One-Shot Audit Risk Calculator

Answer eight questions about how you actually run an AI audit today to see where a one-shot, drop-it-in-ChatGPT process is most likely to fail, and the failure mode most likely to bite first.

Answer all eight questions from 0 (your process already does this) to 4 (this is closer to a one-shot, drop-it-in-ChatGPT habit). Score what your process actually does today, not what you intend to fix.

Rate your process on eight questions

0 = fully disciplined practice, 4 = fully one-shot practice. Each question belongs to one of the three failure surfaces below the form.

The deliverableCan every finding in your last audit be traced back to a specific document, quote, or interview answer?

Every finding cites its exact sourceFindings are a confident summary, not traceable

The deliverableWhen a document says one thing and an interview says another, does your process surface the conflict, or quietly pick one?

Contradictions are surfaced and flaggedOne version gets averaged away, unnoticed

The deliverableIf the client asked 'who says so' about a recommendation, could you point to the exact source, or would you point to the model?

You can point to the exact source, every timeThe honest answer is "the model thought so"

The relationshipDoes what you learn on one workflow or engagement carry forward and inform the next one, or does each answer start from zero?

Findings compound across engagementsEvery engagement is solved in total isolation

The relationshipAs a client's documents and interviews pile up, does your process hold up, or do you start feeding a single window more than it can carry?

Scale adds structure, not strainYou are stuffing one window past its limit

The relationshipMid-engagement, can you find any past finding or document in seconds, or does it live scattered across a drive, a doc, and a thread?

Everything is organized and findable instantlyIt is scattered across drive, doc, and text thread

The implementationDo you assess a client's data, systems, and organizational readiness before you recommend a solution, or does the solution come first?

Readiness is assessed before any solution is namedSolutions get recommended before readiness is checked

The implementationHave you had to walk back a recommendation after the client was already excited, because a readiness gap surfaced too late?

Never. Readiness gaps surface before the pitchRegularly. The walk-back happens after the sale

How this works

A one-shot audit, dropping a client's documents into a general-purpose model and prompting it to "do an audit," does not fail because the prompt was weak. It fails in three specific places at once: the deliverable (can a finding cite where it came from), the relationship (does anything you learn carry forward, or does every engagement start from zero), and the implementation (is a solution sequenced behind a readiness check, or recommended before anyone looked).

The math here is deliberately simple and fully visible. Each of the eight questions above is answered on the same 0-4 scale (0 = disciplined practice, 4 = one-shot practice) and belongs to exactly one of the three surfaces. Every surface score is just the sum of its own questions' answers, turned into a percentage of that surface's own maximum. The overall score is the sum of all eight answers, turned into a percentage of the overall maximum. That is the entire formula: additive, equally weighted, and stated here in full. There are no hidden weights and this is not Audity's internal scoring engine, it is an illustrative model built to make the three failure surfaces concrete.

The stakes are not hypothetical. More than 80 percent of AI projects fail, roughly twice the failure rate of non-AI IT work, and the leading root cause is misunderstanding the problem and optimizing for the wrong metric, not the model (RAND, 2024). And provenance is not a nicety even when a system is grounded: in a study of leading AI legal-research tools, grounded systems still hallucinated in roughly 17 to 33 percent of queries (Stanford RegLab, 2025). If even grounded tools miss that often, an ungrounded one-shot summary is not a shortcut. It is where the risk actually lives.

Related free tools for consultants

Score a client's actual readiness before you architect a solution.
Stress-test your phase order if the readiness surface scored high above.
See every free tool in the Diagnostic Discipline toolkit.

This tool is one page of the method

The Diagnostic Discipline is the full framework behind this tool. Join the community to put it into practice with other AI transformation partners, or read the paper it comes from.

Join the AI Transformation Partner community Read the full Diagnostic Discipline paper