Skip to content
Why GuardPilot

Not conceived in a boardroom. Built from 25 years of practice.

GuardPilot exists because one entrepreneur had the problem first-hand — and refused to accept it any longer. 25 years of executive experience in contract-heavy services produced 385,000 lines of personal knowledge. Translated with AI into a living system that watches 24/7 and asks the right questions.

385.000
input lines of personal knowledge

The moat is not the code. The moat is the knowledge.

Contracts, exceptions, calculations, invoicing rhythms, escalation rules, edge cases from 25 years of practice. Everything an experienced director knows — captured, structured and translated into rules that drive 27 agents. No consultancy model can replicate this; it has to be lived.

Four pillars

Why GuardPilot works where other AI projects stall.

01

Lived knowledge, not a consultancy deck

385,000 lines of input from 25 years at the helm — contracts, exceptions, escalation rules. Captured by the person who had to fix the mistakes.

02

Our own operation as testbed

27 agents run non-stop on our own company since autumn 2025. Every signal, every false positive, every edge case improves the system — before it reaches you.

03

Dutch & privacy-first

Personal data never leaves the network. Cloud analysis only on anonymised data. No US-only vendor lock-in.

04

Reproducible, not a black box

Every signal has a trail: source, model, evidence, decision. Two models must agree — if they disagree, it escalates to a human.

Positioning

Where GuardPilot stands

AI assistants and chatbots

Help one person with one task. They do not know your company.

Consultancy projects

They come, advise, leave. The report starts ageing on day one.

Enterprise data platforms

Powerful, but built for large corporates, with the teams, timelines and budgets that go with them.

GuardPilot is the fourth way. A control layer across your entire company, built by an entrepreneur who runs it daily on his own businesses, delivered with the architect included, and small enough to fit you.

ASKING THE RIGHT QUESTIONS YOURSELF

A question with an offer to help. Never an accusation.

Every question is pre-tested on all facets: laws and regulations, collective labor agreements, contract terms and the file itself — files are read, not just numbers. Tone rules are enforced deterministically before sending: figures as schema, so anyone sees the deviation at a glance.

fixed question structure · client X
priority 1
WHAT WAS MEASURED

18 hours over budget last week. Every hour straight from the source.

PERIOD & SOURCE

Week 29 · hours ledger · previous weeks ran through an extra-work number, this one didn't.

THE QUESTION

“Regular work over budget — why? Or forgotten extra work? Then we create a number and the invoice can go out.”

CHECK IT YOURSELF

Check Tuesday's work order — extra work without a number?

manager 08:41 — “Forgotten extra work — rush job Tuesday.”
GuardPilot 08:41 — “Extra-work number created, 18 hours rebooked. The invoice can go out.”
OVERLOAD PREVENTION — HARD-ENFORCED
  • A small, hard cap on questions per client per run
  • Repeat filter: no identical question within seven days
  • Materiality floors filter out micro amounts
  • Ranking on financial impact decides what comes first
  • And the most important rule: healthy clients receive no question.
Validation before sending

No raw signal ever reaches a manager. Everything first passes elimination, the quality gates and — where configured — human domain validation. The human decides — always.

IF NO ANSWER COMES

Controlled escalation, never stalling.

Unanswered doesn't vanish. Escalation is always a proposal to a human — never an automatic decision.

1

Question to the directly responsible person

Site or project.

2

Controlled retry

On repeated failure, a deterministic fallback question; the chain never stalls.

3

Escalation proposal, one level up

From site via region and district to executive level. Always a proposal to a human.

VERIFY BEFORE SIGNALING

Reproducing a claim is not verification.

This is the difference with dashboards and BI: before a finding reaches a human, it passes four verification layers. What doesn't hold up disappears — with the reason recorded.

LAYER 1

Figure-exact

Every amount tested against the raw source.

LAYER 2

Elimination

Actively try to explain it away: contract, cost type, duplicate, reversal, periodic rhythm.

LAYER 3

Miss check

What is being missed around this finding?

LAYER 4

Maturity

Is the period closed enough for a conclusion? If not: park and retry.

JOIN CHECK

Every data load gets a measured join rate. Below threshold means stop and investigate — no silent data loss.

HEALTHY IS SILENT

Hundreds of signals in one category means wrong logic, not hundreds of real problems — treated as a system fault.

PROOF PER AMOUNT

Agent, source file, timestamp and category — without that lineage the record is rejected by the outgoing contract.

REGRESSION GUARD

Validated client cases run as tests after every change. If the change breaks an outcome, the change reverts — not the test.

