Use this guide with the Innovation or Theater research paper and the Download Excel workbook. The workbook tool page remains available at AI Implementation Assessment Workbook.

The complete research package is also available as a ZIP download.

This guide explains how to use the AI Implementation Assessment Workbook, the companion Excel tool for the AI Implementation Decision Framework.

The workbook answers one question:

Should this AI implementation proceed as proposed?

The answer is strict:

YES - proceed as proposed

NO - do not proceed as proposed

A NO does not always mean "never use AI." It means the current proposal should not move forward in its current form.

The Core Question

The workbook is built around a simple distinction:

The question is not "Can AI do this?"

The question is "Should AI be trusted with this role here?"

Capability alone is not enough.

Workbook Flow

The workbook follows six stages:

Org Posture -> Use Case Intake -> Localized Risk -> Digestion / Flags -> Formula -> Decision

1. Org Posture

Evaluates whether the organization can responsibly absorb AI-related downside.

2. Use Case Intake

Defines the proposed implementation clearly enough to assess.

3. Localized Risk

Identifies what the implementation can break.

4. Digestion / Flags

Converts defined data into blockers, constraints, redirects, and reason codes.

5. Formula

Combines posture, intake, localized risk, AI Role Fit, autonomy, and hard stops.

6. Decision

Produces the final YES or NO as proposed.

Who Should Complete the Org Posture Assessment?

Org Posture should not be completed by one person alone.

Recommended participants:

executive or senior decision-maker
operations leader
IT or systems owner
data/source owner
legal, compliance, privacy, or security voice where relevant
frontline workflow owner
training/process documentation owner
someone optimistic about AI
someone skeptical about AI

A good posture assessment includes people who understand authority, how the work actually happens, what breaks when the work fails, and what downside the organization can absorb.

How Often Should Org Posture Be Reassessed?

Recommended cadence:

Reassess organizational posture every six months, or sooner after a major leadership, regulatory, operational, vendor, policy, training, or AI-use change.

Reassess immediately after:

AI incident or near-miss
new AI policy
major workflow change
shift from internal AI use to customer-facing AI use
new high-impact AI implementation
major vendor/tool change
leadership change
significant training completion

How to Reuse Org Posture Across Projects

The Org Posture assessment is reusable.

Recommended process:

Complete the organization-level posture assessment.
Save the workbook or Org Posture tab as the current posture baseline.
Use that baseline when evaluating specific AI implementations.
Run Use Case Intake and Localized Risk separately for each proposed implementation.
Update posture when the organization changes materially.

Who Should Complete an Implementation Assessment?

The person most excited about the AI implementation should not be the only person defining its risk.

Recommended participants:

implementation sponsor
current workflow owner
accountable decision-maker
user or person who depends on the workflow
technical/tooling owner
data/source owner
person who can identify downside
person willing to challenge whether AI is actually necessary

For high-consequence implementations, add legal, compliance, privacy/security, HR, safety, clinical, financial, or other domain experts as appropriate.

Completing Use Case Intake

Use Case Intake defines the implementation.

It should answer:

What is being proposed?
What workflow, system, decision, or human moment is affected?
How does the workflow operate today?
What is AI supposed to do?
What output or action will AI produce?
Who uses it?
Who could bear consequence if it fails?
What data or knowledge sources will AI use?
What tools can AI touch?
What human review exists?
Who owns the outcome?
What would count as success?
What would count as unacceptable failure?
What non-AI alternatives were considered?

Core rule:

You cannot assess what you cannot define.

Completing Localized Risk

Localized Risk asks what the implementation can break.

The assessment should identify:

Existing Localized Risk
AI-Driven Localized Risk
Consequence Types
Consequence Bearers
Severity
Exposure
Reversibility
Containment
Complexity
Redundancy

Core rules:

Consequence must be identified before severity can be measured.

Low task complexity does not guarantee low Localized Risk.

You cannot digest an undefined failure surface.

Interpreting YES

A YES - proceed as proposed means the implementation passed the workbook's current assessment logic.

It means:

posture meets the required threshold,
the use case is defined,
localized risk is acceptable for the proposed autonomy,
AI Role Fit is sufficient,
hard stops were not triggered,
and the proposed AI role does not exceed the autonomy ceiling.

A YES is not legal clearance, compliance certification, or a guarantee of safety. The organization still owns the outcome.

Interpreting NO

A NO - do not proceed as proposed means the current proposal should not move forward in its current form.

A NO may be triggered by:

undefined use case
weak posture
incomplete intake
high localized risk
unclear consequence bearers
excessive autonomy
no accountable owner
broad tool access
weak containment
failed redundancy test
deterministic tool preferred
theater risk

A NO should be treated as a design signal. Ask:

What would need to change for this to become acceptable?

How to Handle Disagreement

Disagreement is useful.

When participants disagree:

Decide whether the disagreement is factual or interpretive.
Use evidence before preference.
Do not let optimism fill missing data.
Do not let fear inflate every answer.
Record major disagreement if it affects the assessment.

A good rule:

When the assessment team cannot resolve a critical uncertainty, the workbook should not pretend certainty exists.

Practical Usage Rules

Score the functional organization, not the theoretical organization.
Score the proposed implementation, not the best possible future version.
Preserve the worst credible consequence.
Separate support from authority.
Treat tool access seriously.
Treat human moments seriously.
Do not use AI to automate confusion.
If the safety layer does the trusted work, AI must justify its presence.
Fit does not equal permission.
The workbook structures judgment; it does not remove responsibility.

Recommended File Practice

Use separate files for organization posture and implementation assessments.

Org_AI_Posture_Assessment_YYYY-MM-DD.xlsx
AI_Implementation_Assessment_[ProjectName]_YYYY-MM-DD.xlsx

For each implementation assessment, record:

posture baseline version/date
implementation name
accountable owner
assessment participants
decision date
final verdict
reason codes
required changes
reassessment trigger

What the Workbook Does Not Replace

The workbook does not replace legal review, compliance review, cybersecurity assessment, procurement review, safety engineering, clinical review, HR/legal decision review, domain expert review, formal AI governance, or executive accountability.

The workbook is a decision-support tool.

It helps organizations make a more structured case for or against AI implementation.

It does not absorb responsibility.

Closing Principle

AI implementation should not be decided by the loudest optimist or the loudest skeptic in the room.

It should be assessed through Org Posture, Use Case Intake, Localized Risk, autonomy, AI Role Fit, and consequence.

The workbook exists to make that assessment more structured.

The core question remains:

Should AI be trusted with this role here?