Full Episodes | Insurtech Leadership Podcast

Introduction

Carriers are rewriting decades-old systems, and more and more of the new code is written with AI. If the AI that wrote the code is also the thing checking it, who is actually validating what reaches the policyholder?

Alex Rodov has spent more than thirty years in software testing, and his answer is that the testing has to be independent and has to start from a complete map of how a carrier's systems actually work. He explains why most carrier testing is still manual, why Emtech keeps AI out of the testing itself, and why the same map turns out to be what a carrier's enterprise AI has been missing.

Guest Bio

Alex Rodov is the founder of Emtech Group, which builds autonomous software testing for insurance carriers and insurtechs, starting with life insurance. He founded QA Consultants in 1994 and built it into one of the largest independent software testing firms in North America. When ALTEN acquired the firm in 2023, he carved out its emerging technology division, a team of PhDs and engineers building testing accelerators, and it became Emtech. He is the author of The Book on Testing and sits on LIMRA's Strategy and Innovation Committee.

Key Topics

-Who checks the AI's work - If AI tests code against the same understanding it used to write it, flawed code passes every test, so validation has to come from someone outside the build.

-Testing is still mostly manual - The knowledge of how a carrier's systems behave sits in the heads of long-tenured testers, and Alex says manual coverage at most carriers lands around 10 to 20%.

-The digital blueprint - Emtech maps a carrier's systems from the e-app through underwriting and policy admin to the issued policy, which Alex says adds up to hundreds of millions of test cases.

-No AI in the testing - Emtech builds the blueprint from deterministic data and runs on-site on desktops, because most carriers won't expose their back end to an outside AI.

-The enterprise AI licenses carriers aren't using - Alex argues that carrier AI is blind to the back end, and that giving it an accurate blueprint is what makes it useful, from user guides to product filings.

-Regression across legacy and acquired systems - End-to-end regression covers mainframes, acquisitions and issued PDFs, and runs in parallel on testers' desktops in days instead of months.

-The future QA team - Alex expects testers to become the stewards of AI inside carriers, because they are the people who understand the systems end to end.

Notable Quotes

"We need technology to test technology, but it cannot be the same technology."

"The knowledge that people have in their heads is often not documented."

"If you take that and you give that to the enterprise AI, that's the missing piece."

"Testers need to become the stewards of AI enablement within organizations because they're the subject matter experts really understanding end-to-end."

Resources

Guest:

Emtech Group: https://emtechgroup.com/

The Book on Testing: https://books.friesenpress.com/store/title/119734000184141596/Alex-Rodov-The-Book-on-Testing

Alex Rodov on LinkedIn: https://www.linkedin.com/in/alexrodov/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

Insurers have spent two years funding AI pilots, and their boards are now asking what those pilots delivered. Why can so few carriers give a clear answer?

Jim Quick leads the insurance advisory sector for PwC in the US, and PwC is four years into studying which companies get real value from AI. His finding is that carriers taking an incremental approach are seeing the least of it. The carriers getting the most picked their hardest problems, such as new business and underwriting or how fast claimants get paid, and rebuilt the whole process.

Guest Bio

Jim Quick is the Insurance Advisory Sector Leader at PwC US, where he oversees the firm's consulting platforms for insurers, including deals, AI, engineering, strategy, customer, and risk and regulatory work. He has spent about 25 years consulting to the insurance industry across personal and commercial P&C, life and annuities, and group benefits, and most of his direct client work today is in life and group benefits. PwC has more than 5,000 insurance-focused professionals in the US and more than 14,000 globally.

Key Topics

-Going big beats going incremental - Across PwC's AI performance study, the companies getting the most value attack their hardest problems, which Jim calls a culture and leadership challenge more than a technology one.

-A practical sequence - Get hands-on with the tools, spend a couple of weeks rethinking the work, go after the toughest problem with leadership backing, and move to the next hardest problem if the first one doesn't work.

