The marine insurance market is running on vibes

Right now, as I write this, 369 navigational warnings are active across the world's oceans. The Bosphorus is flagged as elevated with a disruption score of 30. Malacca Strait is disrupted at 50. Suez Canal sits at elevated, also 30. These numbers change every few hours. Sometimes faster.

The Joint War Committee — the body whose listed areas directly determine war risk premiums for most of the London market — meets quarterly. Their last circular is weeks old. Their next one is weeks away.

I'm not saying the JWC gets it wrong. They're experienced underwriters making considered judgments. But there's a structural problem when the data surface updates every 15 minutes and the pricing framework updates four times a year.

---

How war risk actually gets priced

For anyone outside the marine insurance market, here's the short version. The JWC publishes a list of areas where vessels face elevated risk — war, piracy, terrorism, related perils. If a vessel transits a JWC listed area, additional premium is typically required. The list drives billions in premium flows.

The process looks roughly like this:

An underwriter evaluating a voyage through the Gulf of Aden today has the JWC listing (which confirms it's a listed area — useful, but binary), an IMB piracy map that may be a few days old, and whatever they remember from the last NGA advisory they read. If they're good — and many are — they also have 20 years of pattern recognition telling them which routes, seasons, and vessel types carry more exposure.

That pattern recognition is real. It has value. But it's not auditable, it doesn't scale, and it walks out the door when someone retires.

The best underwriters in London are essentially running a neural network in their heads. The problem is that neural network has no API, no version history, and a retirement date.

What "vibes" looks like in practice

I've watched underwriters assess voyage risk. The workflow typically goes:

1. Check if the route passes through a JWC listed area (yes/no)
2. Look up recent piracy reports in the region (if they remember to)
3. Glance at weather (maybe)
4. Consider the vessel's flag and age (from the slip)
5. Apply a premium rate based on "what we charged last time for something similar"

Step 5 is where the vibes live. "What we charged last time" is a heuristic, not an assessment. It works when conditions are stable. It fails badly when they shift — and in 2025-2026, conditions have shifted dramatically. Red Sea diversions. Houthi attacks on commercial shipping. Record piracy in the Singapore Strait. New sanctions regimes every quarter.

That screenshot is from ArcNautical — the platform I've been building. One query. One route. Ten intelligence signals scored against the actual path the vessel would take through the water. Risk score 50, ELEVATED. JWC listed area exposure: 100 out of 100. Distance: 2,222 nautical miles. Fuel burn: 368.2 metric tons of VLSFO. CII rating: C.

An underwriter seeing this doesn't need to remember whether the Gulf of Aden is listed. They can see that it maxes out the JWC signal, then look at what else contributes — piracy incidents, conflict intensity from GDELT event data, sanctions exposure, country instability indices for Iran, Yemen, and Somalia.

The latency problem

Data staleness is the quiet risk in marine underwriting. Not because the data doesn't exist — it does, in real time — but because the workflow doesn't surface it.

Signal	Raw data cadence	Typical underwriter access	Latency gap
NGA navigational warnings	Continuous (15-min batches)	Checked weekly, if at all	Days
IMB piracy incidents	Daily	Monthly ICC report	Weeks
GDELT conflict events	Every 15 minutes	Not used	Never seen
JWC listed areas	Quarterly	At publication	Low
OFAC/UN sanctions	Weekly	At binding (screening tool)	Days
Weather / sea state	6-hourly	Rarely checked for risk	Variable
AIS vessel positions	Real-time	Tracked for operations, not underwriting	Disconnected

Look at the GDELT row. The GDELT Event Database processes global news and categorizes conflict events every 15 minutes. It can tell you whether the political temperature around a transit country spiked this morning. Almost no underwriting desk uses it. Not because it's irrelevant — a spike in Goldstein-scale conflict events around a country directly correlates with maritime risk in its waters — but because nobody has built it into their workflow.

The "good enough" trap

Marine insurance has operated this way for centuries. Lloyd's literally started in a coffee house. And the market's loss ratios have been... fine. Not great, not catastrophic. Fine.

That "fine" is the trap. When your combined ratio hovers around 95-100% and you're writing enough premium volume, there's no burning platform for change. The six-figure maritime intelligence subscriptions that do exist are bought more for due diligence optics than for changing actual pricing decisions.

But here's what's shifting:

The velocity of change has outpaced the update frequency of the tools. A JWC listing that was accurate in January might not reflect March reality. An underwriter's mental model from 2024 might misjudge a 2026 transit.

What would "not vibes" look like

I'm not arguing that institutional knowledge has no value. It does. A senior underwriter who's seen three piracy cycles knows things that no dataset captures — the political dynamics, the negotiation patterns, the seasonal variations in attack frequency.

But that knowledge should be an overlay on data, not a substitute for it. Here's the difference:

Aspect	Current state	Data-driven state
Route risk	"It goes through a listed area"	Composite score 50/100, 6 of 10 signals active, JWC exposure 100, piracy 35, conflict 28
ETA confidence	"About 7 days"	P10: 5.8d, P50: 6.6d, P90: 7.4d (from 500 Monte Carlo simulations)
Vessel quality	"She's a decent ship, classed with DNV"	Vetting score 72: 0 PSC detentions, flag grey-list, age 12yr, ownership opacity 45
Premium rationale	"Similar to what we charged for the Aden transit last month"	Risk delta +12 vs. last scored voyage on same route, driven by piracy signal increase
Audit trail	Email chain + spreadsheet	Timestamped score with signal breakdown, route geometry, data sources, confidence level

The second column isn't a fantasy. It's what our platform produces right now for a Fujairah-to-Djibouti query — risk score 50, 2,222nm, 368.2 MT fuel, CII rating C, with every signal broken down and every data source timestamped.

The uncomfortable part

I should be honest about something. Building a platform that produces auditable, real-time risk scores is implicitly an argument that the existing way of working isn't good enough. That's an uncomfortable thing to say to a market full of experienced professionals who've been doing this successfully for decades.

So I'll frame it differently: the existing way of working was excellent for the environment it was designed for. Stable routes, slow-moving threats, quarterly changes. The environment changed. The tools didn't.

The argument isn't "trust our numbers blindly." It's "your underwriters should have these numbers in front of them when they apply their judgment." The institutional knowledge doesn't go away. It gets something to work with.

---

I know this piece will read as self-serving to some. I built a maritime intelligence platform, and here I am arguing that the market needs maritime intelligence platforms. Fair criticism.

But I'll ask this genuinely: if you're underwriting war risk today, what does your workflow actually look like? Are you working from the last JWC circular and your own experience, or have you found data sources and tools that close the latency gap? I'd like to hear from people who've solved this problem differently than I'm imagining — or who think the problem doesn't exist.