The Intelligence Suite works differently.
Nobody submits a risk. The universe is already graded.
A whole class of subject, enumerated and assessed before anybody asks.
A register counts what suits a registrar; a policy attaches to something else. The collapse onto the real subject is stated rather than assumed, and everything that survives it is graded and priced before anybody asks.
Start with everything that's published.
Not a prospect list and not a vendor extract. The record for a class of subject, all of it, dated, and keyed to an identifier anybody can follow back.
Read the code. Never read its description.
Every register has a field whose name says one thing and whose contents say another. A population screened on an unresolved code looks perfect, joins cleanly, and is wrong in the one way nothing downstream can catch.
Collapse it onto what you actually cover.
A register counts what suits a registrar. A policy attaches to something else, one insured, one deductible, one renewal date. The join between the two is stated, not assumed.
Then say what interests you, and only then.
Written once against a published file, and reproducible by anybody who disagrees with it. That is the difference between a market and a pipeline.
Every one of them has a price before anybody asks.
Read, graded, priced and sortable, because the model can tell them apart. That is what a universe is for: not a longer list, an answered one.
Hover any mark. Real subject, real composite, real premium.
Generated from four kinds of signal, across every domain of the risk.
Public, joined, proprietary, vendor, and the breadth is what replaces the form.
Published files, joins nobody has taken, what the risk reports about itself, and what a specialist sells. Held on every subject in a class, each on its own clock — and none of it requested.
A statement of values is a handful of self-reported cells.
Year built, construction, protection, the schedule, the loss runs. Typed by somebody with an interest in the answer, checked by nobody, and the entire basis of traditional risk assessment.
Everything that matters is already published.
Assets, location, exposure, registrars, assessors, hazard agencies, inspectorates, courts. Dated files, keyed to identifiers anybody can follow back.
It is the floor rather than the edge, every competitor can read the identical file. A market that cannot reliably read the record has no business claiming to read anything harder.
The facts nobody publishes, because they live between two files.
A regulator publishes its open findings. It separately publishes the work authorised against them, and on each finding, the reference of the work that answers it. So the delay between a problem being recorded and anybody acting on it is the publisher's own join, and it has never been taken.
The same shape gives the second reading of every fact: as it stands, and as it has behaved. That one exists only if somebody was already holding last year's snapshot.
The signals a risk emits itself.
Security cameras, water-leak detectors, HVAC telemetry, IoT, OEM telematics etc. Anything that produces a continuous, verifiable, behaviour‑level signal can become a proprietary input. Crucially a risk can be priced in absence. So, the inclusion of proprietary signals doesn't punish those without, it simply allows those who do to have a more accurate composite score, tier and technical price.
The vendors who specialise in a given domain.
Geospatial, behavioural, financial, operational, and environmental datasets that cannot be collected at the same fidelity via public sources.
Each provides greater insight into a risk to provide an ever more nuanced assessment.
The asset, the ground, the governance, the conduct, the money, the security, the people, the counterparties.
No single one of these is new. What is new is holding all of them, on every subject in a class, before anybody asks — and being able to say which of them moved this week.
The board is a sample and not the library: every register a jurisdiction publishes is another column, and the joins between them are another again.
What we saw outranks what they printed.
Breadth is worth nothing if everything on the board is believed equally. Each reading carries the grade of the evidence under it, and a direct observation ranks above a structured attestation — because a published file is a claim and a reading is a measurement.
The third rung is the one that stings. A self-reported answer — the entire input of a traditional submission — sits mid-ladder.
A next-generation model that learns, evolves, and explains itself.
One engine, open maths, and no second path to the number.
A signal nobody could read is removed rather than scored zero, and its weight redistributes across the ones that answered. Scores reach money through the relativities and nowhere else, so any price decomposes back to its reading.
Twenty readings, each carrying the grade it earned.
Not twenty fields on a form. Twenty separate questions answered against separate sources, each stamped with how good the evidence behind it was.
Weighted into groups, and a group knows what it is.
What can hurt it. What it is. How it is run. Keeping those apart is what lets a model say why a price moved.
An unread signal is removed, not scored at zero.
Drop a reading. Its weight does not vanish, it redistributes across the readings that did answer.
This is the quiet failure in every model of this kind. A rubric can aim a quarter of its weight at questions returning the same number for every subject, and nothing anywhere fails. A weight is only a judgement if the thing it weighs can move.
Scores reach money one way, and only one way.
Through the relativities and nowhere else. No load bolted on the side, no second engine, no table maintained by hand, so you can always ask which reading moved the price, and there is always an answer.
The curve is fitted to outcomes, not to an opinion.
Realised losses land on the measurements that preceded them and the fit moves. Which only works if something was reading before the loss.
An edge protected by its own structure, delivering more than was ever possible.
Not a feature anybody can buy, because the readings behind it cannot be bought back.
An edge is cost or knowledge, and there is no third one. Thirty years of competition have very nearly closed the first; the second needs readings taken before anybody thought to ask for them.
An edge is either cost or knowledge. There is no third one.