-Growth over efficiency - Most of the market started with cost savings, while PwC's most innovative insurance clients are aiming AI at growth, including bundling products and answering quote requests faster.

-Tokenomics - Some companies have watched AI savings get eaten by token and data center spend, and the real cost of deployment has become a C-suite question.

-Agents on the org chart - Carriers are working out how to account for and manage AI agents, and whether AI engineering should sit in a central office or inside each business unit.

-Build is back, and hardening is the scarce skill - Jim expects far more custom development on the hyperscalers and frontier models, and says the hard part is getting a pilot into production with token usage contained and an audit trail in place.

-Finish the core - Carriers still need a modern, auditable policy, billing and claims core, so Jim's advice to anyone mid-implementation is to finish it and use AI to deliver it faster.

Notable Quotes

"It's really not a technology problem at its core. It is a cultural, it's a leadership, and it's the willingness to really reimagine and be willing to take risks."

"I like to say we're kind of making underwriting fun again, or making underwriting underwriting again."

"It's pretty easy to build things in a pilot environment. What's much harder is to actually get it into production, contain the token usage and harden the solution so that it's auditable."

"If you haven't done it, I think you should. And if you're in it, finish it."

Resources

Guest:

PwC US Insurance: https://www.pwc.com/us/en/industries/financial-services/insurance.html

PwC, "Reinventing insurance: An industry beyond the tipping point": https://www.pwc.com/us/en/industries/financial-services/library/reinventing-insurance.html

Jim Quick on LinkedIn: https://www.linkedin.com/in/jiquick/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

-Introduction

What does a mid-sized brokerage have that it cannot actually use? A specialized book and the data behind it, and none of the underwriting, capital, filing, and admin machinery it would take to turn that book into a program it owns. The ten largest firms solved this by buying MGAs. Cole Riccardi built Authentic to rent that machinery to everyone else, and nine of the top ten retail brokerages now work with him on digital programs anyway. In this conversation he explains why he refuses to call Authentic a software company, what he got wrong in the first two years, and why he went after Main Street before he went after the hard market.

Guest Bio

Cole Riccardi founded Authentic in 2022 after five years at Aquiline Capital Partners, where he invested across the insurance value chain and became convinced that nobody was building affinity insurance programs at scale. He built the company with a co-founder and COO who spent fifteen years as an actuary at AmTrust and a CTO who came out of six years at Amazon, and the platform was architected as multi-carrier, multi-tenant, and multi-product from the first day. Authentic runs out of New York and Dayton, Ohio, is backed by FirstMark Capital, and now sits behind six carriers with programs spanning Main Street, energy, and inland marine.

Key Topics

-Programs should not be limited to the top ten - The largest brokerages own their own MGAs. Riccardi's argument is that everyone else already has the book and the data, and only lacks the machinery, which is a rentable problem.

-An MGA that gives its software away - Riccardi is emphatic that Authentic never wanted to sell software and makes its money in the transaction flow. The technology is what makes it a better trading partner, not the product.

-Affinity is product creation, not just distribution - For PushPress, Authentic went through the entire book and built four or five custom class codes that mapped to CrossFit gyms and similar facilities, rather than rating them off generic fitness codes.

-Main Street first, on purpose - Rather than chase hard-market classes where premium comes fast, Authentic built a sizable Main Street book to prove that the configured programs were what moved business over, not market scarcity.

-Same infrastructure, new practices - A Chubb energy underwriter joined with a thesis about independent power producers being underserved on casualty. That program launched in May, and an inland marine practice was announced the day of this recording.

-The captive he overbuilt - Riccardi is direct that as a first-time founder he overcomplicated the early years by building custom captive cells on the back end. The captive still exists and takes smaller slices, but it is no longer the pitch.

-Fast nos and the right to win - Authentic has no trucking underwriter and no capacity partner who will raise a hand for it, so brokers asking about long-haul trucking get turned down quickly rather than strung along.

