# Using AI in Business 101 **Get an hour back on the boring stuff - without learning to code, and without the hype.** This free course takes you from "I've typed into a chatbot once" to confidently finishing real business tasks with AI - a customer reply, a quote, a week of social posts - using your own business as the example. It's honest about what AI is good at, where it gets things wrong, and how to keep your data and your reputation safe. By the end you'll have actually done the work, not just watched a demo. **Who this is for** Small-business owners and operators who are not technical - trades, retail, professional services, agencies, hospitality. If you run a real business and you're busy, sceptical, and short on time, this is built for you. No coding. No spreadsheets full of formulas. No prior AI experience assumed. **What you'll be able to do** - Finish a real customer reply, quote, chase-up email and social post using AI, in your own voice. - Write a clear, specific ask that gets a usable answer first time (no "prompt engineering" jargon). - Catch AI when it confidently makes something up - before it reaches a customer. - Know exactly what is safe to paste in and what must never go near a chatbot. - Pick the highest-value first task in *your* business and roughly work out if it's worth it. - Build one reusable template that turns a 15-minute job into a 90-second one. - Write your own one-page "AI house rules" you and your team can actually follow. **Prerequisites** Almost none. A device with a web browser, an email address, and a real task you've been putting off. A free account on any one of ChatGPT, Claude, Gemini or Copilot - we'll set that up in the first module. **Total time:** About 2.5-3 hours total. Designed in short, self-contained lessons (5-10 minutes each) you can do in the gaps between jobs. **Format:** Self-paced video + worksheets, or facilitated live. Every lesson produces one finished thing. You can drop in, do one lesson, and come back next week without losing your place. --- ## Module 0 - Start here (15 minutes) *Goal: get set up, send your first message, and know exactly what you'll walk away with.* ### Lesson 0.1 - What this course is (and what it isn't) "Use AI in your business" is too big and too vague to act on. This lesson shrinks it to something concrete: a handful of real tasks you'll finish today, using the work you already do. **Key points** - You don't need to "understand AI" to get value from it - you need to finish one real task. - This is the free on-ramp. It covers solo chatbot use: drafting, summarising, tidying. It does **not** cover automation, connected tools, or AI that runs unattended. Those belong in the next hands-on workshop. - Honest framing all the way through: what it's great at, where it fails, and how to stay safe. **Try it (5 min):** Write down the one email, message or document you've been avoiding this week. Keep it on hand - you'll use a real one of these in Module 2. ### Lesson 0.2 - Pick a tool and set up a free account The exact tool matters less than the habit when you're starting. For the text-only exercises in this course, ChatGPT, Claude, Gemini or Copilot are all fine choices: you type, it types back. **Key points** - For the core text-only exercises, any of the four is fine. Our live demos may use one tool, but the habit you are learning is tool-agnostic. - Setting up takes about five minutes: an email address and a password. - Start with a free tier. You should not need a paid tier for the core exercises, though limits and features vary by tool. **Try it (5 min):** Create a free account on one tool. Then type this and read the reply aloud: > *"Explain in 5 simple bullet points how you could help someone who runs a [your trade, e.g. plumbing] business. No jargon."* That's it - you've now used AI. The reply proves the tool works and gives you a few ideas to react to. ### Lesson 0.3 - The whole skill in one picture Almost every useful first task follows the same loop. Learn the loop and you've learned most of beginner AI. **Key points** - The loop: **give it your context -> ask in plain words -> check it critically -> edit and use.** - The first answer is a *draft to react to*, never a final. You always do the last 20%. - "Check it" is not optional - it's where you stay the boss. **Try it (3 min):** Take the reply you just got and type back: *"Make that shorter and friendlier."* Watch it change. You've just done the whole loop once. **You'll walk out able to…** open a free AI tool, send a message, get a reply, and improve it - and know exactly what the rest of the course will give you. --- ## Module 1 - What AI actually is, in plain English *Goal: get a correct-enough mental model so your decisions (and your safety habits) make sense.