Independent · No position to protect · No narrative to maintain
The Problem
The output looks finished. The evidence chain isn’t.
Generative AI is exceptionally good at producing plausible, coherent, well-formatted output. That is not the same thing as analysis.
The question is not whether the document looks convincing. The question is whether anyone can prove why it should be believed.
The hallucination is not the failure. The failure is the institutional belief that formatted output is analysis.
No reproducibilityThe same prompt can produce a different result.
No audit trailThe conclusion cannot be reconstructed from the output alone.
No falsificationConfidence in the document is not evidence for the claim.
The Chain
Every AI-assisted decision runs the same chain.
Seven links. The three in the middle are missing exactly the properties a risk committee would demand — and the link that restores them is the one most chains do not have.
Interrogation sits before the signature — or nowhere. Verification after the decision is called an inquiry.
When You Need Us
The problem usually appears before the error does.
01
A board paper reads perfectly, and nobody in the room can say who verified it.
02
Two reports disagree, and both are confident.
03
An investment committee is about to commit capital on AI-assisted analysis.
04
A regulator asks how output is verified — and there is no answer on file.
05
Something was published under the firm’s name that turned out not to exist.
Review AI-generated analysis already in circulation — board papers, treasury reports, due diligence and financial analysis. Claims traced, assumptions tested, contradictions surfaced and unsupported conclusions identified.
Written forensic brief within 48 hours. No scoping call required.
02 · Build what you need
Forensic Intelligence Engagement
from £10,000
Build verification into the analytical process itself. Forensic MI architecture, model audit, decision-chain analysis and capital or allocation risk mapping.
20-minute scoping call. No obligation. No pitch deck.
AI output · financial models · board papers · due diligence · decision chains
The Practitioner
Built inside institutional finance. Applied to AI.
Thirty years inside the machinery that produces institutional analysis — modelling, treasury, financial systems, consulting and forensic reconstruction.
The methodology was not developed in a seminar. It comes from working inside the systems, models and reporting chains that sit underneath consequential decisions.
The published forensic record — working papers, audits and corrections — is at paulfaulkner.com.
Practitioner RecordJPMorgan Chase · Global Treasury PwC · Consultant SG Kleinwort Hambros Bradford & Bingley · Model architect, 1999 CIMA qualified · 30 years institutional
The Method
Six questions. One fixed sequence.
Every engagement applies the same interrogation protocol. The full method is deliberately published rather than hidden behind a sales conversation.
01
Provenance
Where did it come from?
02
Assumption
What had to be true?
03
Contradiction
What works against it?
04
Confidence
Does certainty match evidence?
05
Reconstruction
Could it be reproduced?
06
Falsification
What would prove it wrong?
Most AI review asks whether the answer is correct. We ask whether it survives cross-examination.