Notable Quotes

"We never wanted to sell software. I always wanted to be in the transaction flow. We make money as an MGA today. That's always how we will make money."

"Nine of the top ten work with us today on digital programs."

"We've actually gone after Main Street to start and built up a pretty sizable book there. We wanted to do that so we could prove to ourselves that our technology mousetrap mattered and we weren't just writing business because it was hard to find markets elsewhere."

"Can we challenge the status quo that MGAs need twelve points to operate? If you're built on modern systems, I don't think you do."

Resources

Guest:

Authentic: https://authenticinsurance.com/

Authentic, For Brokers: https://authenticinsurance.com/brokers

Cole Riccardi on LinkedIn: https://www.linkedin.com/in/cole-riccardi-63a243127/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

What is an insurer actually buying when it signs an AI contract? Dan Griffith's answer is that the failures came from companies buying AI because they needed AI, and the wins came from companies buying a migration that finishes faster at a fixed cost. Tim Vieyra runs Comotion AI, a Cape Town firm that spent twelve years building analytics and policy administration software other companies put their name on, and is now selling in the US under its own. Dan is advising Comotion on that entry, so this conversation runs two threads at once: what has to be true about an insurer's data before any AI project is real, and what has to be true about a company before it tries to sell into the American market.

Guest Bios

Tim Vieyra is a qualified actuary who left corporate financial services in 2013 to co-found Comotion, a software and data firm serving life insurance, investment, and healthcare clients. Comotion built the analytics engine behind TAI Insights and has worked with RGA South Africa. Comotion AI is the platform version of that work, putting governed, queryable structure over insurance, finance, and mortgage data, with data migration and profiling as the lead use case. Dan Griffith spent thirty years in enterprise sales, including a run as the first US hire for a South African company he helped scale from $3M to $150M, and now runs Greater Gain Group, which helps international software companies land and scale in US insurance and financial services.

Key Topics

-Buying an outcome, not AI - Griffith argues the failures came from companies that bought AI because they needed AI. What Comotion's customers buy is a migration that lands faster at a fixed acquisition cost.

-Profiling in week one instead of month six - Vieyra explains that migration teams have historically found their data problems only after loading a target system and running reports. Comprehensive profiling on ingest moves that discovery to week one or two.

-Governance resonated least - Griffith expected governance to carry the pitch and found it landed weakest, because governance is not a problem until it becomes one and the industry has not settled what the word means.

-Narrowing the message on purpose - Vieyra describes shifting from "we can help with a broad array of problems" to two named use cases. That limits bandwidth, but it cuts through the noise faster.

-South Africa versus the US - A smaller market rewards land-and-expand relationships and tolerates a muddier value proposition. The US demands a dialed-in problem statement and credibility, and then it opens up.

-ICP first, everything else after - Griffith walks through the full-day workshop that set Comotion's ideal customer profile, and why a pilot should be priced as a percentage of the full deal and bounded by either time or scope.

-Reset advice for stalled AI programs - Vieyra's answer for anyone whose AI initiative is not working is to re-evaluate the project against what the team knows now, not what it knew when the decision was made.

Notable Quotes

"Hold on to your paradigms loosely. Be humble enough to know that what you know now is not what you knew when you made the decision. And be willing to cut your losses and move to things that work."

"We see failures all over the place because people just bought AI because they needed AI."

"Governance isn't a problem until it's a problem."

"In the US you have to have a much more dialed in and focused value proposition. You have to have a very, very clear problem that you're solving and credibility on why people should believe you. If you tick those boxes, the market is open to you."

