AI is here to stay in insurance: what's working now, according to Altamont's Joe Zuk
Bottom line: AI is already useful in insurance - especially as a research tool and in predictive modeling - but success depends on real insurance expertise paired with real tech skill.
- Using AI as a research copilot is the most reliable current application
- Predictive modeling for perils and loss trends is gaining ground
- Insurtechs need deep insurance knowledge and credible teams, not just code
Where AI delivers today: research and document intelligence
Across Altamont Capital Partners' portfolio, AI works best as a research assistant. Think fast ingestion, sorting, and surfacing of relevant details across large volumes of documents and data. The value is simple: it flags what matters so underwriters, claims pros, and actuaries can make better calls faster.
Tools that act like a "copilot" help teams find anomalies, summarize filings, and highlight unusual exposures. AI does the grunt work; humans decide.
Data focus: exposure vs. experience
Zuk sees progress with experience data (what actually happened), while exposure data (what could happen) remains the hard part. That gap shows up in property schedules, vendor data quality, and the context needed to make exposure truly predictive.
Translation: you can speed up underwriting with better experience data pipelines today, but you still need disciplined processes to clean and structure exposure data.
Visual AI: imagery that sharpens claims and exposure
Altamont has worked with vendors using imagery to detect small changes a human might miss. That could be roof wear, property additions, or scene validation for auto claims. Post-event analysis can also test whether a claim narrative is physically plausible.
Visual AI won't replace adjusters, but it gives them sharper eyes and better prompts for investigation.
Predictive modeling: perils, probabilities, and practical deployment
The other high-impact lane is predictive modeling. Using historical weather and catastrophe data to estimate the likelihood of events and probable damage within time windows helps pricing, aggregation, and portfolio management.
Insurers pairing internal loss histories with trusted external data sources are seeing the most lift. For reference frameworks and governance, see the NIST AI Risk Management Framework here, and for climate and weather baselines, NOAA's historical datasets here.
What too many insurtechs get wrong
Zuk calls out a common issue: underestimating insurance. Regulatory nuance, program structures, and distribution realities aren't trivial. Many founders assume brokers are going away - he's not convinced.
Teams that ship fast but skip insurance fundamentals burn trust and runway. The market notices.
What Altamont wants to fund
- Real domain expertise: Understand the exact part of the value chain you're fixing.
- Deep AI/tech skill: Clear architecture choices, data strategy, and deployment plans.
- Credible, diverse teams: Not all insurance people. Not all tech people. A balanced bench that's actually shipped in production.
Funding outlook: interest is high, money is pickier
Deal flow is there, but last year was down and capital is more selective. Interest rates and liquidity still set the tone. AI is real - the open question is returns for investors at scale.
Expect cycles. Quality will get funded; vague pitches won't.
Practical moves for carriers, MGAs, and brokers
- Start with document and research copilots across underwriting and claims. Measure time saved and decision lift.
- Build a clean experience-data pipeline first; then tackle exposure data with clear ownership and standards.
- Pilot visual AI on specific lines (property, auto) with tight feedback loops and human review.
- Stand up a governance checklist aligned to recognized frameworks (bias, auditability, data lineage, model monitoring).
- Staff the intersection: pair senior insurance operators with seasoned ML engineers. No tourists.
The takeaway
AI is here to stay in insurance, but the winners keep it practical. Research copilots, predictive modeling, and targeted vision use cases are paying off now. Pair real insurance expertise with real AI skill - and build teams that can ship, learn, and iterate.
If your team needs structured upskilling for underwriting, claims, or data roles, explore industry-focused AI learning paths here.
Your membership also unlocks:
