AI for Business Textbook
Find every way your business pays for AI and put a ceiling on each one. Then take one task through a test you can trust, and decide keep, fix or stop from your own numbers.
Map. Test. Price. Protect. Run. In order.
17 Chapters, 2 Appendices
What you will have in 30 days: your true monthly AI cost across every spend path with a cap on each, and one task taken through a known-answer test and a timed both-ways trial, with a keep, fix or stop decision written down.
- The 20-Minute Spend-Path Audit. Guess, count, compare. The 4 paths money leaves by, and the cap arithmetic for a worst week.
- What the evidence actually says. The 2025 support study, the jagged frontier, and why there is no productivity multiplier in this book.
- What an assistant is, mechanically. Why fluency is not correctness, and how to describe any tool by input, output, action and owner.
- The AI Advantage Framework. Map, Test, Price, Protect, Run, each with an instrument.
- MAP: the Decision Map. Input, interpretation, judgment, action, feedback. Size the prize in minutes per week, size the risk by worst-error class.
- TEST: the Known-Answer Test. Material errors defined first, 20 known cases including the messy ones, 5 held back, a timed both-ways batch.
- Working beside it. Who decides the facts, how every promise traces to a source, prompts as tested settings, new staff draft first.
- PRICE: true cost and whole-task value. How AI bills grow, the card reconciliation, the whole-task formula and the keep rule.
- When to turn it off. Pause conditions set before the test, and why assistance never licenses automatic sending.
- PROTECT. Approved sources with owners, drift and retesting, the NIST framework as 4 questions, an off switch 2 people can use, assistants that can act, and the FTC rule on reviews.
- RUN and the 30-Day Sprint. One workflow at a time, week by week.
- Worked examples. 3 labeled composites, including the owner who approved a whole category from one demo.
- The One-Page AI Scorecard.
- Metrics that matter, with your own baseline and no target from us.
- 13 common mistakes, each with the signal and the fix.
- The 30-day and 60-day look-back.
- Proof: what is and is not in this book.
Appendix A: tool categories, dated. Appendix B: using the spreadsheet and templates.
SAMPLE · CHAPTER 1The 20-Minute Spend-Path Audit
Do this before you read anything else. You need your card statements and 20 minutes.
Step 1: Guess first
Write down what you think the business spends on AI each month. Every tool, all in. Do not look anything up. That guess is the first half of your baseline.
Step 2: Count the 4 paths
- Subscriptions. Flat fees and per-seat add-ons. Per-seat bills every login, used or not.
- Metered usage. Per message, per minute, per page, per credit. AI features inside your phone system or help desk often live here.
- Scheduled and automated spend. Anything that runs without a person pressing a button. It bills while you sleep.
- Expensed and personal-card spend. A tool on someone's own card, or a trial that turned paid on a card nobody watches.
Step 3: Compare
Write your guess, what you counted and the gap. The gap is the most useful number on the page. The counted figure is your baseline.
In the book's hypothetical example, a cleaning company owner guesses $60 a month and counts $150 across 5 lines. 3 of the 5 were not on his mental list.
Step 4: Cap every metered and scheduled line
Illustration: an automation charges 1 cent per task and starts re-running itself, 2,000 tasks an hour.
The cap does not stop the mistake. It stops the mistake from becoming a bill.
The question that finishes the audit: is there any way the business can pay for AI that is not on your list?
FROM CHAPTER 6The Known-Answer Test
Before the tool touches a customer, give it cases where you already know the right answer. Count what it gets wrong, and which wrong answers would have cost you.
- Define material errors first: wrong policy, invented fact, wrong recommendation, missed escalation, inaccurate commitment.
- Collect 20 known cases: 12 ordinary, 3 with missing information, 2 with conflicting sources, 2 out of scope, 1 asking for a promise you do not make.
- Hold 5 back. Tune on 15. Run the 5 once, at the end.
- Run 10 live instances both ways, timed. Switch which goes first.
- Decide keep, fix or stop from the log, not from the demo.
94% sounds fine. Missing 3 of 4 serious cases is not. Set your limit for consequential misses before you test.
FROM CHAPTER 8The Whole-Task Formula
Price saved time at the cost per hour of whoever does the work today, not your own rate. And never book saved minutes as revenue: only claim cash when a cash cost changes.
FROM THE TEMPLATESA Tested Prompt
The book teaches how to test a prompt, not a pile of them. Every prompt in the library carries the task it serves, the date it was tested and who may change it.
FROM CHAPTER 14The 4 Numbers That Decide It
Questions
Is this a list of AI tools to buy?
No. Tool names sit in one dated appendix as examples of a category, with no prices. The body teaches how to find what you spend, test one use and decide whether to keep it. Those do not date.
Will it tell me how much time AI saves?
No. Nobody can know that for your business. It gives you the whole-task formula and a timed both-ways test so you measure it yourself.
I have never used AI. Is this too advanced?
Chapter 1 works whether you pay for nothing or for 10 tools. The test in Chapter 6 needs one task you already do and 20 past examples of it.
Does it cover AI that can send emails or take actions?
Yes, in Chapter 10: least privilege, human approval for anything that sends, books, pays or deletes, and treating outside content as data, never instructions. It follows the OWASP Top 10 for LLM Applications, 2025 edition.
Is this legal advice?
No. The data and reviews sections are general information, dated. For health, financial or children's information, the book tells you what to ask your attorney.
Do This Today
Write down your guess at what the business spends on AI each month. Then open last month's card statement and count it. The gap between the two is your first finding.
Then put a cap on every metered line that offers one. The rest of the book can wait until the weekend. The audit takes 20 minutes.
Read it in your library.
This is included with The Standard, along with every other guide in the library. Your monthly re-audit tells you which one to open next. $497 to start, $297/mo to keep it running. Cancel any month and keep everything.
See The Standard →Not sure yet? Take the free Bucket Audit or tell us what's broken.