* ### Lesson 1.1 - It's a fast, eager new hire who's never seen your business The most useful picture isn't "robot brain" or "magic" - it's a sharp, brand-new junior who's read everything but knows nothing about *your* business and will confidently guess if you don't tell them. **Key points** - Brilliant at first drafts, tone, summarising, rephrasing. Terrible at knowing your prices, your customers, your local rules. - It will fill gaps with a confident guess rather than say "I don't know." - You'd never let a day-one junior send an email unread. Same rule here. **Try it (5 min):** Ask it to write a short "About us" for your business including *"your three most popular products and your opening hours"* - but **don't tell it any of those facts.** Watch it invent them. This one experiment teaches more than any explanation: it shows you exactly how and when AI makes things up. ### Lesson 1.2 - Why it makes things up (and why that's normal) That invented opening hours moment wasn't a bug or a broken tool. It's how these tools work - and knowing why makes the fix obvious. **Key points** - AI predicts plausible-sounding words. It doesn't "know" facts or look them up unless told to. - Fluent and confident does **not** mean correct. The fluency is the danger. - Do not plan your safety habits around this being fixed. Treat checking as permanent, no matter which tool you use. **Try it (5 min):** Ask for *"5 statistics about [your industry] in [your country], with sources."* Spend a few minutes trying to verify two of them with a quick web search. Count how many were real, vague, or invented. ### Lesson 1.3 - What it's brilliant at vs bad at (the golden rule) A one-page split so you instantly know which jobs to hand over and which to keep. **Key points** - **Great at:** first drafts, rewriting, changing tone, summarising long text, brainstorming, tidying messy notes. - **Bad at / never trust blind:** exact numbers, dates, names, prices, laws, anything with a single right answer. - **The golden rule: it drafts, you decide.** Nothing leaves the building unread. **Try it (5 min):** Make two lists from your week: five tasks that are "mostly typing or reading" (good for AI) and three where a wrong answer costs you money or trouble - tax, legal, final pricing, medical (keep these human). You've just sorted your own work. **You'll walk out able to…** explain in one sentence what AI is, predict when it'll guess, and apply "it drafts, you decide" to anything. --- ## Module 2 - Your first real win: writing things faster *Goal: finish a real customer message with AI, in your voice, in under 15 minutes.* ### Lesson 2.1 - The 4-part recipe for a good ask Beginners think they need a secret skill. They don't - they need to give context and ask plainly. One small recipe, reused a hundred times, beats twenty techniques you'll forget. **Key points** - The recipe: **Role** (who it should act as) -> **Context** (your business, audience, facts, and tone) -> **Task** (what you want done) -> **Format** (how you want it back). Keep it on a card by the till. - Most weak prompts only do the Task part - "write an email" - and leave out the other three. - It's a checklist for clear thinking, not a magic spell. The goal is clarity. **The RCTF card** | Part | What to include | Fast example | |---|---|---| | Role | Who the AI should act as | "Act as my friendly office manager" | | Context | The real situation, audience, facts, and tone | "Regular customer, invoice overdue 14 days, we want to keep the relationship warm" | | Task | The exact job | "Draft a payment reminder email" | | Format | Length, layout, and constraints | "Under 80 words, no legal threats, clear next step" | **Copy-paste template: everyday business ask** ``` Act as [ROLE] for a small [TYPE OF BUSINESS] in [LOCATION]. Context: [WHO THIS IS FOR], [WHAT HAPPENED], [KEY FACTS], and our tone is [TONE]. Task: [WHAT YOU WANT AI TO DO]. Format: [LENGTH / BULLETS / EMAIL / TABLE / OPTIONS]. Before you answer, flag anything important that is missing or risky. ``` **Try it (5 min):** Type the bare version first: *"Write a chase-up email for an overdue invoice."* Note how generic it is. Then the full version: > *"You're my office manager. Write a polite chase-up email for an overdue $480 invoice to a regular customer named Dave who usually pays late but always pays. Keep it under 80 words, friendly, no legal threats."* Compare the two. The gap is the whole skill. ### Lesson 2.2 - Give it your real stuff Vague in, vague out. The single biggest upgrade is pasting in your actual specifics - the real message, the real price, the real name. **Key points** - Paste the actual email you received, the actual review, the actual rough notes (names removed if sensitive - see Module 4). - The AI isn't psychic. It can only work with what you give it. - If your prompt feels thin, tell it: *"Before you write it, ask me the 3 things you most need to know."