You can process a risk for less than anybody else, or you can know something about it that nobody else has priced. Every durable advantage in this market is one of those two.
The market competes on cost, because the files are shared.
The same pack, the same broker, the same questionnaire, read by everyone at once, so the only variable left is what it costs you to read it. Roughly a quarter of premium is expense rather than loss, and about ten points separate the most efficient from the average.
That gap is the margin, and it has very nearly been competed away.
A moat is a claim, so it should be checkable.
Origin, lift, the loop, time, resistance to synthesis. Each one is a question with an answer you can go and look at, rather than an adjective.
Most claimed moats fail on time.
A rival can copy the reader tomorrow. It cannot buy the readings it did not take yesterday — and the strongest signals are differences between dated snapshots, so the second one is worthless without the first.
Knowledge is power. Written as appetite, it is profit.
State your read once, against any signal, and the book arrives ranked to it.
A criterion names a signal, a direction and a threshold — so appetite can finally be about something no submission form has a box for. And most of what it holds back is silence rather than a finding, which is what prices the next source.
Not "this class, this size". This subject's conduct.
A criterion names a signal, a direction and a threshold. Which means your appetite can finally be about something no submission form has a box for.
An agent's appetite ranks. A provider's binds.
One record, two enforcement semantics. On a broker's desk it sorts the worklist. On a capital provider's it is a rule at the match.
And a rule never fires on a reading nobody holds, which is why the next figure matters as much as the first.
Most of a young book is held by silence, not by a finding.
Move the third slider. Coral is what fails because nothing answered, not because something did.
That number prices the next source before anybody buys it. A traditional file cannot produce it, because it cannot tell an absence from a zero.
Your read of a shared universe is the part that is yours.
The assessment is the same for everyone on purpose, a market cannot run on a number each side computes privately. What is yours is the threshold you set against it, and the fact that nobody had to produce the reading underneath.
A graded risk arrives at the venue, not at a list of names.
It comes already assessed, with its evidence book attached. So the question is not "what is this?" — it is "does this clear the floor I already published?".
Capital states its shape in advance, once.
A signal, a threshold, an evidence floor, a capacity. Standing, no expiry, no correspondence, and nobody asked to read something they were never going to write.
Admission is decided on the ladder, not on the wording.
A floor is satisfied by a rung's position, not its name. Reorder the ladder and every floor anybody wrote as a word silently moves with it, which is exactly what happened here, and referred a whole book until it was caught.
Bound means every layer signed. Not a slip that looks full.
Two routes once held two separately-maintained rules for it, and hundreds of risks read bound to the client and in placement to the producer.
Identity and floors collapse before they reach a desk.
An agent-facing route may not hold a refusal at all. A neutral venue is not a promise in a contract — it is an absence in a response.
Linking risk features to loss unlocks inference nobody could run before.
The outcome returns to the readings that preceded it, and the fit moves.
A realised loss lands on the dated readings that came before it, never on today’s, and the fit moves — so every subject carrying that reading re-prices, not only the one that had the loss. Nothing a participant tells us can reach it.
A curve is a hypothesis with a date on it.
Fitting frequency and severity is not the interesting part; everybody does that. Having a dated reading against every subject before the outcome is.
Analysed against the evolution of signals over its lifespan.
Not to today's readings, to the evolution of those readings in the period before loss. A stale, stable feature gives a beautiful, monotonic, worthless result. A living signal, that changes provides inference and analysis on what actually occurred.
The relativities re-tune, and the whole book re-prices.
Every subject carrying that reading moves, not only the one that had the loss. An outcome on one risk is information about every other risk that shares it.
And nothing a participant tells us can reach it.
A relayed fact moves its own client's price and stops. A model that learns from what participants tell it eventually learns what participants want it to believe.
Every loop improves the next, and deepens the data moat.
The six stations close, and the rail itself becomes the circuit.
Six stations, and the rail itself becomes the circuit. Every loop leaves the record deeper than it found it — which is the part a late starter cannot buy back.
A market without waste is a profitable market.
The circuit opens out flat, and what it has been producing hangs beneath it.
The assessment is produced once, for everybody.
An assessment is never owned or forked, because the destination is a neutral market. Four houses looking at one risk stop being four desks producing the same number four times.
produced once, and read by everyone who is owed it
You see the risks inside your appetite. Only those.
Stated once, evaluated against the whole universe, so the book arrives ranked. Nobody sends you a submission you were never going to write, or asks you to read one to find that out.
a threshold on a signal — not a class and a limit
Every price decomposes, so every loss explains.
When something goes wrong you can see which reading was held, when it was taken and where it came from, rather than reconstructing a judgement from a file note.
every reading carries its source, its date and its citation
Ask the universe a question.
Per signal, per subject, per date, so a targeted portfolio is a query rather than a project. And no count is ever printed for a query nobody has run.
a portfolio is a query, and the answer is already standing
Tell us what you write.
We will come back with a book, graded, priced and triaged to your appetite before a single submission exists.
Three lines, and a person replies.
Tell us who you are and what you would want to see. We will come back with a time and a book to look at — there is no sequence of emails behind this.