Appendix D: Decision Flowcharts
You’re at your desk. You have a task. You’re wondering whether AI can help—and if so, how.
This appendix connects the book’s frameworks into a single decision path. Start at the Master Flow, then follow the arrows to whichever chapter’s flowchart applies. Each flowchart is designed to get you from question to action in under two minutes.
The Master Decision Flow
Every AI task decision follows this sequence. Start here every time.
┌─────────────────────┐
│ I HAVE A TASK. │
│ Should AI help? │
└──────────┬──────────┘
│
▼
┌──────────────────────────┐
│ STEP 1: SCORE IT │
│ Is this a good AI task? │
│ → Flowchart 1 (Ch 5) │
└──────────┬───────────────┘
│
┌──────┴──────┐
│ │
Score < 100 Score ≥ 100
│ │
▼ ▼
NOT YET. ┌──────────────────────────┐
Build AI │ STEP 2: ASSESS THE RISK │
skills on │ What could go wrong? │
easier │ → Flowchart 4 (Ch 11-12) │
tasks first. └──────────┬───────────────┘
│
┌──────────┴──────────┐
│ │
Red flag Green / Yellow
│ │
▼ ▼
STOP or get ┌──────────────────────────┐
executive │ STEP 3: BUILD IT │
sign-off. │ What's the workflow? │
│ → Flowchart 2 (Ch 7-8) │
└──────────┬──────────────┘
│
▼
┌──────────────────────────┐
│ STEP 4: FEED IT │
│ What inputs does it │
│ need to do this well? │
│ → Flowchart 3 (Ch 9) │
└──────────┬──────────────┘
│
▼
┌──────────────────────────┐
│ STEP 5: REVIEW IT │
│ How carefully should I │
│ check the output? │
│ → Flowchart 5 (Ch 14) │
└──────────┬──────────────┘
│
▼
┌──────────┐
│ SHIP IT │
└──────────┘Flowchart 1: Is This a Good AI Task?
Framework: FFCC Scorecard (Chapter 5)
When you need this: You have a task and you’re wondering if AI would actually help—or if you’d spend more time wrestling with it than doing it yourself.
IS THIS TASK A GOOD AI CANDIDATE?
═════════════════════════════════
Do you do this task at least weekly?
│
├─ No ──→ ⚠ Caution. Low frequency = slow learning,
│ low ROI. Score Frequent: 1-2.
│ (Still continue—but set expectations.)
│
└─ Yes ──→ Score Frequent: 3-5.
│
▼
If AI gets it wrong, can you catch and fix it
before anyone outside your team is affected?
│
├─ No ──→ ⚠ High stakes. Score Forgiving: 1-2.
│ You'll need heavy review (see Flowchart 5).
│ If errors are irreversible → seriously
│ consider skipping AI for now.
│
└─ Yes ──→ Score Forgiving: 3-5.
│
▼
Can you describe exactly what good output looks like?
(A template exists, format is defined, success is measurable)
│
├─ No ──→ ⚠ Fuzzy specs = generic results.
│ Score Clear: 1-2.
│ Fix this first: build a template
│ or define your criteria (Ch 9 can help).
│
└─ Yes ──→ Score Clear: 3-5.
│
▼
Can this task run without touching 3+ other systems,
teams, or approval chains?
│
├─ No ──→ ⚠ Dependencies multiply problems.
│ Score Contained: 1-2.
│ Can you isolate a sub-task that IS
│ contained? Score that instead.
│
└─ Yes ──→ Score Contained: 3-5.
│
▼
MULTIPLY your four scores.
400–625 → ✅ Excellent. Start here.
200–399 → ✅ Good. Second-wave candidate.
100–199 → ⚠ Proceed with caution. Validate first.
Below 100 → ❌ Not yet. Build skills on easier tasks.The trap: We chase impressive projects (comp analysis, client presentations) while ignoring daily tasks that score far higher. Teams spend a quarter on an AI proposal generator that quietly dies while daily email triage sits ignored. Trust the math, not the glamour.
- Department leaders: Score your team’s top 5 recurring tasks—the ones that eat hours across multiple people every week.
- Individual contributors: Score your personal daily friction points—the tasks you dread that follow a predictable pattern.
- Small company CEOs: Score the decisions that repeat but don’t need your judgment—just your time.
