Reading-Group Facilitation Guide

For TEC Canada / YPO / EO / Vistage chapters and internal executive teams reading AI That Pays for Itself together.

What this is for. A peer group of 5–15 operators gets far more out of this book reading it together than alone, because the book's whole argument is that the hard part isn't the technology — it's the operations, the ownership, and the honest conversation nobody's having internally. A reading group forces that conversation. This guide turns the book into six 90-minute sessions (compressible to four — see the note at the end), each with the chapters to read, discussion questions, an exercise, and a between-session assignment that makes the next session real instead of theoretical.

How to run it. One facilitator (a chair, a moderator, or the most senior operator). Everyone reads the assigned chapters before the session. The discussion questions are conversation starters, not a quiz — pick three or four that fit the room. The exercise is the point; protect time for it. The between-session assignment is what separates a book club from a working group: each member does one concrete thing on their own firm before the next session, and the next session opens with what they found.

No prep tools required, but they help. The free companion templates ([URL]/vault) map directly onto the exercises. Members can pull the relevant worksheet before each session.

Session 1 — The Reality and the Boring-AI Thesis

Read: Chapters 1–3 (Why Most AI Projects Fail · The Boring-AI Thesis · Operations First, Not Tool First).

Discussion questions: 1. Everyone names one AI project — yours or one you watched up close — that didn't pay off. Was the failure technological, or was it one of the operational causes in Chapter 1 (no agreed metric, no data readiness, lost sponsorship, no owner)? Be honest about which. 2. The book argues the AI that pays is the boring kind. Where in your firm is the genuinely boring, high-volume, repetitive work that nobody brags about — and is anyone looking at it for AI, or is everyone chasing the impressive thing? 3. "Automating a bad process makes bad things happen faster." Where have you seen a tool laid on top of a broken process? What happened? 4. Chapter 1's three success factors — pre-approved metric, data readiness, sustained sponsorship — which of the three is your firm weakest on, and why that one? 5. If you had to pick one workflow in your business as the "boring first project," what would it be, and what's stopping you from starting?

Exercise (25 min): Each member writes their single worst workflow on a card — the one that wakes them up, the one everyone complains about. Then, in pairs, each person tries to talk their partner out of automating it ("fix the process first") and the partner defends it. The point is to feel the operations-first instinct, not to win.

Before next session: Run the one-page Process Audit ([URL]/process-audit) on that worst workflow. Map where the work actually goes and write down what "right" looks like in numbers. Bring the filled-in sheet.

Session 2 — The Method: The Operating Stack

Read: Chapters 6–10 (The Operating Stack · Assess · Illuminate · Accelerate · Sustain).

Discussion questions: 1. The Operating Stack is five layers, and it's a stack because the layers depend on each other. Which layer is your firm's weakest — Foundations, Process, Governance, Architecture, or Cadence? (The book bets most of you will say Layer 1 or Layer 5. Did you?) 2. Assess says: when "we already started building something else" collides with what the assessment found, the assessment wins. Have you ever been in that fight? Who won — the evidence, or the sunk cost? 3. Illuminate says write the kill criterion first, before the metric, before the scope. Could you write a specific, numeric "we'd stop here" condition for your top use case right now? If not, what does that tell you? 4. Sustain is the chapter everyone skips. Be honest — for any AI or automation you've already deployed, whose job is it to re-check it quarterly? Does anyone actually do it? 5. The 90-day cadence depends on setting the redesign date, not just the deploy date. Why do you think that's the date most firms never put on the calendar?

Exercise (30 min): As a group, take one member's Process Audit from last session and run the front half of the stack on it out loud: Assess it (is the data ready? is it high-volume and repeatable?), then draft an Illuminate pilot scope — the metric, the kill criterion, the 90-day dates. The owner leaves with a real pilot sketch.

Before next session: Complete the Data Readiness Audit scorecard ([URL]/data-readiness-audit) for your candidate workflow. Bring your score and the one thing it surprised you about.

Session 3 — How AI Actually Works (without the hype)

Read: Chapters 11–14 (How Agents Actually Work · Multi-Agent Orchestration & HITL · Memory, Tools, and Skills · The Plumbing). Skim 15–17 if the group is technical.

Discussion questions: 1. An agent is "a model given tools, memory, and the freedom to decide its next step." Did that demystify anything you'd been treating as magic? Where had a vendor made it sound more complicated than it is? 2. The book says start with the simplest pattern that works and make the system earn every step up to multi-agent. Where in your firm has someone reached for the complicated version of something that a simple one would have handled? 3. Human-in-the-loop gates belong at every irreversible action. What are your firm's three most irreversible AI-touchable actions — the ones where being wrong is permanent or expensive? 4. The idempotency example (the agent that sends the invoice twice because the network hiccupped) — the model did nothing wrong, the tool was built wrong. How much of your AI risk is actually software-engineering risk that nobody's framing that way? 5. "Is it MCP-compatible? A2A-aware? Durable?" — would your last vendor have been able to answer those? Should those be in your next RFP?

Exercise (25 min): Each member maps their three most irreversible AI actions and the human gate each one needs (the Chapter 12 worksheet, [URL]/hitl-gates). Share one around the room — the goal is to hear how differently each firm's "irreversible" looks.

Before next session: Add the three RFP questions (MCP-compatible? A2A-aware? Durable?) to your actual procurement checklist, or draft them if you don't have one. Bring it.

