I've had some version of this conversation with a dozen underwriters and portfolio managers over the past year. Ask any of them if they have enough data, and the answer is almost always yes. Ask if they trust that data enough to make a fast decision on it, and the answer gets a lot more complicated.
That gap is the real story in P&C right now. Not a data shortage, a complexity problem. NOAA's National Centers for Environmental Information counted 403 separate weather and climate disasters in the US since 1980 that reached at least a billion dollars in damage, 27 of them in 2024 alone. Every one of those events throws off another layer of peril data, sensor data, and model output, and connecting all of it into one coherent answer, fast enough to act on, is a harder problem than most organizations are set up to solve.
The problem isn't what most people think it is
Ten years ago, the constraint was access. Carriers didn't have parcel-level wildfire scores. They didn't have satellite imagery refreshed weekly. They didn't have flood models that ran below the county level. That constraint is gone. What replaced it is a volume problem. Underwriters and portfolio managers are now working with a dozen data sources at once, none of them built to talk to each other, and making sense of all of it inside a submission deadline or a renewal cycle takes real technical work that most teams don't have the time or tooling to do consistently.
Picture the actual moment. A submission comes in, and the underwriter is looking at a wildfire score from one vendor, a flood layer from another, and a client-submitted Statement of Values that may or may not reflect what's actually built on the parcel today. Nobody failed to assign ownership here. The honest problem is that fusing a dozen disconnected sources into one defensible answer, fast enough to hit a quote deadline, is difficult work that most teams aren't equipped to do by hand at the volume the business now requires.
That's the actual problem. Fragmentation and volume, not scarcity, and not a missing job description. It shows up differently depending on who's sitting in the seat. A CUO pricing per-risk at the point of quote and bind needs a fast, defensible answer on one property. A CRO managing portfolio-level aggregation and capital modeling needs the same underlying data rolled up across thousands of properties without the fusion work breaking down at scale. Right now, most carriers are solving both problems with different tools, different vendors, and different versions of what should be the same underlying record.
What fragmentation actually drives
Watch what this does inside an underwriting or portfolio team, and the pattern repeats itself almost exactly the same way from carrier to carrier.
Underwriters spend real hours reconciling conflicting outputs from different vendor models before they can even price a submission, and that time tax compounds across a book of business. Portfolio managers run concentration checks at fixed intervals—a quarterly review, maybe a renewal cycle, instead of continuously, which means accumulation risk in a coastal zone or a wildfire corridor can build for months before anyone sees it move. That risk is not evenly distributed either. Severe convective storms are the single most frequent peril behind US billion-dollar disasters since 1980, 203 events by NOAA's count, more than tropical cyclones and flooding combined, which makes continuous concentration monitoring a bigger problem than most portfolio reviews are built to handle. Two analysts on the same team can land on different underwriting theses for a similar risk simply because they pulled from different sources, or weighted the same inputs differently, and there's no shared record forcing agreement.
Capital gets deployed against a picture of exposure that looked complete at bind and wasn't. That gap rarely shows up right away. It shows up in loss ratio a year or two later. It shows in mispriced AAL and PML once a loss event tests the assumptions behind them, and the numbers behind that test are not small. Aon's data, compiled by the Insurance Information Institute, puts insured losses from US natural catastrophes at $79.6 billion in 2023 alone, across 89 separate events, and tropical cyclones carry the highest average cost per event of any peril NOAA tracks at $23.0 billion. That's the scale of assumption a fragmented data process is quietly resting on. It shows up when a treaty renewal conversation turns into a debate about how TIV was actually calculated, and nobody in the room can point to one number everyone agrees on.
None of these are hypothetical. It's the operating reality for a lot of underwriting desks right now, and it's the reason “we have plenty of data” and “we can move fast with confidence” have become two very different claims.
The basic fix, and where most attempts at it fall short
The fix isn't more data. It's a single source of truth.
That phrase gets thrown around loosely, so it's worth being specific about what it actually has to mean here. A single source of truth is one parcel-level record that every team in the organization, underwriting, portfolio management, actuarial, capital markets, is looking at and trusting at the same moment. Not five dashboards that each claim to be authoritative. One record, with the fusion work already done, that still holds up when a reinsurer or a regulator asks where a number came from.
Building that requires a few things most point solutions don't actually do. Real data fusion, meaning satellite imagery, environmental sensors, third-party models, and proprietary datasets combined into one coherent parcel-level view, rather than one more layer sitting next to the layers you already have. Deterministic logic underneath whatever AI is doing the heavy lifting, because a probabilistic guess isn't defensible when it feeds into a number a regulator or capital provider is going to test. And the output has to live inside the actual workflow, at the point of quote and bind, inside the portfolio review, not in a report that gets opened once and forgotten.
Skip any one of those three and what you've built is another data source. You haven't solved fragmentation. You've added to it.
How Prometheus actually delivers this
This is where our approach diverges, and it's the reason we built Prometheus the way we did.
Go back to the three requirements a real single source of truth has to meet. Actual fusion across sources, not another layer sitting next to the ones you already have. Deterministic logic underneath the AI, so the number holds up when someone tests it. And the answer living inside the actual workflow, not in a report nobody opens. Most tools in this market solve for one of those, occasionally two. Prometheus is built to hold all three at once, and that's the difference between another data feed and an actual single source of truth.
Prometheus isn't another model bolted onto the stack, and it isn't another peril layer competing with the five you already license. It's the fusion layer underneath all of it, pulling satellite imagery, environmental sensor data, proprietary datasets, and third-party models into one parcel-level record that functions as the single source of truth for the account. Fulcrum handles the cat modeling and concentration analysis on top of that record. Pyre handles wildfire-specific scoring at the parcel. But the record underneath both is the same one, which is the entire point. Underwriting, portfolio management, and capital markets end up working from the same answer instead of reconciling three different ones every time a question comes up.
We saw this play out directly with Roosevelt Road Specialty, a Managing General Underwriter that came to us with what their team calls the SOV Problem. A client-submitted Statement of Values can look complete on paper and still hide under-declared value or outdated construction detail. Manually verifying it across a full portfolio used to eat entire days of an underwriting team's time. With Prometheus, Roosevelt Road can run a broad-brush pass across an entire portfolio and immediately see what needs a second look, turning reviews that used to take hours or days into something measured in minutes. Their CUO of Property, Brendan Cook, put it about as plainly as it can be put.
“The worst way to find out you have a bad risk is through a loss.”— Brendan Cook, CUO of Property at Roosevelt Road Specialty
That's the single source of truth actually doing its job. Not a nicer interface sitting on top of the same fragmented inputs. One record that holds up when it gets tested, whether the test is a treaty renewal, a regulator's question, or an actual loss event three years down the road.
The moat
Every carrier can eventually buy the same peril data. Almost none of them have one record that every team trusts enough to act on without re-checking it first. That's the actual moat, a single source of truth that holds up under pressure, and it's not something you assemble by adding another vendor contract. It takes the fusion work, the deterministic logic, and the workflow discipline all built together, which is the specific combination Prometheus was designed around from the start.
The data was never the hard part. Knowing what to do with it always was.
Marco Puebla, VP, Head of Revenue, Neural Earth