Appendix A · Chapter 2 · AI That Rebuilds the Business

The Five-Question Screen

This is the complete instrument from Chapter 2. It answers one question: should your business attempt an AI-first redesign now? Be clear about what passing buys you: not a green light to build, but the qualification to begin the whole-design planning process — the drawing of the complete design that the strategic roadmap (Appendix C) then deploys in managed stages. Run it before a dollar moves, and run it in writing — a screen you keep in your head is a screen you’ve already passed.

Score each question 1 to 5. A score of 3 or below is a fail on that question — fix it before you start, not after. Every failure autopsy in this book has the same shape underneath it: a gap someone could see at the beginning and decided to carry forward instead of close, because closing it felt slower than the alternative. It is never slower than the alternative. Fail two or more questions, and the honest verdict is not yet — not never, not lesser. A failed screen is a sequencing signal, and it points one of two directions. If the gaps are buildable from where you stand — Depth unnamed, Distance unhired, a cushion two patient years from real — rethink the approach and build the missing prerequisite first. If the scores say the business is earlier than that — no cushion and nothing yet generating one, the workflow-level gains still unclaimed — the right sequence is the other end of the spectrum first: work AI That Pays for Itself, then return to this screen when the season changes. Same reader, different season (see the Preface). One exception: Question 2, Distance, is disqualifying on its own. A redesign with no real outside perspective does not proceed until that gap is closed, whatever the other four scores say.

Question 1 — Depth

Can you name, in writing, the specific people who hold the tacit knowledge of how your business actually works — including the undocumented exceptions, workarounds, and judgment calls that never made it into an SOP?

A 5 looks like: named individuals for each core process, and you can say what each one knows that isn’t written down anywhere.
A 3 looks like: you know roughly who the veterans are, but nobody has ever mapped what leaves the building when they do.
A 1 looks like: the SOP binder is your answer.

How you will mis-score this. The wrong Depth nominee is almost always the most technical person — sometimes a senior manager who nominates himself. Depth is not technical skill. It’s exception-knowledge: the person who knows why invoices from that one customer get handled differently, and what the system quietly does wrong on the last day of the quarter. The tell that you have Depth but haven’t captured it: when someone asks for the SOP, the answer is “talk to Denise, she’s the process.”

Question 2 — Distance

Can you actually bring in someone with zero loyalty to your current process and no revenue riding on your answer — a genuine outsider, not a new hire who will blend in within eighteen months?

A 5 looks like: a named person, structurally independent, whose fee is the same whether they say go, go differently, or stop.
A 3 looks like: “our board member asks hard questions.”
A 1 looks like: nobody — or the substitutes below.

How you will mis-score this. The most common substitute I hear is “I’ll ask ChatGPT.” It fails twice. Sycophancy — the tendency to just tell you what you want to hear — is measured and admitted: the training objective itself rewards agreeing with you (Sharma et al., 2023), and OpenAI publicly rolled back a model update in 2025 for exactly this. And even a model that doesn’t flatter you is the same shared tool your competitors are prompting — the instrument that narrows everyone’s ideas toward the same small set of answers cannot also be your source of outside perspective. The second substitute is the internal contrarian — the person whose unofficial job is to disagree. A mere contrarian has no solutions; “that won’t work” people never ask “why can’t you do it this way,” and they carry insider bias anyway. Score a 4 or 5 here only for a real, structurally separate outsider — Appendix F is the vetting checklist.

Question 3 — Financial cushion

Can the business absorb this redesign costing a multiple of the initial budget — not an increment — without threatening payroll, debt covenants, or the core operation?

A 5 looks like: you’ve reserved multiples of your working budget figure, and the worst case is survivable on paper.
A 3 looks like: the plan works if nothing runs over. Plans run over.
A 1 looks like: this bet needs to succeed for the year to work.

How you will mis-score this. By omitting four costs, in the order operators reliably omit them: operating and maintenance, tokens and consumption (the small per-use charges an AI system racks up every time it’s asked to do something), rework, and contingency — what you do if something makes this stop in the future. The demo price is not the price.

Question 4 — Business-first

Does this redesign start from a specific business problem your Depth source can name in one sentence, with a number attached — or from a technology you’ve decided to acquire?

A 5 looks like: “our engagement rework rate is X and it’s costing us Y” — and AI is the tool that makes fixing it survivable to attempt.
A 1 looks like: “we need agents.” A technology-first idea names a technology to buy. A business-first idea names an outcome to change, with a number.

How you will mis-score this. Technology-first ideas arrive wearing business language. “Customers expect a chatbot now” sounds like strategy; it’s an artifact acquisition. Test it: delete the technology’s name from the sentence. If nothing is left, you have your answer.

Question 5 — Time horizon

Are you prepared to treat this as a repeatable capability — with scheduled reviews and re-runs of the loop — rather than a project that ends?

A 5 looks like: you accept that the advantage starts decaying the day it ships, and someone will own watching for it (Chapter 12’s triggers).
A 1 looks like: you’re planning a finish line.

If you’re PE-owned, Horizon isn’t entirely yours to answer. The clock that governs is the fund’s hold period. Check the review cadence above against the fund’s timeline before you score this a 5 — alignment gets verified, not assumed.

The three dimensions you will over-score anyway

Across operators, the same three: ownership and reporting structure (an enthusiast with admin rights is not an owner), SOP quality (a binder written for a 2019 audit describing a workflow two migrations ago), and data readiness (“it’s all in the system” — which is precisely the problem). Three field tests. First, the month-four diagnostic: if this project should be killed in month four, who kills it, and what happens to their bonus? A pause means ownership was never assigned. Second, ask three people who run a process daily to describe the same handoff — three confident, different answers is your real SOP score. Third, request one report — revenue by customer by service line, trailing twelve months — and time it. More than a day, or two conflicting answers, and the data score just announced itself.

Final instruction

Run the screen twice: once alone, once with someone who has no stake in your ambition and nothing to lose by disappointing you. The gap between the two score sheets is information. It is usually the most expensive information on the page.