Innovation or Theater: AI Implementation Decision Framework

Publication note

Innovation or Theater is a working paper for deciding whether a proposed AI implementation should proceed as proposed. It introduces the AI Implementation Decision Framework, a deterministic assessment model that separates organizational AI posture from localized implementation risk.

The complete paper is available as a Download PDF. The companion Download Excel workbook turns the framework into an Excel-based assessment flow, and the workbook user guide explains how to complete and interpret the assessment.

For offline review or sharing, download the full research package ZIP with the paper PDF, workbook, and user guide.

Core question

The paper is organized around one implementation decision: should AI be trusted with this role here?

The framework asks organizations to define the use case, identify what AI newly makes possible to break, evaluate consequence bearers, set an autonomy ceiling, and decide whether the proposal should proceed as proposed.

The five things it makes you decide

The framework separates two questions most AI conversations blur together: is the organization ready, and is this specific use safe? It moves through five checks, in order.

1. Organizational AI posture — the readiness, skills, and constraints the organization brings before any single use case. A strong use case inside an unprepared organization still fails.

2. Use-case intake — the actual task, the accountable owner, and the data sources. "Use AI for support" is not a use case; "draft first-pass replies to billing questions, reviewed by a named person, from these sources" is.

3. Localized risk — what this AI role newly makes possible to break, and who carries the consequence when it is wrong. The risk is specific to the role, not to "AI" in general.

4. AI role fit — whether AI is actually suited to the task, or is being added because it is novel. Some work has the language and pattern for AI to help; some does not.

5. Autonomy alignment — how much independence the AI should have, and whether the proposal stays under that ceiling. Most failures are autonomy granted before trust is earned.

What "proceed as proposed" means

The output is deliberately narrow: not "is AI good here?" but "should this proposal move forward as written?" A NO is rarely a verdict against AI — it usually means a more bounded version of the same idea is fine. The framework is built to make the safer redesign obvious instead of killing the project.

Hard stops

Some gaps cannot be averaged away. Regardless of how strong the rest of a proposal looks, the framework forces a NO when the use case is undefined, no one is accountable for the result, the data sources are undefined, the localized-risk picture is incomplete, the autonomy is excessive for the role, or the redundancy cannot justify its own value. These are the conditions under which a confident answer would be meaningless, so the framework refuses to produce one.

A YES and a NO

A YES looks like a defined task, a named owner, known sources, a risk that is understood and survivable, and an autonomy level matched to earned trust — assistance with a human accountable for the result.

A NO looks like "let's let AI handle this" with no owner, vague inputs, and a quiet assumption that the output goes straight into real work. The demo was impressive; nothing underneath it was decided. That gap is the difference between innovation and theater.

The full paper develops each step with reason codes and scoring, and the Excel workbook turns it into a repeatable assessment. For the complete version, Download PDF.