Intelligence no longer has limits.
A new world is emerging.
Frontière serves the leaders it will test.
Whether the decision commits capital, a company or a name: Frontière prepares it with the most advanced AI capabilities — the complete case file, the options priced, the useful contradiction. The discernment that concludes remains human — and it signs.
Decisions of capital · Decisions of consequence · Marseille — France & international
The frontier of intelligence, in the service of those who lead.
In the end, you are the one who signs.
You have been briefed, advised, reassured. Then everyone goes home — and what remains is you, the decision, and what you don't yet know.
- If this fails, what should I have known?
- Who around this table has nothing to gain by saying yes to me?
- What I think I know about these people — who told me?
- Am I deciding — or merely ratifying?
- At what point can I no longer turn back?
— the eleven p.m. questions
No one around you is in a position to answer without an interest of their own: whoever championed the deal has already defended it, whoever sits on the committee has a candidate, whoever advises you is selling what comes next. And no one has the time to take the whole file apart again. That is precisely what we do: the entire case — investigated, priced, contradicted — delivered before the deadline, by an eye that has nothing else to sell.
Intelligence has become abundant. Judgment, rare.
A general revolution is under way — closer in nature to electricity than to a tool. Frontier AI models now produce, in minutes and at near-zero marginal cost, cognitive work at senior-analyst level: analysis, synthesis, modelling, monitoring. What sold yesterday in man-days is produced today in machine-hours — and the hierarchy of value inverts.
- Dataabundant — it overflows.
- Analysiscommoditised — machines produce it.
- Judgmentrare — it is exercised, and it commits.
- Responsibilityinalienable — it is signed.
A world where analysis is free is a world where judgment is the only differentiator left. The leaders who win this world will be those who arm themselves with analytical systems augmented by frontier AI — and governed by human judgment that concludes and signs. Frontière builds those systems, and puts them in service of one thing only: your decisions of consequence.
The frontier beats the craftsman; governed judgment beats the software.
A system of intelligence, designed for your decision.
An entire case file, in days, on a singular decision: that cannot be bought off the shelf. It has to be designed. For each engagement we assemble a chain of agents dedicated to your question — which sources, in what order, under which grids, with which controls. This craft has a name: AI design. It is ours, and it is what makes everything else possible.
The agentic system.
Tesuji, our engine: specialised agents built on Claude, Anthropic's frontier model — each with its task: read the documents, cross-check, hunt for the contradiction, verify the previous agent's work. We design this chain for your decision, never for general use. That is the difference between asking an AI a question and having a case investigated.
The matrices.
Power without a grid produces volume. Goban is our repertoire of proprietary conceptual matrices: for each family of decision, the grid that says where to look, in what order, and what counts as proof. It is what lets Tesuji approach every case with the exact frame it calls for — and hunt there for the tesuji, the move that reverses the dynamic. The first of these matrices reads a leadership team as a system: six dimensions, two loops, one fulcrum.
The partners.
You never talk to a machine: you talk to a partner — from investment banking, strategy consulting, executive search — who frames the question, challenges the evidence the chain brings back, settles what it leaves open, and puts their name to it. A committee of senior advisors — experienced executives and investors — brings a second eye to every reading. The interface of the whole system is human: that is what you are buying.
A partner requalifies the question, settles which documents must be gathered and sets the controls the chain will have to pass.
Accounts and filings, articles and shareholder agreements, board minutes, key contracts, registers, press, transcribed interviews, inventories. In full: nothing is summarised before it has been read.
One agent is tasked with attacking the previous agent's conclusion. Every figure traces back to its document, its date and its formula; what withstands is retained.
The partner arbitrates, settles what the chain leaves open, marks what remains unestablished — and signs.
The volume a team of analysts takes weeks to absorb passes in hours — and the time saved goes back into judgment, never into margin.
The machine reads everything. The model knows what to look for. The partner decides — and signs.
The red line, contractual: the machine never assesses a person · no decision is automated · consent first, individual debrief after · every conclusion is signed
At the frontier of intelligence, judgment remains human — committed, signed, tested.
Decisions of capital. Decisions of consequence.
For us, preparing a decision means: every document read, every cross-check made, the options priced through to their consequences, the contradiction raised before the decision is taken — and a recommendation that a name commits to. We prepare the decision; you make it. Two doors, depending on your seat.
An intelligence does not decide where to look.