- Senior leaders: Score the information-gathering that precedes your real decisions, not the decisions themselves.
Flowchart 2: What’s the Workflow?
Framework: Five-Component Workflow Structure (Chapters 7–8)
When you need this: You’ve scored a task and it’s a good candidate. Now you need to turn it into a reliable, repeatable process—not just a one-off conversation with ChatGPT.
BUILD YOUR WORKFLOW
═══════════════════
STEP 1: NAME YOUR TRIGGER
What kicks this off?
│
├─ A specific time ──→ Best. "Every Friday at 3 PM."
│ (Most reliable) Schedule it now.
│
├─ A work event ────→ Good. "New support ticket arrives."
│ (Event-based) Define the exact event.
│
└─ "When I remember" → Weakest. Will be skipped when
(Manual) you're busy. Can you convert to
time-based or event-based?
│
▼
STEP 2: LIST YOUR INPUTS
What does AI need to do this well?
│
├─ Core content ────→ The actual material to process
├─ Context ─────────→ Audience, history, goal
├─ Constraints ─────→ Length, format, things to avoid
└─ Examples ────────→ What good output looks like
│
For each: "If this were missing, how would
output quality suffer?" Cut what doesn't matter.
(See Flowchart 3 for deeper input design.)
│
▼
STEP 3: DEFINE THE PROCESSING
What does AI do with the input?
│
Pick ONE specific verb:
│
├─ Summarize ├─ Analyze
├─ Draft ├─ Compare
├─ Categorize ├─ Translate
└─ Extract └─ Other: _______
│
⚠ If you need multiple verbs, break into
sequential steps—not one mega-prompt.
│
Also define the output format:
"3 bullet points" / "500-word draft" /
"sorted list with scores"
│
▼
STEP 4: SET YOUR REVIEW
What are you checking?
│
├─ Accuracy ────→ Are facts correct?
├─ Tone ────────→ Does it sound right?
├─ Completeness → Did it miss anything?
└─ Appropriateness → Is this safe to use?
│
Set a time budget: _____ minutes per output.
(If review takes as long as doing it yourself,
fix the input—don't just push through.)
│
▼
STEP 5: CONNECT THE ACTION
What happens after you approve?
│
Pick a specific verb:
Send / Publish / Save / Escalate /
Assign / Schedule / Archive / Submit
│
⚠ If there's no action, there's no workflow
—just a conversation.The trap: Most workflow failures come from Step 2 (inadequate input), not Step 3 (bad prompting). If output is consistently off, fix your inputs before rewriting your prompt. Also: run the workflow 3 times before judging it. The first run always feels rough.
- Department leaders: Your triggers are often time-based (weekly summaries, monthly reviews). Your actions usually flow outward—to your team or upward to leadership.
- Individual contributors: Your triggers are often event-based (new assignment, client email). Your actions are personal deliverables.
- Small company CEOs: Your triggers are often daily (morning briefing, inbox processing). Your actions are decisions and delegation.
- Senior leaders: Your triggers are often cyclical (before board meetings, quarterly reviews). Your actions are organizational communications and policy.
Flowchart 3: Are Your Inputs Good Enough?
Framework: Layered Input Pattern (Chapter 9)
When you need this: Your workflow exists but the output is generic, inconsistent, or requires heavy editing. The fix is almost always in the input, not the prompt.
DIAGNOSE YOUR INPUT QUALITY
════════════════════════════
Is the output generic—it could apply to
anyone's situation, not specifically yours?
│
├─ Yes ──→ You're missing CONTEXT.
│ Add: who this is for, what they
│ already know, the history, why
│ this matters right now.
│
└─ No ───→ ▼
Is the output the wrong format, length,
or tone?
│
├─ Yes ──→ You're missing CONSTRAINTS.
│ Add: word count, format spec,
│ tone guide, what to include,
│ what to avoid.
│
└─ No ───→ ▼
Does the output vary wildly between runs—
good one time, bad the next?
│
├─ Yes ──→ Your inputs are AMBIGUOUS.
│ AI is filling gaps with guesses.
│ Use the Expert Test: hand your
│ input to a colleague with no
│ other context. If they ask
│ clarifying questions, those
│ reveal your gaps.
