also a place where AI can help by improving early estimates using richer signals: repair network pricing, injury indicators, historical settlement patterns, and claim text.

If you need internal buy-in, frame it this way: better reserves improve capital efficiency and reduce end-of-year volatility. That gets attention.

Omnichannel insurance isn’t optional—and AI is what makes it feel coherent

The podcast called out another stat that should make any growth leader uneasy: lots of consumers start online, but many still want a human to finish—especially for complex products.

That’s not a failure of digital. It’s a failure of continuity.

Customers won’t tolerate repeating themselves. They expect the “shopping cart” model: start on mobile, continue on web, finalize with an advisor, and everyone sees the same context.

What “omnichannel” actually means operationally

Omnichannel isn’t “we have a chatbot and a call center.” It’s this:

• Shared customer record across channels
• Conversation history available to agents and advisors
• Consistent recommendations (coverage, limits, deductibles) across touchpoints
• A clean handoff from automation to a human without losing data

AI plays two roles here:

1. Decision support: recommending next-best actions and coverage options based on context.
2. Context packaging: summarizing what happened so far so the next human (or system) can pick up instantly.

The interview referenced a McKinsey metric that organizations with top-tier personalization can drive up to 25% revenue lift. Regardless of the exact percentage in your business, the mechanism is clear: better fit → higher conversion → higher retention → better cross-sell.

Integration is the real bottleneck (and most teams underestimate it)

Carriers don’t struggle to find InsurTechs. They struggle to vet, integrate, and scale them.

Drabik described carrier demand for curation and “pre-built integrations” because time-to-market is everything. Here’s the truth most companies learn the hard way: an AI pilot that can’t plug into your core platforms is a demo, not a capability.

A simple integration scorecard for AI in insurance

Before you sign a contract—or before you extend a pilot—pressure test the solution with questions like:

1. Data path: Where does it pull data from, and where does the output go?
2. Workflow fit: Which user role changes behavior on day one?
3. Audit trail: Can you explain model outputs to compliance and customer advocates?
4. Security and privacy: What’s stored, what’s retained, and what can be deleted on request?
5. Time-to-value: What can go live in 90 days that actually reduces cost or increases conversion?

Snippet-worthy rule: If the value can’t be measured in production, it can’t be managed.

A practical 90-day plan to start (or restart) an AI + InsurTech program

If you’re trying to generate leads, grow conversion, or reduce expense ratios in 2026, you need execution paths that fit real insurance constraints. Here’s a 90-day plan I’ve seen work.

Days 1–30: Pick one journey and define the metric

Pick a lane:

• Quote-to-bind (conversion rate, drop-off rate, time-to-quote)
• FNOL-to-first-contact (cycle time, customer satisfaction, leakage)
• Submission-to-decision in commercial lines (touch time, referral rates)

Define one primary metric and two supporting metrics. Don’t overcomplicate it.

Days 31–60: Implement “assistive AI,” not full automation

Assistive AI is easier to govern and easier to adopt:

• summarization
• extraction
• recommendations
• routing

You’ll get faster adoption because adjusters and underwriters feel supported, not replaced.

Days 61–90: Instrument, learn, and scale one step

Instrument everything:

• model accuracy (and drift)
• time saved per case
• conversion lift
• exception rates

Then scale one step: another product line, another region, another claim type. Scaling beats reinventing.

Where AI in insurance is headed next

The direction is clear: more ecosystem data, more personalized pricing, and more automation in underwriting and claims. But the winners won’t be the companies with the most AI proofs-of-concept. They’ll be the ones with trustworthy, integrated AI workflows that customers and employees actually use.

Insurance will keep having “human moments,” especially for high-stakes products. The opportunity isn’t to eliminate humans. It’s to make the handoffs clean, the products understandable, and the decisions faster and more consistent.

If you’re planning your next AI initiative, here’s the question to pressure-test it: Will this make insurance easier to buy, easier to understand, or easier to recover from when something goes wrong? If the answer is yes, you’re probably building the right thing.