* Then answer. **Try it (10 min):** Open a genuinely awkward customer email you've received (remove the customer's name). Paste it in and ask: *"Draft a calm, professional reply. Friendly but firm. Under 120 words."* Read it, change one thing so it sounds like you, and you've rescued a real task. ### Lesson 2.3 - Fix it by talking, don't start over The most common beginner failure is one mediocre try, then "AI doesn't work for my business." The conversation *is* the tool. **Key points** - Improve an answer with plain follow-ups: "shorter", "friendlier", "more formal", "add a line about our Saturday hours". - You never need to retype the whole thing - just say what's wrong. - Three small nudges usually beat one perfect first prompt. **Try it (5 min):** Take any answer above and improve it three ways in a row: *"Make it shorter"* -> *"Make it warmer"* -> *"Add a clear next step."* Proves you don't need special words - you just keep talking. **You'll walk out able to…** turn a real customer message you'd been dreading into a finished, on-brand reply in minutes. --- ## Module 3 - Making AI sound like YOU *Goal: stop the generic-robot output that makes owners quit, and build your first reusable asset.* ### Lesson 3.1 - Why the first draft sounds generic, and the fix Generic output is the number-one reason owners abandon AI. The fix is simple: give it your voice to copy. **Key points** - A cold draft sounds like a robot because it has nothing of *you* in it. - Feed it two of your own past emails and say *"write the new one in this same style."* - Write one reusable **"about my business"** paragraph - your name, what you do, your town, your tone - and paste it at the top of asks. **Try it (10 min):** Paste two of your own past emails, then ask for a new one *"in the same voice and style as these."* Compare it with a cold generic version side by side. The difference sells the habit. ### Lesson 3.2 - Your first reusable asset is a saved prompt Forget Zapier-style wiring. The honest entry point to "automation" is a reusable template - and it solves a problem every owner recognises: retyping the same five messages every week. **Key points** - Build a template once with `[BLANKS]` you fill in: name, amount, date, job. - Turn any good answer into one: *"Save this as a reusable version with blanks for the bits that change."* - Keep templates where you'll actually find them - a pinned note, a saved doc, a notes app. - Be honest: this is a saved prompt, not a real automation tool. Connected workflow tools are beyond this 101. **Try it (10 min):** Take your overdue-invoice email from Module 2 and ask: *"Rewrite this as a reusable template with blanks for the customer name, the amount, and the due date."* Save it. You now own a copy-paste asset that turns a 15-minute job into a 90-second one. **You'll walk out able to…** make AI write in your voice and keep one reusable template you'll open again on Monday. --- ## Module 4 - What's safe to type in (and what's not) *Goal: handle real business data without the privacy lecture - one rule, one list, one habit.* ### Lesson 4.1 - The three buckets and the never-paste list Data safety is the easiest thing to get wrong on day one and the most important to get right. You don't need privacy law - you need three buckets and a short list. **Key points** - **Public** (safe to paste): your website copy, a published job ad, a general question. - **Internal** (be careful): draft pricing, rough internal notes. - **Sensitive** (never paste into a general chatbot): customer details, card/payment data, passwords, tax identification numbers, health info, staff/HR records, signed contracts. - The **postcard / front-page test:** if you'd be uncomfortable seeing this text on the front page of the paper or in a competitor's inbox, don't paste it. **Try it (10 min):** Look at three real things on your desk (a customer's address, a supplier quote, a staff member's pay detail) and sort each out loud into Public / Internal / Sensitive. Builds the instinct as a quick judgement, not a policy to memorise. ### Lesson 4.2 - Redact and still get a great answer You almost never need the sensitive bits to get a useful draft. Swap them out and carry on. **Key points** - Replace names with "Customer A", amounts with "$X", addresses with "their suburb". - For drafting, summarising and brainstorming, redacted context works just as well. - Settings and policies vary by tool and plan. Before using business information, check the tool's data controls and whether your chats may be used to improve the service or reviewed. Even with stronger settings, sensitive data still should not go in. **Try it (5 min):** Take a real customer email you'd never paste raw. Redact the name, amount and address, paste the cleaned version, and ask for a reply. Confirm you got an equally useful answer with zero sensitive data exposed. Then go find the privacy/data-controls setting in your tool. **You'll walk out able to…** decide in seconds what's safe to paste, redact the rest, and still get a usable answer. --- ## Module 5 - Is it true? The 10-second safety check *Goal: build the verification reflex that turns "I played with a chatbot" into "I use this safely on real work."