Resources

Guests:

Comotion AI: https://www.comotion.ai

Comotion Business Solutions: https://www.comotion.us

Tim Vieyra on LinkedIn: https://www.linkedin.com/in/timalive/

Greater Gain Group: https://www.greatergaingroup.com

Dan Griffith on LinkedIn: https://www.linkedin.com/in/dangriffithsr/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

Why would an industry that governs almost everything still leave AI sitting on the back burner in the workflows where it would matter most? Earnix research puts AI in most or some workflows at 81% of insurers, and 92% of them run formal AI governance reviews, but fewer than one in three executives strongly agree those reviews are keeping pace with regulation. Aaron Wright, VP of Professional Services for the Americas at Earnix, spends his working life on the far side of that gap, and he argues it is not really a technology problem at all.

In this episode, Josh Hollander and Aaron work through what separates transformation from transformation theater, why modernization programs tend to fail three months before go-live rather than at it, and where AI is already moving numbers a CFO would recognize.

Guest Bio

Aaron Wright is VP of Professional Services, Americas at Earnix, and a Fellow of the Casualty Actuarial Society. He started out of school with a statistics degree as an actuarial analyst at Farmers Insurance, and was on the phone with a Department of Insurance defending the math behind a rate change within six months. He then spent roughly twenty years at USAA across pricing, underwriting, state management, catastrophe management and data science, closing out with a modernization mandate. He now leads the teams that arrive after a carrier has bought the technology and someone has to make it actually work.

Key Topics

-The governance gap - Formal reviews are nearly universal, but the confidence that they track where regulation is heading is not, so real operational decisions move at a measured pace.

-"We have AI" versus AI on the balance sheet - Aaron separates the places where AI is moving cost per ticket and handle time from the places where it is still hype.

-Where programs actually fail - Not at go-live, but months earlier, when somebody has to kill a customization or force a business unit to change and nobody holds the authority to make that call.

-The scaffolding problem - Carriers have built so many workarounds around legacy systems that the workarounds, not the systems, are often what hold them back.

-Parallel-process fatigue - Running the old process alongside the new one creates burnout that can kill a project's momentum, and only leadership can set expectations around it.

Vibe coding has a ceiling - Aaron will build a prototype to show the art of the possible, and he will not put it into a governed insurance environment.

-AI that enables expertise - One Earnix customer sees about twelve months of average tenure in the call center against eighteen months to train a representative fully, which is the kind of math that makes embedded, governed AI worth the effort.

Notable Quotes

"Saying you have AI and actually having it be something that hits the balance sheet that your CFO cares about is a very different thing."

"Most programs are not going to fail at the go live. They're going to fail three months before it, when somebody has to make that unpopular call and there's nobody with authority to actually make it happen."

"The average tenure for call center representatives is about twelve months at their company, and it takes eighteen months to get them fully trained. Well, that doesn't work."

"Our view of how AI should be used within insurance is to enable expertise."

Resources

Guest:

Earnix: https://earnix.com/

Aaron Wright on LinkedIn: https://www.linkedin.com/in/aaron-wright-fcas/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

Ask an insurance operator to write down how a policy endorsement actually gets processed, and you will usually get about six steps. The real number is closer to fifty. The other forty-four live in the head of whoever has been doing the job longest, and they leave when that person does.

Giancarlo Stanton, Co-Founder and CEO of Magiq, has built a company around that gap. His argument is that you have to know what the process actually is before you automate it, and most operations don't. The industry is buying agents to handle work nobody has written down, which is how you end up with a system whose decisions nobody can explain.

In this episode, Josh Hollander and Giancarlo get into what it takes to pull a process out of a competent person's head, why the exceptions matter more than the normal path, and what changed in one support operation once someone finally mapped the work.

Guest Bio

Giancarlo Stanton is Co-Founder and CEO of Magiq. He started in complex commercial litigation, moved in-house, and became General Counsel and VP of Claims at Swyfft before serving as Chief Operating Officer at Cover Whale. Along the way he advised Obie, FRG Specialty, and Sunfish. That path put him in the back office of regulated businesses during periods of fast growth, which is where he formed the view that the minutiae are the whole game. He founded Magiq with Nick Eich, previously at Groupon and then on the engineering team at Kin, to build the thing he kept needing as an operator: a way to go from what should get done to proof that it actually happened.

