Why the framework exists
It started in 2023 as an experiment, in the middle of day-to-day consulting work. After thirty years of advising financial-services firms, I wanted to know what generative AI could carry on real deliverables: diagnostics, concepts, decision papers. From 2025 the experiment became a system: first in preparing for my interim mandate, then as the scaffolding of my daily work.
The first insight was sobering: individual results could impress, but quality would not repeat. What worked precisely today turned arbitrary tomorrow, because the standard depended on the day's dialogue. In a business where reliability is the product, that is disqualifying.
So I began codifying method: the diagnostic grids, root-cause analyses, evaluation logic and steering routines of three decades of project work became reusable building blocks. Results improved markedly. That exposed the next problem: building blocks without an order are not yet delivery. A backbone was needed. That became the framework: a phase model with gates at which every intermediate result is checked before work proceeds — begun inside a live interim mandate, then systematically extended.
Then came the most instructive phase. With every extension, side effects grew: rules contradicted each other, versions drifted apart, the checking apparatus sprawled. What organisations rightly fear about generative AI — creeping quality loss under growing complexity — was now sitting in my own system. With one difference: I could measure it.
The consequences went to the core of the design. Version control as the backbone, where no change lands unchecked. Conserved reference cases — golden cases — against which every change is tested automatically before it counts. A rigorous pruning of the checking apparatus down to the checks that demonstrably catch errors. A documented delta for every version step. And the most important rule: every sign-off rests with a human. The system structures, checks and keeps things consistent — the decisions are mine.
Today the framework uses AI for structure, consistency, diligence and verifiability — and leaves to the expert what belongs to him: judgement, context, accountability. Client data runs through a strict pseudonymisation regime; confidential information never leaves the protected perimeter.
I am not selling the claim that AI-assisted delivery works. I have lived through its failure modes, measured them, and closed them. Hence: proven delivery.
