Newsletter · · Ashutosh Agarwal
AI Goes After Insurance's Boring Middle - Vertical Spotlight: Insurance - Week of August 11, 2026
How AI is landing in insurance's back office, from natural-language pricing to voice agents and agentic prospecting, drawn from operator and founder podcasts for the week of August 11, 2026.
Vertical Spotlight: Insurance
Week of August 11, 2026: AI Goes After Insurance's Boring Middle
This week's rotation comes back around to Insurance. The theme across the podcasts: nobody was building a robot underwriter that replaces human judgment. They were pointing AI at the parts of insurance that people quietly hate, cleaning data, answering the same phone call for the tenth time, reading 20-page rate filings, chasing renewals, and the reason keeps coming back to one thing: the industry can't hire fast enough to do that work by hand anymore.
The Landscape
If you want to know where AI is actually landing in insurance right now, don't look at the moonshots. Look at the pricing desk, the front desk, and the prospecting list. That's what this week's episodes were about, and the through-line is unusually consistent: the technology is being aimed at the drudgery, not the judgment, and the thing pulling it in isn't hype, it's a labor shortage.
Start with pricing, because it's the most technical corner and the most telling. On the InsTech podcast, Dani Katz, co-founder of the London pricing-software firm Optalitix, described building an AI feature that lets an underwriter "talk to quotes in natural language" instead of clicking through a system, ask a question in plain English, get the answer, move on. What's striking is why he says he built it that way. Optalitix ran a series of interviews on "pricing transformation" (the industry's phrase for dragging pricing off spreadsheets and onto modern software), and the people meant to benefit were the least impressed:
"Underwriters seem the least excited about the progress that they made, with only 34% seeing significant progress in their company's pricing transformation efforts, which is pretty fundamental."
So the design brief flipped. Instead of forcing underwriters onto the "best" tool, Optalitix built AI around the tool they refuse to give up, Excel. In their survey, underwriters rated Excel above every other pricing tool on flexibility, speed, and collaboration; the one and only category where it lost was future-proofing (it doesn't connect to everything). Katz's point for founders: the winning tool isn't the most powerful one, it's "the one that the humans like and can use easily." A separate survey number he gave lands the same nail, in the world of MGAs (managing general agents, the specialist firms that underwrite on a carrier's behalf), actuaries were "spending 50% of their time sorting the data out before they actually got around to doing their job." That's the half of the workday AI is going after.
Then there's the front desk. On Insurance Town, Francisco Lopes, co-founder of the voice-AI startup Sonant, made the labor point bluntly: agencies adopt an AI that answers the phone not because they want to fire anyone, but because "it's really hard to hire good people. You have turnover, natural turnover... people that are retiring." The same retirement wave showed up, from the owner's chair, on Business Refocused, where agency principal Kyle Becker said he has four staff who have "crested 60" and could retire tomorrow, and is deliberately choosing to backfill them with AI tools rather than people. His line for anyone still dithering: "We're burning the boats. We are absolutely not going back."
The third corner is prospecting, and this is where the week got genuinely modern. On the Power Producers Podcast, Rob Gifford of the data firm InsuranceXDate described wiring public records (trucking-safety filings, nonprofit tax returns, workers'-comp data, and state rate filings) into a system that a producer's own AI assistant can query directly. More on the mechanics below, but the punchline is that a salesperson can now type one request and turn a single sales appointment into six well-researched ones. Host David Carothers, not a man given to understatement, stopped him mid-sentence: "I don't know what else you could possibly want as a producer other than people calling you to say, hey, I want to buy from you."