Key Topics

-Your best people cannot describe their own jobs - When something becomes routine it becomes invisible, so asking an expert to write down what they do all day produces a partial answer at best.

-The mirror problem - Magiq shows an organization its own process as steps one through fifty, and the response is almost always that leadership knew about step one and step fifty and assumed the rest.

-Insurance runs as an ecosystem, not as silos - When billing misses a cancellation for nonpayment, claims pays a loss that should never have existed, which is why Giancarlo argues point solutions aimed at one function cannot fix the operation.

-The exceptions are where the value sits - The flat cancel and rewrite that one person handles differently today than yesterday is what turns a bordereau into a conversation with your reinsurer about why the math doesn't math.

-What mapping the work actually did - A support operation went from a one-in-five call response rate and an 8.8 day email response time to 96-97% call response and 5.8 hours, seven months in.

-Sales is a process too - Mapping the inbound lead decision tree and empowering producers to end calls that were never a fit drove a 23% improvement in per-producer output and a $10 million gross written premium lift over a seven month engagement.

-The staffing question - Giancarlo makes the case that doing more with the team you have lands better with buyers than headcount reduction, and that a business only profitable with no employees was never really profitable.

Notable Quotes

"You can't automate things you don't understand. And that, I think, is just objectively true. But that doesn't stop people from trying."

"No one is sitting there running a business saying, I need more AI. They're saying, I need less cost, or I need more sales, or ideally both. And probably fewer errors."

"People don't do things incorrectly because they want to. They do things incorrectly or inefficiently because they think they're doing them correctly, but it's eight thirty at night and it's their fiftieth endorsement."

"If the only way your business is profitable is for there to be no employees, then you probably just don't have a profitable business to begin with."

"It's really about making sure the mistakes don't happen, not making sure the people aren't there."

Resources

Guest:

Magiq: https://usemagiq.com/

Giancarlo Stanton on LinkedIn: https://www.linkedin.com/in/giancarlo-stanton-1b011a59/

Magiq on LinkedIn: https://www.linkedin.com/company/usemagiq

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

What does an insurer actually own after a few years of buying AI one use case at a time? Feathery co-founder Zack Khan argues that the answer decides whether AI ever produces more than incremental gains, because a workflow you cannot change without filing a vendor ticket is a workflow you are renting. Fresh off a $30 million round led by Portage, with Allstate and Erie both in as strategic investors, Khan walks host Joshua R. Hollander through what it takes to put workflow control in the hands of the operating teams themselves.

Guest Bio

Zack Khan is Co-Founder of Feathery, the AI operating and decisioning system for financial services. He started as a software engineer building complex forms at Robinhood and Nextdoor alongside co-founder Peter Dun, then became the fifth employee at Hightouch, where he led marketing as the company grew toward a $3 billion valuation. Khan and Dun founded Feathery in 2021 as a developer-focused form builder and grew it into a platform that now serves more than 300 firms, including Tokio Marine, Hiscox, Baldwin Group, and Hylant, orchestrating submission intake, virtual inspections, quoting, and benefits workflows across carriers, MGAs, and brokers. Feathery ran profitably before raising its $30 million round in July 2026.

Key Topics

-Vendor-led vs. team-owned workflows - Why the last generation of insurance software put every change behind a vendor ticket, and what changes when underwriting and operations teams define workflows in natural language instead.

-Copilot licenses vs. step changes - Giving everyone a copilot produces small gains; redefining one workflow end to end took a carrier from an eight-hour time to quote down to fifteen minutes.

-Orchestrate rather than replace - Feathery's agent, Robin, works across existing rating engines, AMSs, and even legacy desktop applications, automating the data entry instead of ripping out the system.

