A year ago this month we were three people in a co-working space in Mouraria trying to explain to a Portuguese reinsurer why a small startup with no book of business should be able to white-label their actuarial tables and issue policies in three minutes. The meeting lasted forty minutes. They said they would think about it. We never heard back.
That story captures most of what year one taught us about building insurance in Portugal: the product assumptions we brought in from studying UK and German insurtechs turned out to be less transferable than we expected. The market is not behind — it is just structured differently, and we had to re-learn several things from scratch.
This is a candid account of what surprised us, what broke, what we changed, and what we still have not figured out.
What We Got Wrong About the Distribution Model
We built v1 on the assumption that direct-to-consumer would be straightforwardly appealing to Portuguese buyers. Buy online, skip the broker, save 15-25% on commission costs. That logic is not wrong. But we underestimated how embedded the corretor (broker) relationship is in this market.
Insurance brokers in Portugal are not just distribution channels — for many households, they are the primary financial advisor, claims shepherd, and trusted contact for anything involving contracts. When your Renault Clio gets rear-ended on the IC2, you call your corretor, not an insurance company hotline. The personal relationship has decades of accumulated trust behind it.
Our first cohort of users were almost entirely under 35. Not because we targeted that demographic deliberately, but because people with established broker relationships did not see a reason to switch. We spent the first four months trying to address this, and the honest answer is: we adjusted our messaging rather than solved the problem. We stopped arguing that brokers add no value and started focusing on the specific case where they genuinely do not — renewals. A broker's value at renewal is close to zero, and that is the moment we can compete.
The v1 Pricing Model: What It Got Wrong
We launched with six input variables for auto pricing. Our actuarial model, which Marta spent six months building and calibrating against Fundo de Garantia Automóvel claim data and our partner insurer's historical book, was technically sound. What we got wrong was the interface.
We asked users to self-report their annual mileage. The data we got back was fiction. Portuguese drivers systematically underreported by 30-40% — not necessarily through intent, but because most people genuinely do not know how far they drive in a year. We were pricing on stated 8,000 km/year and getting customers who were actually driving 14,000 km/year. Our loss ratios in the first two quarters were worse than we had modelled, and this was the primary reason.
We fixed this in v2 by replacing the mileage question with a trip frequency question (how many days per week do you typically drive?) cross-referenced with vehicle age from the Registo Automóvel and average Portuguese commute distances by postal code. It is a less precise input individually, but the aggregate prediction turned out to be more accurate because it does not rely on people knowing something they do not know.
Regulatory Surprises: ASF Was Not the Problem We Expected
Before we launched, we were most worried about the ASF approval process. We had read accounts from insurtechs in other markets about multi-year licensing fights with regulators. Our experience was different.
The ASF's framework for intermediários de seguros (insurance intermediaries) under Decreto-Lei n.º 144/2006 is detailed but navigable. The genuinely hard part was not the ASF — it was the incumbent insurance infrastructure. The SGOA (Sistema de Gestão de Ocorrências de Acidentes) data exchange, the ISP legacy systems, the lack of standardised APIs from incumbent insurers. Portugal's insurance back-end runs on interfaces designed in the 1990s, and every integration we needed was custom work on our side.
We spent roughly four months on integrations that we had budgeted six weeks for. That delta came entirely from undocumented data formats and APIs that behaved differently in production than in their technical specifications. We are not naming the parties involved — they are partners and the problems were not bad faith — but anyone building on top of Portuguese insurance infrastructure should budget significantly more integration time than they think they need.
The Trust Signal We Did Not Anticipate Needing
In January 2024, a user in our beta group tried to use their Indie policy document during a GNR stop on the A1. The officer accepted it. The user posted about it in a Portuguese finance Facebook group. That post drove more new registrations in the following two days than all our paid acquisition in the previous six weeks combined.
The lesson: in insurance, the most valuable marketing moment is not acquisition, it is a real-world validation scenario that proves your product works when something goes wrong. We had been spending almost nothing on social proof and too much on SEO. We redirected resources accordingly — into claims handling speed, into having a real human available by phone within two business hours, into the on-demand verification QR code that the GNR officer used in that story. Word of mouth in insurance is overwhelmingly driven by claim and verification experiences, not by pricing or UI.
What the Numbers Looked Like After Twelve Months
We are a small team and we are not disclosing revenue figures here. But we can share operational metrics that describe the shape of the business:
- Average time from first quote to bound policy: 4 minutes 12 seconds. Our target was three minutes; we are still working on this.
- Policies that renewed at the first renewal anniversary: roughly 71%. We expected around 65% for a new digital insurer with no established trust history, so this exceeded our model.
- Claims where we hit our 48-hour acknowledgement target: 89%. The 11% where we missed were all multi-party auto accidents requiring coordination with other insurers.
- Customers who came through organic or word-of-mouth channels (not paid): 58% by Q4 2024, up from 34% in Q1.
The renewal rate is the number we watch most carefully. It tells us whether the pricing was fair, whether the experience was good enough, and whether the digital-only model builds trust over time. So far, the trend is positive.
What We Still Have Not Figured Out
Health insurance is harder than auto. Not technically — the pricing model is more complex, but that is just actuarial work. The hard part is what health coverage actually means to someone. Auto insurance is mostly about property and legal liability. Health insurance is personal in a way that makes customers much more anxious about exclusions, waiting periods, and network coverage. Our health product is functional, but the communication around it is still not as clear as it needs to be. People buy it and then call us to ask questions that should be answered by the policy document. That gap is our problem, not theirs.
We also do not have a good answer yet for customers with complex risk profiles — classic cars, high-value homes, business vehicle fleets. Our model is built for standard risks. Anything outside a narrow band either gets declined or gets a price that is uncompetitive because our uncertainty is baked in as margin. We are working on this, but it requires more partner capacity than we currently have.
What Year Two Looks Like
We are not making product announcements here, but the direction is clear: better claims communication, a revised health product that does a better job of explaining what it covers before purchase, and an improved renewal flow that proactively tells customers whether their price has changed and why.
The Portuguese market has room for what we are building. The incumbent insurers are not bad at claims — they are slow at everything else. Digital quoting, instant binding, transparent pricing, and a customer portal that actually works are not ambitious goals. They are table stakes that the market has been slow to demand. That is changing, and our job is to be ready when it does.
Year one was harder and slower than we expected, and we made enough mistakes that we are glad we did not try to scale before we had a product we were proud of. The shape of what we built is right. The details needed — and still need — a lot of work.