Decision Framework

When Not
to Use AIA Decision Framework for Teams

Everyone is having the adoption conversation. Nobody is having the other one: when do you say no? Using AI in the wrong place does not just produce bad output there. It erodes trust in every output you produce, including the ones where AI is actually helping.

Get the framework · $29 See what's inside
One PDF, no filler Usable in under an hour Instant download

“Battle-tested in production environments.”

Jared Spataro, Microsoft’s CMO for AI at Work, citing the author’s public Copilot prompt library on LinkedIn. This framework holds itself to the same standard: the decision lines in it come from real rollouts, including the calls that went wrong.

The false premise

Most teams ask the wrong question.

"Can AI do this?" The answer is almost always yes. A model will generate the report, score the candidates, draft the response, classify the tickets, fluently and fast. That is not the useful question.

What teams ask

"Can AI do this?"

What this framework asks

"What happens when AI does this wrong, and nobody catches it?"

Wrong and uncertain gets caught.
Wrong and confident gets shipped.
The failure mode that makes AI dangerous is not bad output. It is fluent, confident, well-formatted output that is wrong, and a reviewer who accepts it.
What's inside

A working document, not a reading document.

Print the checklists. Fill in the matrix. Share the scripts before the vendor demo, not the week after you committed. Every tool here is built to run in under an hour for any proposed use case.

5-Question Pre-Flight Check

Ten minutes. Two or more yes answers and you proceed with caution. Three or more and you do not deploy until conditions change.

Use-Case Scoring Matrix

Score any use case on four dimensions and get a verdict. With a worked example and a contrast example.

Red Flags Checklist

12 observable signals in three categories: output risk, process risk, governance risk. Check a box, you have a concrete problem to fix.

"No AI" Conversation Scripts

Exact language for three situations where you need to say no without becoming the anti-AI person in the room.

The Decision Tree

The full workflow, top to bottom. Stop at the first branch that creates a constraint. No skipping ahead.

10 Real Use-Case Verdicts

Yes, no, and yes-with-guardrails on ten real workflows, each with the full reasoning so you can adapt it to yours.

The AI Use Policy Clause

Copy-pasteable, jurisdiction-neutral language for your acceptable-use policy. Review and accountability requirements included.

The Week-After Checklist

Seven moves for the week after you say no, so the decision sticks instead of quietly reversing six weeks later.

Applying It to M365 Copilot

A surface-by-surface mapping: grounding, agents, DLP, and the draft-vs-act line that changes the whole risk picture.

The 5-question pre-flight check

Answer these before you deploy AI anywhere.

They take ten minutes. Each one catches a specific, expensive failure mode that vendors will not raise in the demo.

Does the task require original judgment that cannot be verified without deep domain expertise?

Is the cost of a confident wrong answer higher than the cost of a slow right answer?

Does the output require a human signature, legal accountability, or professional liability?

Is the input data incomplete, unverifiable, or based on context the model cannot access?

Is the relationship or trust context more important than the information content?

Two or more yes  = proceed with significant caution.  Three or more  = do not deploy until the conditions change.
The scoring matrix

Score it. Get a verdict. Move.

Rate any use case on reversibility, verifiability, and task consistency, then subtract its stakes. The number tells you which of three things to do.

Deploy
Score 5 to 9

Strong AI candidate. Deploy with standard review practices.

Guardrails
Score 2 to 4

Proceed only with specific review gates in the critical path before any output is used.

Don't deploy
Score below 2

The conditions are not right. Fix the gap, or keep the human in the decision.

Score = ( Reversibility + Verifiability + Consistency ) − Stakes

Ten real verdicts, with the reasoning

YesSummarise a weekly status report for PM review before it goes out
NoScore and rank job candidates without the manager reviewing the rationale
GuardrailsClassify and route incoming IT support tickets
NoAuto-reply to customer service emails with no human review before sending
GuardrailsGenerate contract boilerplate from a pre-approved template library
NoProduce a root-cause analysis of a workplace incident for a regulator
Bonus chapter

Applying the framework to a Microsoft 365 Copilot rollout

Copilot is where AI arrives in most organisations whether or not anyone decided. This section maps every part of the framework to the real surfaces: where grounding on your Graph fails silently, why a citation is a place to check and not proof, and the draft-vs-act line where reversibility collapses and an agent needs a gate before it ships.

Who it's for

For the person who has to make the call.

IT operationsProject managersTeam leadsAI championsAnyone handed an "add AI" mandate
This is not anti-AI. It is what makes the AI you do use trustworthy. The goal is to say no in the right places, so that yes actually means something.
Get the framework

Get in, get clarity, move.

  • The 5-question pre-flight check and the scoring matrix
  • 12 red flags, 3 conversation scripts, the decision tree
  • 10 worked verdicts, the policy clause, the week-after checklist
  • The Microsoft 365 Copilot mapping
$29
One PDF. Instant download. Yours to keep.
Get the framework now
Licensed for individual and team use.

No filler. No chapters about "the AI revolution." Get in, get clarity, move.

Questions

Before you buy

What format is it?
One PDF, delivered instantly on purchase. It is built as a working document: print the checklists, fill in the matrix, and share the scripts with your team.
Is this an anti-AI book?
No. It exists to make the AI you do deploy trustworthy. The point is to say no in the right places so that your yes means something, not to lower your adoption rate.
Who is it for?
IT operations, project managers, team leads, AI champions, and anyone handed a mandate to "add AI" who has to figure out where it actually belongs. No technical background required.
Does it cover Microsoft 365 Copilot?
Yes. The framework is tool-agnostic, and a dedicated section maps it surface by surface to a real Copilot rollout: grounding and oversharing, agents that act versus draft, DLP, and the accountability register.
How long does it take to use?
The pre-flight check is ten minutes. A full evaluation of any single use case is designed to take under an hour. If it takes longer, the use case is not defined clearly enough to evaluate yet.