-Rules where you want them, judgment where you need it - Deterministic guardrails handle rating inputs, while objective-driven agents handle tasks like checking a virtual inspection for a pool or a tree touching the roof.

-The Baldwin Group benefits example - Custom employee benefit guides that took 20 to 30 or more hours to build now generate in minutes, across 10 to 15 assets and multiple languages, and the team's hours moved to consultative work.

-What a platform actually is - Khan's test is control: if you cannot apply your own business logic to your own workflow without a vendor ticket, you bought a point solution.

-Start in specialty lines - Fast-growing lines with less tech debt and less red tape are where AI transformation finds its first champions inside a large carrier.

Notable Quotes

"An actual platform gives you control. If you don't have fundamental control over the actual experience, and you're applying your own firm-specific business logic to your workflow, that is a point solution."

"If your rating engine is working great, it's just the annoying part is typing data into it. You don't rip out the whole rating engine; you just automate the data entry part of it."

"Some of our customers have gone from an eight-hour time to quote to less than fifteen minutes responding to producers. That's the stuff that actually will impact your bind ratio."

"For AI to truly have the impact you want, you need to redefine the workflow, think about the end-to-end ideal process that you want your best underwriter, your best claims adjuster, your best producer to be following."

Resources

Guest:

Feathery: https://www.feathery.io/

Zack Khan on LinkedIn: https://www.linkedin.com/in/zackkhan101/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

Most carriers can say how many AI pilots they are running. Far fewer can say where every automated decision in the business sits. That gap is the subject of this episode.

Josh Hollander is joined by Karl Canty and Paul Laurent of Artefact, the global data and AI consultancy, for a working conversation about the four frameworks converging on US insurance: the NAIC model bulletin, California's ADMT rules, Colorado's algorithm governance regime, and New York's DFS circular letter. Paul argues that this combination has made US insurance the most completely defined AI governance environment anywhere, and that the scope reaches well past the AI programs carriers are proud of. Rules engines, credit scores, telematics feeds, and vendor models are all inside it.

Karl's argument is that carriers moving on this for the first time are not first movers, they are late, and that the inventory work the rules force on you is the same work that makes your own data usable for growth.

Guest Bio

Karl Canty is Partner and US Insurance Lead at Artefact, where he is building the firm's North American insurance practice. He has spent about twenty-five years in insurance data, technology, and operations, including advisory roles at PwC and EY, a data and AI leadership role at EXL, and a partnership at Capco. He writes for Digital Insurance on where AI actually pays for carriers.

Paul Laurent is Head of AI Risk and Data Trust at Artefact, running the firm's AI governance advisory practice across North America. He describes himself as a technical lawyer: criminal prosecutions early on, a master's in computer science, then a decade at Oracle translating regulatory obligations into working architecture, followed by identity, data security, privacy, and now AI. He has contributed to NIST framework work across cybersecurity, privacy, and AI risk management.

Key Topics

-The four timelines - New York in force since July 2024, an NAIC pilot ending in an AI systems evaluation, California live with enforcement from January 2027, and Colorado already amended toward California's standard.

-Scope is wider than AI - California's rules reach deterministic tools and legacy rules engines, which pulls decades-old underwriting and claims infrastructure into the conversation.

-The vendor chain is not a place to put the risk - The NAIC treats external consumer data and information systems, including credit scores and telematics, as the carrier's responsibility. -Compliance cannot be delegated to a third-party vendor.

-Human in the loop rarely survives the test - California wrote three bright-line conditions for human review. In Paul's assessments, more than eighty percent of processes carriers describe as human-reviewed do not meet them.

-Nobody can produce the inventory - The first question in every engagement is where the business makes significant decisions with automation. Paul gets the same answer every time: we have some, we don't know where all of it is, and there is probably shadow.

-CalPrivacy is already staffed and fining - Roughly 150 complaints actioned, capacity for a hundred at a time, and eight-figure fines appearing.