│
└─ No ───→ ▼
Is the output close but missing your
unique perspective or voice?
│
├─ Yes ──→ That's REWRITE territory.
│ Use the AI output as raw
│ material and rewrite in your
│ own voice. That's working as
│ designed (see Flowchart 5).
│
└─ No ───→ ✅ Your inputs are solid.
If output is still off, the
problem may be task fit (revisit
Flowchart 1) or review calibration
(Flowchart 5).Build better inputs with four layers (in order):
| Layer | What It Includes | Example |
|---|---|---|
| 1. Context | Who, history, constraints | “For a 50-person marketing team transitioning from agency to in-house” |
| 2. Content | The actual material | The data, document, or conversation to process |
| 3. Constraints | Format, tone, inclusions, exclusions | “500 words, professional but not formal, include metrics, no jargon” |
| 4. Request | What to do with the above | “Draft a summary highlighting the three biggest risks” |
The trap: People spend 80% of their effort on prompt wording and 20% on input quality. Results come from the inverse: 80% input quality, 20% prompt wording. A mediocre prompt with excellent context beats a perfect prompt with thin context. Every time.
- Department leaders: Your biggest input gap is usually aggregation—you have the data across your team, but you’re not feeding enough of it to AI. Pull from project tools, standup notes, and cross-functional updates.
- Individual contributors: Your biggest gap is implicit expertise—you know things about the client or project that you haven’t written down. Make it explicit.
- Small company CEOs: Your biggest gap is signal vs. noise—you have breadth but need to curate what’s relevant. Don’t dump everything in.
- Senior leaders: Your biggest gap is the strategy-to-operations bridge—connecting your high-level context to the specific task at hand.
Flowchart 4: What’s the Risk?
Framework: Permission Framework (Chapters 11–12)
When you need this: Before expanding AI into a new area—especially one involving sensitive data, external communications, or decisions that affect other people.
ASSESS THE RISK
═══════════════
QUESTION 1: What's the specific benefit?
│
Can you say it in one sentence with a
number in it?
│
├─ No ──→ 🛑 STOP. Define the benefit
│ before proceeding. "Saves time"
│ and "improves efficiency" are
│ not specific enough.
│
└─ Yes ──→ Write it down: _______________
│
▼
QUESTION 2: What's the actual exposure?
│
What data does AI see?
│
├─ Public info only ────────→ Low exposure
├─ Internal, non-sensitive ─→ Medium exposure
├─ Customer-facing ─────────→ High exposure
└─ Regulated / PII / HR ───→ Very high exposure
│
Who could be affected by errors?
Just you → Low. Your team → Medium.
Customers → High. Public → Very high.
│
▼
QUESTION 3: What are the failure modes?
│
Name the 3 most likely ways this goes wrong:
1. _______________________________________
2. _______________________________________
3. _______________________________________
│
For each: how quickly would you detect it?
│
├─ Before it ships (human review) → Good
├─ Within hours (monitoring) ────→ Acceptable
├─ When someone complains ───────→ ⚠ Danger
└─ Only when sued / audited ─────→ 🛑 Stop
│
▼
DECISION:
│
├─ Clear benefit + Low exposure +
│ Recoverable failures
│ → ✅ GREEN LIGHT. Proceed.
│
├─ Clear benefit + Medium exposure +
│ Recoverable failures
│ → ⚠ YELLOW. Proceed with specific
│ safeguards. Name them:
│ - What extra review? ___________
│ - What monitoring? _____________
│ - What's the kill switch? ______
│
├─ Clear benefit + High exposure
│ → 🔶 PILOT FIRST. Test with limited
│ scope before broader rollout.
│
└─ Vague benefit, OR unrecoverable
failure modes, OR very high exposure
→ 🛑 RED FLAG. Don't proceed without
executive sign-off and legal review.Data access rule of thumb: Start with minimum viable access. Customer summaries before full records. Financial dashboards before transaction data. Specific communications before archives. You can always expand later—clawing back access after an incident is much harder.
The trap: “Proceed with safeguards” without naming the specific safeguards is not a decision—it’s a deferral. If you can’t name the safeguards right now, you haven’t finished the assessment.
- Department leaders: Your exposure is usually team-multiplied. An error in your workflow gets distributed to 5-50 people. Factor the blast radius, not just the single-output risk.
