The natural way to tell the story of our tool should be chronological. First the paper with our investigation. Then the product.
That is not how it really happened.
The method and the machine were born in parallel. Chicken and egg. Neither came first. Every insight came from trying to build the machine. Every constraint in the machine came from what the method demanded. And we started with our passion: Business Continuity Management.
And this article is about the second half — because the first two articles were about the first half, and the machine is where the whole thing had to land.
The idea did not come from a strategy meeting. It came from a psychology app that I saw on Instagram in 2025.
The app offered to let you talk with your clone. Type in, share, chat with a version of yourself. I tried it. It was interesting in the way those things are interesting — a novelty that says something real underneath.
I did not stay on the app. I stayed with the question.
Could I clone myself digitally? My brain, and the brains of my colleagues. Twenty years of BCM practice. Not the surface — the interior. The reflexes. The reading of the room. The knowledge of what a standard actually means when it meets an executive committee.
That was the seed. But when you want to do something like that, you have to respect the rules.
Because the truth is, “everybody loses their minds” (Joker voice – Heath Ledger) with LLMs. They are accessible. Public. Free at the entry point. It feels like the big AI manufacturers handed us a little toy to play with a piece of AI — without us really understanding what is behind it, and worse, without respecting the rules of AI. Or more accurately, the rules of LLMs.
Everyone is writing prompts. Everyone is calling that AI. It is not AI. It is a chat with a text engine that occasionally produces something useful and often produces something that sounds useful, which is worse.
You cannot build an AI BCM tool like that.
So we set constraints. Hard ones.
No hallucinations
In BCM, a hallucinated clause is a compliance failure. A hallucinated recovery strategy is a client whose plan does not work at three in the morning. Hallucination is not an amusing quirk. It is disqualifying.
Data protection
Client data is client data. No leakage. No storage on our servers. We don’t keep your data. You generate something, you don’t keep it? It’s on you.
No training on our data or client data
The AI you use to draft your plan should not become smarter by consuming what you typed into it. Ever. Full stop.
Guaranteed results
Not “good enough on average.” Not “solid in most cases.” The output has to hold, every time, or the tool does not deserve to be in a practitioner's hands.
Four constraints. Any one of them would knock out most of what the market is currently selling as “AI for professionals.”
Meeting all four took thousands of hours. It was hard. Not because the technology does not exist — because putting it together correctly demanded that we stop treating LLMs as the answer and start treating them as one component in a larger system. We tested everything. Build a tool from scratch. Build a SaaS on top of platforms.
The path we found to the BCM AI is not a prompt. It is not an LLM with instructions. It is something with a BCM brain: RAG with legit documents, vectorization, right instructions, categories to optimize the answer. We are at the first stages, it’s not an AI engine but it’s already working incredibly great.
For our BCM, we use our investigation and the Centaurization methodology. Then D/M/S told us where to point everything.
(D)eterministic tasks (100% AI) are useful. They save time. A compliance matrix that populates itself, a maturity assessment that computes from structured input — these are real gains. But they are not what makes a BCMer's day.
(S)trategic tasks (100% human) are sacred. The machine does not touch them. Not because it cannot try — because it should not. Arbitrating an unrealistic RTO in front of the executive committee is not a prompt problem. It is a career, a relationship, a moment of political judgment. No system should be in that room pretending to help.
(M)ixed is the reason the Centaur exists.
The mixed zone — the collaborative zone — is where the BCMer produces most of their working hours. Continuity plans. Risk treatments. Strategy analyses. BIA narratives. Audit responses. Long, structured, contextual documents that require expertise to build and expertise to sign. This is where a BCMer spends a lot of time and energy, unfortunately. And this is exactly where AI is the most seductively wrong on its own — where it can produce a beautiful draft that says nothing, or a coherent-sounding plan that would collapse under stress.
M is where the Centaur lives. Not to replace the practitioner. To augment. Iron Man is the right metaphor if you understand it: the suit does not fight the battle. Tony Stark does. The suit lets him do what he cannot do alone. And he has Jarvis.
The Centaur is the suit. The BCMer is the reason it exists.
I am not going to describe the app. The best way to know it is to use it.
But there is one rule I want to leave on the table, because it is not a marketing rule — it is a rule of the thing itself.
Garbage in, garbage out.
If you feed the Centaur bad data — imprecise, missing context, no scoping, a five-line brief where a complete file is needed — it will do what it can with what you gave it. It will not invent competence you did not provide. That is not a defect. It is a design choice. A tool that hides bad inputs behind confident outputs is exactly the kind of tool the Centaur is built to be the opposite of.
If you feed it quality — precise scope, real context, the details a practitioner would collect anyway — in ninety seconds it produces content that even a twenty-year BCM veteran could not match, either in substance or in speed.
That is not a claim. That is what the tool does when the rules are respected.
Here is why this matters.
In BCM, we have a support problem. We get little in the way of resources. Little money. Little headcount. Little organizational patience for a discipline that only proves its value when something has already gone wrong. So the BCMer's time gets consumed by what is measurable and reportable — documentation, plans, analyses, reports, policies, compliance evidence. The work that produces artifacts.
That is not what saves a company in a crisis.
What saves a company is culture. Interaction. Strategic thinking. Exercising the business until the response is a reflex instead of a document. Sitting with an IT director and understanding, without a form, what will actually work when the primary site is gone. Exercising with operatives on the field. Training people and creating preparedness.
Those are not the only things that count. But they are the things that get lost first, because the paperwork never stops.
The Centaur is built so companies can put a serious professional tool in their BCMer's hands — one that carries the artifact production, so the BCMer can carry the rest. It does not replace anyone. It gives them back the part of the job that the job was actually supposed to be.
The result is the ResiliencIA app. The first BCM AI expert. Built on Centaurization.
The paper is on SSRN, Zenodo, and ResearchGate. The Centaur is at resiliencia.app.