Table of Contents
Introduction
Ask an AI model for a risk assessment or a cost analysis and it will produce one, confidently, whether or not it is any good. That confidence is the problem. Nothing in a single AI response tells you if it is a genuinely strong output or a plausible-sounding first draft.
The Single-Draft Problem
A single AI draft has no built-in check against its own mistakes. It can miss a risk category entirely, state something as fact that is actually an assumption, or produce a recommendation that reads well but does not survive scrutiny. Without a second pass, the person receiving it has no signal about which of these happened.
The Draft-Then-Audit Loop
Every production tool in True North Solutions' agentic AI lineup runs the same structure: one AI produces the output, a second, independent AI grades it against a fixed rubric, and it gets sent back for revision until it clears the bar. The drafting AI does not grade its own work. That separation is what makes the grade meaningful rather than circular.
Rubrics Are Not Interchangeable
The rubric changes by tool, because what counts as a good output changes by task. The ESG Risk Analyzer is graded against McKinsey-grade criteria for analytical rigor, actionability, and completeness. Microbot Miner's automation candidates are scored against a 7-criteria readiness rubric covering frequency, duration, error rate, and systems touched. A generic quality check would miss what actually matters in each case.
What You Actually Get
The practical difference shows up in the deliverable: a documented quality score and an audit trail per recommendation, not just a polished document. If a recommendation is questioned later, there is a specific, graded reason it was included, not a guess about what the AI was thinking.
FAQ
What does 'independent AI auditor' actually mean here?
One AI produces the draft output. A second, separate AI call, prompted only to grade the work against a fixed rubric, reviews it. The drafting AI never sees or influences its own grade.
What happens if the output fails the audit?
It gets sent back for revision and re-graded. This repeats until the output clears the rubric, so what reaches you has already passed a defined quality bar rather than being the first draft.
Does this apply to every tool, or just some?
Every production tool True North Solutions ships runs the same draft-then-audit loop, from the North Star Compiler's Hoshin Kanri X-Matrix to the ESG Risk Analyzer's supply chain risk assessment.
Is the audit rubric the same for every tool?
No. Each tool is graded against criteria suited to its output. The ESG Risk Analyzer is scored against McKinsey-grade criteria for analytical rigor, actionability, and completeness; Microbot Miner's automation candidates are scored against a 7-criteria readiness rubric.