- Individual contributors: Your primary risk is usually reputational. AI doesn’t take responsibility—you do. Ask: “Am I comfortable defending every claim in this output?”
- Small company CEOs: You often have full authority but limited safety nets. No legal team to catch problems, no compliance department to flag issues. Be more conservative, not less.
- Senior leaders: Your risk is often organizational—setting precedent. If you use AI carelessly, your team will too. Model the behavior you want to see.
Flowchart 5: How Carefully Should I Review?
Framework: Review Spectrum (Chapter 14)
When you need this: AI produced output. You need to decide how much time to spend checking it before you use it.
CALIBRATE YOUR REVIEW
═════════════════════
What are the stakes if this output is wrong?
│
├─ Low: Internal, reversible, limited
│ audience (notes to self, brainstorm
│ docs, first drafts for further editing)
│ │
│ └──→ Baseline: SCAN (10-30 seconds)
│ Glance for obvious errors.
│
├─ Medium: Some external visibility,
│ correctable errors (routine client
│ emails, standard reports, internal
│ presentations)
│ │
│ └──→ Baseline: SPOT-CHECK (2-5 minutes)
│ Check 2-3 key claims or numbers.
│
├─ High: External-facing, hard to correct,
│ reputational (published content,
│ proposals, presentations to leadership)
│ │
│ └──→ Baseline: DEEP REVIEW (15-30 minutes)
│ Verify every claim. Check sources.
│
└─ Critical: Regulatory, legal, high-value
decisions (contracts, board materials,
compliance docs, safety-critical)
│
└──→ Baseline: DEEP REVIEW or REWRITE
Use AI output as raw material only.
Rewrite in your own voice.
NOW ADJUST for these factors:
│
Is this a new workflow or task type?
├─ Yes → START AT DEEP REVIEW.
│ New = no reliability data yet.
│
Is AI generally reliable at this specific task?
(Summarizing = reliable. Factual claims = less so.
Math = often wrong. Creative interpretation = variable.)
├─ Less reliable → Move ONE LEVEL DEEPER.
│
Are you rushed, tired, or distracted?
├─ Yes → Move ONE LEVEL DEEPER or WAIT.
│ Pressure makes you miss more, not less.
│
Has this workflow been running well for 30+ days?
├─ Yes → You MAY move ONE LEVEL LIGHTER.
└─ No ──→ Stay at current level.Your review level: _______________
Sustainability check: If your review time is more than 50% of the time AI saved you, something is wrong. Fix the input (Flowchart 3), tighten the task scope (Flowchart 1), or accept that this task isn’t a good AI fit right now.
The trap: I reviewed every AI output word-by-word—72,000 messages a month—until review fatigue set in and I started missing errors. That’s the worst of both worlds. The goal isn’t maximum scrutiny on everything. It’s the right scrutiny on each thing.
- Department leaders: Your team’s review calibration matters more than yours. If you set Deep Review but your team defaults to Scan, you have a process gap. Check in individually after the first 5 uses.
- Individual contributors: Start at Deep Review for your first two weeks on any new workflow. Resist the urge to speed up before you have data.
- Small company CEOs: You’re often the only reviewer. That means your attention is the bottleneck. Protect it by starting with high-FFCC, low-stakes workflows—the ones that can earn their way down to Scan.
- Senior leaders: Your review calibration sets the organizational standard. If you rubber-stamp AI outputs, your team reads that as permission to do the same.
Quick-Reference: The Complete Decision Path
For the wall, the bookmark, or the sticky note on your monitor:
┌──────────────────────────────────────────────┐
│ AI TASK DECISION PATH │
│ │
│ 1. SCORE IT → FFCC: F×F×C×C ≥ 100? │
│ 2. RISK IT → Benefit > Exposure + Risk? │
│ 3. BUILD IT → Trigger→Input→AI Processing │
│ →Human Review→Action │
│ 4. FEED IT → Context + Content + │
│ Constraints + Request │
│ 5. REVIEW IT → Scan / Spot-Check / │
│ Deep Review / Rewrite │
│ 6. SHIP IT → Take the action. │
│ │
│ Still rough? Run it 3 more times │
│ before judging. Fix inputs first, │
│ prompts second. │
└──────────────────────────────────────────────┘