Jasmin Flori-Hess · Raindrop
LinkedIn · raindropme.com · CRM #479
Jasmin is a solo-started, now small-team founder (2–10 people) building an insurance ontology: AI that breaks policy wordings into clauses and maps covers, exclusions and definitions into a consistent structure that people and AI agents can rely on. She has just come out of a big POC under extreme pressure, is expanding to the US and Switzerland, is raising capital and hiring, and runs leadership on fractional people (CIO, CPO).
Demand is ahead of capacity. Guidewire Vanguard, InsurTech NY, a carrier POC, Buy Australian AI cohort, US launch partner talks. A tiny team can't onboard carriers, run POCs, raise and expand into three markets at once.
Trust is the product. She keeps writing about evidence, versions, provenance and "whose interpretation is your AI acting on". Enterprise carriers will ask her to prove accuracy. That's the pressure point to listen for.
Corporate career
Senior Project Manager Education at AO Foundation (2017–19), then Marketing & Business Development, Philanthropy Services at UBS Zurich (2021–22). Background: corporate comms, change and project management in finance and medical services. Not an engineer, a builder of bridges and stakeholder trust.
Career break in the outback → the problem
Six months travelling Australia with family. Insuring the trip, it took her 15 hours to analyse eight policies. That's the origin story; she tells it often. Attended the first Swiss InsurTech Hub Summit (Nov 2022) when the idea was called "Retja" (still the legal entity: RETJA Pty Ltd).
Founded Raindrop, seed round, launch
- Founded April 2023 in Darwin.
- Oversubscribed seed round, late 2024 (≈75% from Darwin investors) plus an NT Government innovation grant.
- Platform launched June 2025 after testing with brokers, underwriters and insurers. Database then: hundreds of cyber, business, management liability and strata policies, 6,000+ extracted clauses.
- Stated goal at launch: halve brokers' policy-reading time (≈30 hours/month saved).
Momentum, fast
- Aug: InsurPitch Sydney (Guidewire/PwC).
- Sep: Guidewire Insurtech Vanguard. Only Australian insurtech in the InsurTech NY Match Program (one of 50). Pitched "a System of Meaning, not another tool that reads PDFs". Met a potential US launch partner, US VCs and their US lawyer.
- Sep: Zurich trip, 6–9 meetings a day: Swiss market entry, research collaboration on her methodology, capital raising, possible hires. Joined Swiss InsurTech Hub.
- Oct: Selected for Australia's first "Buy Australian AI Partnership" cohort (Stone & Chalk + National AI Centre). Top 10 go to a Commercial Accelerator in Feb 2027, and she wants in.
- Oct: "Extreme pressure to deliver a big POC." Now done (that's why she said early November).
- 2 Nov: Nominated Best Newcomer, Global Startup Awards.
In her words: "We standardise, categorise and structure the wordings first. Then we take this data and build insurance intelligence in terms of relationships such as dependencies and triggers. A deep PDS analytics that can be used for building, analysing, comparing wording or embedded in existing claims and other workflow tech stacks." (her LinkedIn reply to you, Oct)
"Whose interpretation is your AI acting on?"
When AI drives a claim decision, how do you prove it read the contract right: which policy version, which clauses, what's ambiguous, who approves? Cites AFCA seeing AI-generated complaints with wrong references, and the proposed General Insurance Code of Practice (submitted to ASIC 22 Oct). Raindrop = the interpretation layer with provenance and versioning. You commented on this one.
Adam Jacobs keynote: no "icky buzzwords"
"Slapping an AI wrapper onto an existing workflow often gets labeled groundbreaking innovation… solving a hard, neglected problem does. True value looks like fixing messy logistics, building durable processes, doing the unglamorous groundwork." She'll respect plain talk and dislike hype.
Domain context beats the model
Gallagher Re report: 99% of insurtech money going to AI; "thin wrappers" face platform risk. Her thesis: the moat is the ontology, not the LLM. Escaping "pilot purgatory" with enterprise partners.
Human in the loop
Funny post about an AI photo filter giving people new faces. Her lesson: set boundaries, always check the output, "big advocate for human in the loop and critical thinking". She cares about verification personally, not just as marketing.
She replied fast and booked herself. She's open. Don't waste it with a pitch; earn the second meeting.
Rule of thumb: she talks 70%, you talk 30%. Every answer she gives earns one follow-up ("say more about that", "what does that cost you?"). Take notes on her exact words.
Warm, human, short
- "Congrats again on the Best Newcomer nomination. And you survived the POC!"
- Set the frame: "I've got 30 minutes blocked. I mostly want to understand what you're building and where it's heading. Then we see if there's anything worth doing together. Sound good?"
The POC and the last 60 days
- "How did the POC go? Who was it for, and what were they trying to prove?"
- "What made it extreme pressure? Was it scope, the timeline, or the team?"
- "What happens next with them: production, a paid pilot, or procurement?"
Pick 3–4 threads, follow her energy
- Accuracy and proof: "When a carrier asks how you know the mapping is right, what do you show them? How do you measure that 98%?"
- New wordings: "What happens when a carrier's wording doesn't fit the ontology? Who catches it, and how long does onboarding a new carrier take?"
- Change and drift: "When you change a model, a prompt or the ontology, how do you know nothing quietly broke?"
- Team: "How's engineering set up today? In-house, Jamie, contractors? What's the bottleneck?"
- Expansion: "US and Switzerland at the same time. What has to be true technically for that to work?"
- Agents: "You're positioning for AI agents too. Are customers asking for an API/agent integration yet?"
Play it back, then share one relevant thing
- "So if I've got it right: [her top 1–2 problems in her words]. Is that fair?"
- Only now, one short story: "We've been doing similar work for an insurance document platform: a 12-step extraction pipeline that went from roughly 70% to 90%+ correct, with quality reporting per layer so regressions show up before release." (Don't name the client.)
- "How are you thinking about solving [her problem]? Hire, fractional, partner?"
Land one concrete, small step
- Offer a choice, not a pitch: "Would it be useful if I [took a look at X / put together a short note on how we'd approach Y / jumped on a call with Jamie]?"
- Agree a date before you hang up. Respect the 30 minutes: finish on time, even if it's going well. Offer to continue another day.
The aim isn't to sell on this call. It's to leave with a reason for a second, scoped conversation, ideally with Jamie (CIO) in the room. That's where work gets closed.