-Where the work turns into growth - Karl's case for entity resolution, a real AI roadmap, and a durable enterprise AI competency coming out of the same assessment that satisfies the regulators.

-The pause happening right now - Paul's most governance-mature client has stopped agentic AI development until these questions are answered, and the conversation inside carriers has moved from the CIO to the chief data officer.

Notable Quotes

"The most completely defined spot on the global AI governance workspace right now is the US insurance industry."

"The vendor chain for NAIC is explicit. You cannot delegate compliance responsibility to a third-party vendor."

"I don't think that companies who are moving right now for the first time are first movers. I think you're already late."

"If you could tech your way out of this problem, it would have happened already."

Resources

Guests:

Artefact: https://www.artefact.com/

Karl Canty on LinkedIn: https://www.linkedin.com/in/karlcanty/

Paul Laurent on LinkedIn: https://www.linkedin.com/in/paullaurent/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

What happens to a company that has run two hundred proofs of concept and still cannot point to one that paid for itself? Rafael Broshi says that is roughly where financial services sits after two years of generative AI experimentation, and he does not think the models are the reason.

Broshi has an unusual vantage point on this. He built a brand new insurance product, took it to the admitted market, watched it fail to find buyers, and rebuilt his company around the software his team had written to run it. In this episode he walks host Joshua R. Hollander through why demos survive and production launches do not, what has to be written down before a process can be automated at all, and where an AI agent should never be allowed to improvise.

Guest Bio

Rafael Broshi is co-founder and CEO of Notch, an AI platform for building and deploying agents inside highly regulated industries. He spent nearly ten years in the Israeli military, as did both of his co-founders, and the three of them started Notch about five years ago as something else entirely: a specialty product insuring digital assets including crypto and social media accounts, launched in the admitted market across 47 states with Hartford Steam Boiler. That product did not find a market. The platform the team built to run their own MGA became the company. Notch now deploys agents across customer experience, claims intake, document ingestion, and underwriting workflows for insurers, banks, and asset managers, and is backed by Lightspeed, Jibe Ventures, Munich Re Ventures, and Phoenix.

Key Topics

-Why a new insurance product is the wrong startup bet - Cyber insurance launched in 1997 and took until roughly 2015 to catch on, and no startup survives an eighteen year runway regardless of how much it raises.

-Innovate on distribution, not on the product - Broshi argues that the insurtechs that generated real returns over the past decade changed how insurance was sold rather than inventing new coverage.

-What happens between a working POC and a failed launch - Agents are non-deterministic, so they break as knowledge, guardrails, and process complexity scale, and by the time a project reaches that point its trajectory is easy to change.

-Why he does not believe in forward deployed engineers - He calls the FDE model system integration under a new name, and asks what happens at use case number two hundred inside a five thousand person carrier.

-Why breaking down the process is the part that actually fails - He ties the 75% digital transformation failure rate to the difficulty of mapping a process across departments, and describes AI agents as RPA 7.0 with the same underlying requirement.

-Where agents stay probabilistic and where they must not - Empathy in a first notice of loss conversation cannot be scripted, but the order of the fifteen intake questions cannot be left to a model.

-The subagent architecture behind auditability - Fifteen to twenty subagents run concurrently in a single conversation, each owning one question, so no action happens without a model that can explain it.

-What disqualifies a process entirely - No API access to the data, prediction problems that belong to older machine learning, and any process with no written SOP.

Notable Quotes

"With Gen AI it's extremely easy to show a demo and to show something working in a POC, but when it reaches production, things go south extremely fast."

"Launching a new agent is a business transformation to a specific department. And it's an HR problem as much as it's an IT problem."

"The scars that you now carry on your back, those are your only assets that are worthwhile."

"Everyone says they're doing AI. Don't choose vendors according to presentations. Always do multi-vendor POCs."