HUMAN VALIDATION LOOP

The internal domain expert reviews outcomes; that feedback flows back into the knowledge base. The system learns cumulatively from every round.

FRESHNESS GUARD

Every status layer reads live sources with a freshness check. A dashboard without freshness checks lies without you noticing.

THE THINKING MODEL

Knowledge only counts when it lives as a lens in code.

The most important discovery of the whole journey: loose documents and dictated knowledge don't reach daily analysis on their own. That's why all domain knowledge has been converted step by step into deterministic thinking layers that run on every client every night — auditable and traceable. And not just clients: every mutation — client, contract or personnel file — is tested against laws, collective labor agreements and the organisation's own contractual commitments.

Hours actual vs budgetInvoice rhythm — monthly, quarterly, tour profilesSector norms by industryPeriodic work: agreement vs invoicingObject drill: result per siteMulti-year trend & seasonality — two years backExtra hours vs extra revenueContract status & termination per objectLabor cost, hiring-in and margin in contextPersonnel files & every mutation — read, not just countedLaws, CAO and premium rules per mutationResult layer: weighting on financial impact
LEVEL 1

The lenses measure

Every thinking layer adds measured, neutral facts. Measure, do not judge.

LEVEL 2

The code decides

The sovereign decision layer determines whether a question is asked, about what and with which figures — with a fixed ranking of question types by weight.

LEVEL 3

The model phrases

Only the wording, within strict tone rules. If the model fails, the code builds a fallback question itself — the chain never stops on a model failure.

Understanding before judgment. The system does not judge — it tests hypotheses. Every question carries the context and the numbers it rests on, so the recipient can think along instead of defend. And deciding remains human work.

VISION

From control to intelligence.

Phase 1 — live

Contract monitoring + financial signaling

27 agents, daily audits, management briefings.

Phase 2 — in progress

Operational intelligence

Patterns, seasonal effects, proactive signaling.

Phase 3 — Q4 2026

Strategic decision support

Portfolio analysis, margin optimization, data-driven strategy.

End goal: a digital brain that watches, learns and thinks along 24/7. Not a replacement, but an amplifier — the human decides.

Origin

How GuardPilot came about

I have always wanted to get to the bottom of things. If I do not understand something, I do not let it go until I do. Fourteen months ago I saw what this technology is going to do to companies. Not as an extra tool, but as a change in how a company is run.

Detail

That was the moment for me. The transition everyone is talking about now, I went through back then.

I am Erwin de Jong, an entrepreneur, not a programmer. I had to learn to work with AI, and it was the steepest learning curve I have ever had. 116,482 times I have steered the system since March, measured. Sometimes a whole paragraph, sometimes one sentence, sometimes just a yes or a no. Every time to keep a grip, every time to make it better. Across 2,060 working conversations with the system, between 16 March and 20 August 2026. Nights, weekends, sessions that did not stop.

Detail

This was the biggest task I have ever taken on.

My first instinct was to outsource. That was a mistake, and an instructive one. What is self-evident to me is not self-evident to someone from outside. They see a piece of the puzzle. So I built it myself. All of it.

Detail

How an invoice travels through a company, which mail matters and which is noise, when a problem has no owner, who has to sign what and when. This can only be built if you know companies from the inside.

That is why GuardPilot is not an IT solution. It is twenty-five years of experience, gained in every company I was responsible for, every single day, applied to new technology.

Detail

Every process I know is in there. Every question I would ask myself, the system asks. And it keeps learning, because I keep learning.

We test everything immediately on real cases. Live acquisitions, a three-year fall in results, questions from the tax authority, our own structure. In real environments, with real consequences.

Detail

Every outcome is validated by my own finance people. That is what makes this different from the many good ideas and initiatives out there.

What I see in people who understand it is first enthusiasm and then alarm. They see the implications for their company, and beyond their company.

Detail

I now give seminars and the phone does not stop. That is why it is written down here, so anyone can go through it themselves.

I also know this is only the beginning. The future will look different from today. But we are out in front, we understand what it can do, and we move with it. Every day.

Own measurement 21 August 2026, one environment, earlier months and other environments not included.

Whatever you charge for this, it is too little.
a CFO, after a presentation

Erwin de Jong, founder

New to terms like agents, knowledge circle or six-layer architecture? Browse the glossary.

Ready to look inside the engine room?

We show what's already running — and share how.

We take on a limited number of engagements per year. The first companies build along and notice it. Join the waitlist

Context and history

Alongside the agents, GuardPilot works on an organisation's history and context. Fourteen days of production — counts, cases, working rules and incidents — are described openly and without names.

View context and history →