SCROLL — WATCH IT CONVERGE
PRÆ·VIDERE
SEE AHEAD

What is the probability that a mid-market cyber incident costs your firm €10 million this year?

An uncomfortable question — and a computable one. Praevidere is a swarm of AI statisticians that automates evidence gathering and risk modelling for insurance underwriting. Today such questions get answered by instinct or by a vendor's score. They deserve a number, with the working shown.

P·V§ I  THE PROBLEM

The future cannot be known. Only estimated.

Most business decisions are made on instinct because real statistics is slow, scarce, and expensive. So inventories miss, campaigns overrun, and plans meet a world that was never modeled. The numbers were computable. Nobody computed them.

P·V§ II  THE INSTRUMENT

The question decides the shape of the answer.

The question is fixed first: what would settle it, by when, and in what unit. The causes follow from there, drawn into one graph, and every node on it must reach the answer. Evidence folds onto that graph, records on one channel and studies on the other. One sampler draws it, and every figure you are shown is read from those same draws.

  1. 01 · Ratio

    Fit to the question

    A question arrives in the words of the business that asked it. We fix what would settle it: the quantity, the horizon that closes it, the unit it is counted in. Nothing downstream is allowed to drift from that definition.

    What would settle it · and when

  2. 02 · Causa

    One graph

    The causes are drafted into a single graph, from the question and the facts found for it rather than from domain habit. Every node carries its own distribution family and must reach the answer. Edges the evidence cannot settle keep a confidence class, so their doubt travels into the result instead of vanishing.

    One question · one graph

  3. 03 · Compositio

    Physical and implied

    Evidence enters as folds on the nodes it bears on, each keeping its source, its scale, and its year. A fact observed in the world is physical. A fact reasoned from studies and expert judgment is implied. A node resting on one channel alone is disclosed as such.

    Records and studies · both channels

  4. 04 · Mensura

    Honest uncertainty

    One sampler draws the graph, and every figure is read from those same draws. A mean that does not exist is refused rather than estimated. Where the graph cannot price a number, the run refuses and names what is missing.

    A number · or a refusal

SPECIMEN Nº 001ONE CLASS · ONE HORIZON · 8,000 DRAWS

For a mid-market US healthcare company (200–2,000 employees), the probability that 12-month losses from cyber incidents exceed $1 million is estimated at 5.6%.

Physical — incident records count how often. Implied — loss studies say how large. The year's total is events times cost. The pale green node is P(year > $1M).

The film prices one question. Frequency: 0.405 contained incidents a year, 1.73 severe. Severity: a median $89,000 for a contained incident, $1.85 million for a severe one. The year's total crosses $1 million in 448 of 8,000 draws — 5.6%, about one year in eighteen.

INCIDENT RECORDS · LOSS STUDIESEVERY FIGURE IS ILLUSTRATIVE · NOT A LIVE RATE
P·V§ III  THE TEAM

One vantage point.

“Life’s most important questions are, for the most part, nothing but problems of chance.”— LAPLACE

ARTIFICES

THE MINDS BEHIND THE TECHNOLOGY

We are a team of craftsmen: an applied scientist, a quant, and a jurist. Between us lie years at the frontier of artificial intelligence, studying how intelligent systems break, how they align to your business, and what mathematics shape them; years in asset management, pricing uncertainty for a living; and years spent leading the remodeling of ossified institutions from within, inspiring through exemplary conduct. Together, we insist on the highest standards because we often had to build them ourselves.

P·V§ IV  CORRESPONDENCE

See ahead with us.

For engagements, partnerships, or a first conversation.