Session 4 — Governance That Holds Up

Read: Chapters 18–20 (NIST AI RMF-Lite & the 12 Risks · ISO 42001, EU AI Act & US State Law · OWASP, Red-Teaming & Shadow AI).

Discussion questions: 1. The book says "governance" lands like a tax on a mid-market CEO, so firms do nothing — then wire AI into hiring or invoice approvals with no guardrails. Is that your firm? What's wired in already that nobody governed? 2. The 12 named failure modes of generative AI — pick the one that scares you most for your business. What's your current control for it? (If the answer is "none," that's the useful finding.) 3. Which AI laws actually touch your firm — by where your employees, customers, and data sit? Be specific. Most rooms discover they're more exposed than they assumed (especially anyone with California or EU customers). 4. Shadow AI is governed by giving people a sanctioned alternative, not a ban. What are your people already using that you haven't sanctioned — and would a ban work, or just push it underground? 5. Whose job is AI governance at your firm right now? If the honest answer is "nobody's," what's the smallest credible version you could stand up this quarter?

Exercise (30 min): As a group, fill in the one-page NIST-Lite Govern/Map/Measure/Manage card ([URL]/nist-lite-card) for one member's highest-stakes use case — the one where a bad output costs a customer, an employee, or a lawsuit. Then each member commits to filling one in for their own top use case.

Before next session: Run the one-page jurisdiction checklist ([URL]/jurisdiction-checklist) and the AI Acceptable-Use Policy template ([URL]/aup-template). Bring which laws touch you and whether you have a sanctioned tool named in your policy.

Session 5 — The Operating-Model Reset

Read: Chapters 21–25 (Workflow Redesign Is the EBIT Driver · AI Economics & TokenOps · Management Is the AI Superpower · Org Design · Change Management).

Discussion questions: 1. "The question is never which tasks can AI do — it's how must this workflow operate differently to create value." Take one workflow: what would actually change about how it runs, not just which steps get automated? 2. The $50 demo / $8,000-a-month production gap — has a vendor's demo price ever set an expectation that production blew up? How would you cost a use case differently after Chapter 22? 3. AI amplifies your most experienced people, not your best prompt-typers (the bottom-half-of-performers-gained-most finding). Who on your team would AI amplify most — and are you pointing it at them? 4. The book says most mid-market firms don't need a $300K Head of AI — they need a fractional vCAIO, a redesign owner, and an engineer (or a partner who supplies one). What's your honest read on what your firm needs, and what's the most expensive hiring mistake you could make here? 5. Adoption is won by a champion network and a 6-week cohort, not classroom training, and the success metric is process compliance, not attendance. Who are your three potential champions?

Exercise (30 min): Each member runs the "build the team or borrow it" decision rubric ([URL]/build-or-borrow) on their own next AI move, out loud, with the group pushing back. The room is unusually good at calling out a hire that's really a partner engagement in disguise, or vice versa.

Before next session: Pick one expert and one task to amplify this month ([URL]/latent-expertise), and name your first three champions. Bring both.

Session 6 — Make It Real, and What's Next

Read: Chapters 35–37 (If You Run a Function + CEO-Intro Toolkit · If You Lead IT · The First 90 Days). Members read the role chapter that's theirs most closely. Skim the vertical chapter (26–31) that matches your industry.

Discussion questions: 1. After five sessions, what's the single first project each of you is actually going to run? Say it out loud and name the owner — this is the commitment moment. 2. For the functional leaders: the hard part isn't the use case, it's getting your CEO to yes. What's your version of the one-page memo, and what would make your CFO say yes? 3. For the IT leaders: where's the biggest gap between your current infrastructure and where it needs to be? (The book bets it's the model gateway and the observability layer.) Is that gap smaller than you'd assumed? 4. The First 90 Days chapter says the thing that makes a program work is someone owning it with enough stubbornness to get through the boring phases. Who is that person at your firm? Is it you? 5. The board's five slides — if you had to brief your board on AI next quarter, what goes on them? What would you be embarrassed to not be able to answer?

Exercise (35 min): Each member presents a 90-second "first 90 days" plan for their firm: the one workflow, the owner, the metric, the kill criterion, and the first re-validation date. The group's job is to find the weakest layer in each plan — and there's always one.

Where this points. If your group's discussions kept surfacing "we know what to do, we're just not sure how to start" — that's the signal to use the tools, not just talk about them. Each member takes the AI Readiness Assessment ([URL]/assessment) individually; it'll tell each firm its weakest layer and which boring project would ship first. And for any member who scores 7+ on the vCAIO-fit rubric in the back of the book and wants 30 minutes on it, the calendar's at [URL]/call — with the no-sell promise: if it's not a fit, you'll hear that in the first ten minutes, and a better path with it.

Running it in four sessions instead of six

If the group only has four meetings, combine like this: - Session 1: Parts I–II (ch 1–5) — the reality and who it's for. - Session 2: Part III (ch 6–10) — the Operating Stack. (Don't compress this one further; it's the spine.) - Session 3: Parts IV–V (ch 11–20) — how it works + governance. Technical members go deep on IV; everyone does the governance exercise. - Session 4: Parts VI + IX (ch 21–25, 35–37) — the operating-model reset and make-it-real, ending on the same commitment exercise and the assessment.

Keep the between-session assignments in all four — they're the difference between a discussion and a decision.