Companies to Know
Optalitix, AI-assisted pricing for underwriters and actuaries. London-based, founded by Dani Katz, and now, he says, serving five reinsurers globally alongside primary insurers and MGAs. Over the past year it added a modern "data lake" (Afterlytics) that removes storage limits, a new API-enabled interface, and Python hosting for actuaries. The headline is an underwriting-AI feature launching in the second half of the year that lets underwriters query quotes in natural language, and a parallel capability that lets actuaries pull data via Python, SQL, or plain-English AI queries. Katz's framing of what actuaries actually want is the useful bit, "they don't necessarily want magic. They want simplicity, transparency, and control", which is why, he says, the best AI for an actuary "is going to be almost invisible." The company's origin is a good tell for the whole space: its very first project was converting an Excel model into a system, and it discovered "how important Excel was... but also how complex it was to convert it." InsTech, "Dani Katz, Co-founder & Director: Optalitix: The human side of AI-powered pricing (414)" (August 9, 2026)
Africa Specialty Risks (ASR), AI underwriting for markets no model covers. CEO Mikir Shah, a Kenyan actuary-turned-investment-banker, has built ASR from 8 people in 2020 to nearly 200 today, doubling premium every year and on track for around $500 million of gross written premium this year, with a recent growth investment from private-equity firm Vitruvian Partners. The AI story is what makes it a Vertical Spotlight item. Because ASR launched during COVID, everything was cloud-based from day one, and after five-plus years it has amassed close to six terabytes of data on Africa, "more data on Africa than anybody else does", which it now feeds into underwriting. Its "ASR 24-7" platform is a fully AI-driven system that takes a risk from quote all the way through to bind, then feeds exposure management, policy admin, and credit control off the data it captures. Shah says a fully automated, AI-driven trade-credit underwriting product launches "in the next two months." The most vivid example of AI-enabled underwriting in a data-poor market was a parametric cyclone cover for Mozambique (parametric means it pays out automatically when a measurable trigger, wind speed or rainfall, is hit, rather than after a claims adjuster inspects the damage): ASR divided the country into 40,000 hexagons, each with a value, so it could model exactly what got hit when "the wind blew or the rain fell." When Cyclone Freddy struck, twice, ASR paid out in eight days, which Shah says let Mozambique rebuild bridges and roads fast. His reason for automating the small stuff is pure unit economics: automation "brings that minimum premium down," making it affordable to properly underwrite the small developing-market risks that London underwriters historically wouldn't spend time on. The Voice of Insurance, "Ep313 Mikir Shah CEO Africa Specialty Risks: Creating new Markets" (August 4, 2026)
Sonant, voice AI that answers the agency phone. Founded by Francisco Lopes (a Portuguese physicist-turned-founder who previously built a social-media tool called Link), Sonant builds AI phone agents for insurance agencies and, unusually, owns its voice technology end-to-end, it "built it from scratch" because there weren't robust alternatives when it started. The metric that matters: one agency it works with went from handling 70 tasks a day to 100 a day, "42% more productive", with the same staff, simply because the AI answers routine calls, collects the information, and routes it to the right person. Sonant also cites its own consumer survey: 83.7% of consumers have already used AI in a customer-service interaction, and nearly 70% of insurance buyers say they'll happily talk to an AI "if it helps them get answers faster." Lopes is refreshingly honest that the voice quality isn't the point, the integration is. A generic voice bot is "a glorified voicemail"; the value comes from plugging into the agency management system so the AI can pull a customer's details in real time, "way faster than what a human could even do." It already quotes pet insurance live on a call. And the detail nobody will forget: one agency gave its AI the voice of the owner's late father, so longtime customers still hear a familiar voice on the line. Next stop, per Lopes, is the broader "agentic" wave, AI that doesn't just talk but does the follow-up tasks itself. Insurance Town, "Can Voice AI Transform the Insurance Customer Experience?" (August 6, 2026)
InsuranceXDate (and its X-Rate database), public records wired to your AI. On Power Producers, Rob Gifford walked through what is probably the most concretely "agentic" thing an insurance salesperson can buy today. InsuranceXDate ingests public data sets, Department of Transportation filings, nonprofit 990 tax returns, workers'-comp records, and, newest, state rate filings (the SERFF system, where carriers file the "loss cost multipliers" that drive workers'-comp pricing), and turns the "pile of PDF forms" into a searchable database called X-Rate. The kicker is that they've built an MCP server (Model Context Protocol, the new plumbing that lets a chatbot like Claude or ChatGPT connect to an outside system) so a producer can point their own AI at the database and say, in effect: look at my calendar this week, find five prospects near each appointment, cross-reference their carriers against the rate filings, and flag anyone who just got hit with an increase so I have a reason to walk in. Gifford built the analytics layer using Anthropic's Claude Code, condensing "20 or so pages of PDF filings" per state into a hover-over dashboard showing which lines are getting more expensive, how many policyholders are affected, and how much premium is at stake. Host David Carothers added his own AI workflow on the same episode: dropping a prospect's loss history into Claude to auto-generate a branded PDF report of claim trends, and dragging eight to ten competing excess-and-surplus quotes in to get an instant coverage-gap comparison ranked by errors-and-omissions risk. His result: he says his agency has averaged over $100,000 a week in small commercial premium for the past month, on policies starting as low as $750, a segment he "would have never touched" before AI made it economic. Power Producers Podcast, "Why Better Prospecting Data Creates Better Commercial Producers with Rob Gifford" (August 5, 2026)