Resources

Guest:

Notch: https://www.notch.cx/

Rafael Broshi on LinkedIn: https://www.linkedin.com/in/rafael-broshi-435a77105/

Notch on LinkedIn: https://www.linkedin.com/company/notchapp/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.

Introduction

What happens to a claims operation when the manipulated photo comes from an ordinary policyholder instead of a fraud ring? Verisk's 2026 fraud study found that 55% of Gen Z consumers and 49% of millennials would make a small rule-bending edit to a claim photo or document, and Aviva stopped £233 million of suspected claims fraud last year with a growing share of it built on AI-generated images and manipulated paperwork.

Jeppe Nørregaard spent seven years building AI systems to detect fake content before deciding the detectors were going to lose. He now runs a company built on the opposite bet, which is that you protect real content instead of chasing fake content. In this episode he walks host Joshua R. Hollander through why detection breaks down under generative AI, what signing content at the point of capture actually asks a carrier to change, and why a flat fraud number next year should worry a claims leader rather than reassure them.

Guest Bio

Jeppe Nørregaard is co-founder of InReality, a Copenhagen company that cryptographically signs photos, video, and documents at the moment they are captured so whoever receives them can prove the content is real and unaltered. He holds a PhD from the Technical University of Denmark, where he worked on AI and misinformation, and his research on how fake news detection degrades over time and under attack is among his most cited work. He did a research stay with an American research group in upstate New York studying AI for misinformation detection, and he sat alongside the researchers building deepfakes while he was building the detectors. InReality works today with news organizations and insurers, and is backed by the Deloitte Innovation Accelerator, the IBC Accelerator, and Antler.

Key Topics

-Why detection loses the arms race - Across seven years of research, Jeppe watched AI strengthen the attacker faster than it strengthened the defender, first with bots and automation and then with deepfakes.

-Flipping the problem - InReality does not evaluate fake content at all. It builds cryptographic guarantees around real content and treats everything outside that pipeline the way the internet already works.

-What a flat fraud number actually means - If detected fraud stops climbing while deepfake use keeps rising, the likelier explanation is that carriers have stopped seeing it rather than that it stopped happening.

-Where signing happens inside a claim - The policyholder photographs the damage through the carrier's own app, and the security sits inside that capture rather than as a review step afterward.

-Cryptography instead of prediction - Detection produces false positives because it is a prediction system. A signature either verifies or it does not, and InReality uses post-quantum signatures.

-Why they refused to build on a blockchain - A blockchain proof cannot be removed, which collides with GDPR's right to be forgotten, so they built an alternative way to produce the same cryptographic proofs.

-What to do with a detection budget already spent - Jeppe separates deepfake detection, which he would stop funding, from the rest of the AI claims stack, which he says carriers should keep investing in.

Notable Quotes

"We don't actually deal with fake content at all in our company. We completely ignore fake content. Instead, we build security guarantees for real content."

"I sat next to the guys who did deepfakes. They tried to deepfake Barack Obama, and I tried to find misinformation. So I knew where this was headed."

"A lot of money has gone into deepfake detection, and fundamentally we believe that it was a mistake, but it's a mistake that everyone made across the planet."

"If we start to see a flattening or a decrease in the amount of fraud that we find, then I would say we have a real problem, because it most likely means that we don't catch it."

Resources

Guest:

InReality: https://inreality.io/

Jeppe Nørregaard on LinkedIn: https://www.linkedin.com/in/jeppe-n%C3%B8rregaard/

Host & Organization:

Joshua R. Hollander on LinkedIn: https://www.linkedin.com/in/joshuarhollander/

Horton International (USA): https://www.horton-usa.com/

Insurtech Leadership Podcast (LinkedIn Showcase): https://www.linkedin.com/showcase/insurtech-leadership-show

Subscribe & Review

If you enjoyed this episode, subscribe on your favorite platform and leave a review. The Insurtech Leadership Podcast is available on YouTube, Podbean, Apple Podcasts, and Spotify.