* ### Lesson 5.1 - The five things to always double-check Non-technical users read confidence as competence - that's the expensive trap. A short, scoped check fixes it without making you fact-check everything obsessively. **Key points** - Always verify: **numbers, dates, names, laws/rules, and quotes/sources.** And anything going to a customer. - Low-stakes brainstorming doesn't need a fact-check - scope the habit so you'll actually keep it. - The one question before you hit send: *"Did it invent any fact, price, name, date or promise I didn't give it?"* If yes, fix or cut. **Try it (5 min):** Ask the AI for something you already know exactly - your own opening hours, a specific local council rule, your own pricing. Watch it answer confidently, sometimes wrongly. Seeing it fail on something you can verify lands the lesson permanently. ### Lesson 5.2 - Human in the loop, and what stays human forever Some things AI should never do unread, and some it shouldn't do at all. **Key points** - Nothing reaches a customer, supplier or regulator unread. That's the "human in the loop" rule. - Stays human forever: pricing calls, hiring decisions, legal/medical/tax advice, anything you'd be liable for. - You own the output. "The computer said so" is not a defence if AI invents a refund policy or defames a competitor under your name. - Don't fake it: AI-written "reviews" and testimonials are a reputational and legal landmine. **Try it (5 min):** List three tasks where a wrong answer would cost you money or get you in trouble. Mark each "human-checked, never auto-send." That's your personal red line. **You'll walk out able to…** run a 10-second check before sending, and know which decisions never leave your hands. --- ## Module 6 - Where is AI worth it? Finding (and proving) your first use case *Goal: pick one high-value, low-risk task and roughly prove it's worth your time.* ### Lesson 6.1 - Start from your annoyances, not from AI The best first use case isn't flashy - it's the boring thing you already moan about every week. **Key points** - Pick by **frequency x annoyance x low blast-radius.** High-frequency + high-pain + harmless-if-wrong = start here. - Good first jobs: the email you write 30 times a month, review replies, quote follow-ups, tidying notes. - Bad first jobs: "handle my tax filing", final pricing, anything where a wrong answer reaches a regulator or customer uncaught. **Try it (10 min):** The Moan List + Grid. Write 8 tasks you grumble about weekly. Circle the 3 that are mostly typing or reading. Draw a 2x2 (how often x how painful) and plot them. Top-right is your first AI job. **Quick decision table** | Task | Frequent? | Low-risk? | Easy to check? | Good starter? | |---|---|---|---|---| | Social captions from rough notes | Yes | Yes | Yes | Yes | | Reply to a customer review or enquiry | Yes | Mostly | Yes | Yes | | Summarise a long email into actions | Yes | Yes | Yes | Yes | | Quote cover note where you supply the prices | Yes | Medium | Yes | Yes, with checking | | Tax, legal, medical, or final pricing decision | Usually no | No | No | No | | Signed contract, HR record, health record, or customer payment data | No | No | No | No | ### Lesson 6.2 - Will it pay off? A no-maths business case Owners need to justify their time and any subscription without a spreadsheet, and with the checking time counted honestly. **Key points** - Rough annual time back: **minutes saved x times per week x ~50 weeks**. A 20-minute job done 10x/week is over 150 hours a year. - The dead-simple ROI sum: **(hours saved per week x what your time is worth per hour) - tool cost per week = weekly payback.** If it is positive and you would keep doing the task this way, it is working. - Count the whole task, including checking, fixing, and learning. If you time only the AI draft and forget the human review, your ROI is fake. - A $20-30/month tool only needs to save an hour or two a month to pay for itself, but only chase a tool after you have found a painful task, never before. **Try it (15 min):** The Stopwatch Test. Pick one task. Do it your normal way and time it. Then draft it with AI, edit, and time that. Write both numbers down. One real before/after beats any statistic, and it is honest about the editing time. **Simple tracking worksheet** | Task | Time before | Time after, including review | Weekly saving | Keep, tweak, or drop? | |---|---:|---:|---:|---| | | | | | | **Kill rule:** finish this sentence before you pay for a tool: "If by [date, about 3 weeks out] this has not saved me at least [X minutes/week or $Y], I will stop or try a different task." **You'll walk out able to…** name your single best first use case and put a believable number on it. --- ## Module 7 - Worked examples by business type *Goal: see your own industry in at least one concrete example, then copy it.