A working agency's AI stack (Kyle Becker). Not a startup, but the most honest picture of adoption economics this week. On Business Refocused, agency owner Kyle Becker described a deliberate plan to replace departing staff with software: an AI platform (or an in-house bot) to take over accounting so his operations manager only has to spot-check "maybe a dozen policies a month"; a service to automatically remarket the personal-lines book as the market softens and clients start shopping; and a commercial-lines tool that builds ACORD forms (the industry's standard application forms) and sales proposals so his younger producers "walk in confident." His governing metric is revenue per employee, and his warning is that it's now a slippery number, because agencies have inflated headcount with cheap virtual assistants. He's candid that these tools are unproven and disposable ("they may be here a year and we may go to the next thing... you have to teach it because it's AI"), but the strategic bet is firm: as staff retire, "we may or may not replace them, or all of them." Business Refocused, "#183 - How Benchmarking Drives Smarter Growth for Insurance Agencies w/ Kyle Becker" (August 4, 2026)
Franklin Madison, decades of data newly AI-charged. A quieter but instructive item. On the Insurance Leadership Podcast, an executive at Franklin Madison, a long-established affinity marketer that sells insurance through banks and credit unions, described using AI two ways: to push campaigns out faster, and, the part she's "most passionate about," to know customers well enough to match a specific product to a specific persona before spending on a campaign. The firm has its own data scientists on staff and analyzes historical campaigns "everything down to specific words." The counter-intuitive detail: direct mail is thriving, precisely because inboxes are now the junk drawer, "your inbox is full of junk and people might be excited to open up their mailbox." AI's job here is to make sure the thing in the mailbox is relevant. Insurance Leadership Podcast, "Making an Impact" (August 6, 2026)
One Debate: Does AI grow the insurance workforce, or quietly shrink it?
Here's the week's genuine disagreement, and what makes it interesting is that both sides showed up in the same seven days, one as a thesis, the other as lived experience.
The optimistic case came from Dani Katz on InsTech, who was asked directly whether we're heading for a "job armageddon" where trainee and manual roles vanish and no graduate ever gets hired again. He pushed back hard, and his argument is worth taking seriously because it's specific to insurance, not generic AI cheerleading:
"I think there are going to be more jobs rather than less... One of the big things is that the world is completely underinsured. When a natural catastrophe happens, around 70% of those claims are not actually covered."
His logic: AI removes the drudgery, which frees the people already in the industry to do the thing the industry is bad at, expanding coverage to the enormous population that has none. More coverage means more products, more markets, more people. "Insurance is a people game. Expand the insurance cover, and we will need more people." He even reframed it as a recruiting fix: strip out the boring parts and "suddenly insurance becomes exciting again for young people." Katz's own view is that actuaries should move up the value chain, become "better translators of risk" who sit with underwriters and portfolio managers, not disappear.
Now set that against the same week's agency owners. On Business Refocused, Kyle Becker is buying AI for the explicit purpose of not replacing four staff as they retire. On Insurance Town, Francisco Lopes was clear that agencies come to Sonant because they can't hire, the AI fills a role a human isn't taking. Neither is planning the coverage expansion Katz describes; they're planning to run the same book with fewer people. That's not job creation. It's a hiring freeze wearing a friendlier face.
So who's right? The honest answer is that they're describing two different layers of the industry. At the frontier, pricing new risks, opening data-poor markets like ASR is in Africa, Katz's story holds: automation is a growth lever, and Mikir Shah's answer to "where's the constraint?" this week was simply "we can't hire enough people." But at the distribution layer, the thousands of independent agencies running personal and small-commercial books, the lived reality is Becker's: AI is a way to hold capacity flat while an aging workforce ages out. The optimists and the realists aren't contradicting each other so much as standing in different rooms of the same building.
The tell to watch, if you're investing or building here: whether the productivity AI unlocks gets reinvested into writing the 70% of risk that's currently uncovered, Katz's virtuous circle, or simply pocketed as a smaller wage bill. This week, the specialist underwriters were doing the former and the retail agencies were doing the latter. The vertical's future depends on which behavior wins.