Our matrices do. That is where what we know lives — and it is what makes two engagements comparable.
Give the same leadership team to ten brilliant analysts: you will get ten different readings. Not because they are wrong — because they do not look in the same places, in the same order, or with the same idea of what counts as proof. What sets a seasoned practitioner apart is not what they know on top: it is what they think to look at.
An artificial intelligence has the opposite, symmetrical flaw: it will always answer, and answer well, the question it is asked. Never the one you forgot to ask.
A matrix is what decides where you look. A fixed number of entries, an order, and for each: the question that commands it, what the case file must bring back, what counts as proof, and what that cell does not allow you to conclude. It is not a template for answers: it is a discipline of the question.
It is also what governs the engine. Our matrices do not decorate the method — they frame the agent chain cell by cell: what it must look for, what it must bring back, and the moment it must say it does not know. A machine without a matrix produces the plausible. Under a matrix, it produces the investigated: traceable to the source, and comparable from one team to the next, one portfolio company to the next.
Goban is the repertoire where they all live. The first is complete and tested: human capital assessment, which reads a leadership team as a system. The others are being written, decision by decision — governance, the candidacy, succession, the sale of a business, post-acquisition integration. Each engagement tests the grid, corrects it, and what it learns serves the next. That is where our capital accumulates — not in the reports delivered.
And the matrix itself carries the red line: each cell states where it stops. It reads a system, never a person — and what it cannot conclude, it writes down.
Answers the question it was asked. Produces the plausible, fast. Two case files cannot be compared: they were not examined in the same places.
Investigates every cell, including the ones no one would have thought to open. Every conclusion carries its cell, its proof and its limit.
Are the decisions that matter taken in the right place, by the right people, at the right tempo?
The real path of a recent decision, the observed lead times, the places where arbitration escalates when it should not.
A decision traced and dated, two concurring interviews, a measurable gap between what is written and what is done.
The worth of a leader. It describes an allocation, not a person — and says so.
One cell of six, in the human capital assessment matrix. The full repertoire unfolds on the Method page. Read the method →
What you receive is not a document.
It is an address — and it stays alive until the decision is made.
You know the scene. The report arrives at the end: eighty pages, two months after the question, three actual readers. The decision, meanwhile, was made along the way — one Tuesday morning, on whatever each person had in mind.
We deliver differently. Every engagement opens with a Kifu: one link, one password, and the entire file. Thirty seconds tell you where the work stands, what remains to be settled and which deadline commands. Each piece opens to the depth you choose: ten seconds, one minute, or everything — and every figure carries its source.
It opens on a phone, between two meetings. It does not close at delivery: it moves between your visits, and it keeps the record. Each circle of readers sees only what concerns them — the one who commissioned, those who decide, those who contributed.
In the game of Go, the kifu is the score of the game: every move, recorded and dated. The Goban reads, the Tesuji plays, the Kifu records.
Yours would open on day one of the engagement.
Composite case — company and names are fictitious · no password required
Alice Arnaud — Founder
Photographie : Andrea Graziosi — Studio Voo · Sky Center, Tour La Marseillaise
One partner carries your case, end to end.
No junior works on your file: there are none. A senior partner frames the engagement, conducts the interviews, arbitrates the reading and signs the conclusion — the chain carries the volume, the partner carries the judgment. A committee of senior advisors, experienced executives and investors, brings a second eye to every reading.
Devoted to intelligence — in all its forms.
The firm was founded by Alice Arnaud, who has spent her life gathering its forms. The intelligence of markets: M&A at Goldman Sachs in London. Of hypergrowth: Rocket Internet. Of organisations: the Data and AI transformation of a major mutual insurance group, up to its executive committee. Of people: years of executive search and leadership advisory. Of machines: Frontière. She thinks in systems — which is where organisations, and decisions, become legible.
Read the full biography →The questions we are asked.
Thirty minutes on
a real situation.
Bring a case — a portfolio company, a succession, a build-up, a leadership team, a decision that fits no box — and leave with a reading. It is the best way to judge the method.
Reply within 48 working hours, two slots proposed · Confidential by default · References available on request, under contractual anonymisation
“The Blind Spot” — human capital, due diligence and value creation.
The white paper examines the blind spot of due diligence — and opens the method: the Goban model, triangulation, the course of an engagement. Send your business email; we send you the PDF. No mailing list.