* These are self-contained 10-minute demos. Pick the one or two closest to your business. Each works in the chat account you set up in Module 0, with no extra tools, plugins, or paid tiers. ### Lesson 7.1 - Trades, retail & hospitality **Key points** - **Trades (plumber/electrician):** dictate a 30-second voice note about a job, paste the text, ask for a clean itemised quote and a follow-up - then check it didn't invent a price. - **Retail/e-commerce:** paste one real review, get three on-brand reply options in your voice; write product descriptions from a one-line offer. - **Cafe/hospitality:** a week of social posts for your actual seasonal special, plus a polite supplier email, in one sitting. **Try it (10 min):** Pick whichever fits. For trades: dictate a messy job update to your phone, paste it, and ask: *"Turn this into a short, polite customer email with a clear itemised quote. Flag anything you're unsure about so I can fill it in."* ### Lesson 7.2 - Professional services, agencies & the inbox **Key points** - **Accountant/lawyer/consultant:** plain-English summaries of long documents - *"what they want, what I need to do, by when."* - **Agency/marketing:** first-draft proposals and client recap emails you then make yours. - **Everyone:** turn a wall of customer feedback into the top 3 recurring themes, then judge whether you agree. **Try it (10 min):** Paste a long, rambling email thread or 8-10 real reviews and ask: *"Give me what they want, what I need to do, and by when, as three short bullet lists"* (or *"the top 3 recurring themes"*). Check it against the original. **You'll walk out able to…** point to at least one worked example in your own trade and reproduce it. --- ## Module 8 - Your safe first week, and what's next *Goal: leave with a bounded plan, your own rules, and a clear next step.* ### Lesson 8.1 - Your one-page AI house rules Most small businesses have nothing written down, so staff either avoid AI or use it recklessly. A fridge-sized page fixes more than any feature. (Call them "house rules" - not "policy", "compliance" or "governance.") **Key points** - One page, max. Anything longer and nobody reads it. - It contains: approved tools, approved uses, the never-paste list, the "always check before sending to a customer" rule, and **a named person to ask when unsure.** - The owner goes first - modelling use makes it safe for the team to copy. Frame it as "removes the dull work everyone hates," never "replaces people," or staff will quietly sabotage it. **Try it (15 min):** Fill in the blanks: *Approved tools: ___. We use AI for: ___. We never paste: ___. Before anything goes to a customer we: ___. When unsure, ask: ___.* You leave with a usable artefact, not a concept. ### Lesson 8.2 - Your week-one plan and personal cheat-sheet Confidence comes from a small, bounded plan, not a feature tour. **Key points** - Week-one plan: one task a day you already do. One tool, one use case, one daily one-line "win log." - Your cheat-sheet: your top 3 repeated tasks and the exact ask you'd type for each. - Spotting when something's gone wrong, and undoing it calmly - it's a draft, nothing's been sent without you. **Try it (10 min):** Build your cheat-sheet. Write your three most-repeated tasks and, for each, the exact prompt (Role-Context-Task-Format) you'd paste. This is the thing that makes you open AI again on Monday. ### Lesson 8.3 - The ceiling of solo use, and what's beyond it You've now hit the natural limit of doing this one-message-at-a-time in a chat window. That ceiling is exactly what the next step removes. **Key points** - There's a ladder: **Chat** (you type, it answers) -> **Cowork** (you hand it your files - a PDF quote, a sales spreadsheet, a contract - and work on them together) -> **Agents** (you train it once on a repeatable job and it runs the steps, like a staff member following a checklist). - "Agent" = a trained staff member following a checklist - not a real-estate agent. Always picture the checklist. - This 101 is the bottom rung. You only climb when the lower rung gets boring. - Beyond this 101, the next rung is working alongside AI on your own files, training repeatable agent-like workflows, connecting tools, and building a measured business case. **Try it (10 min):** Rung self-sort. Write 6 real tasks from your week. Place each on Chat / Cowork / Agent. The Cowork and Agent ones are your map of what's next. **You'll walk out able to…** run your first week safely, hand the team one page of rules, and see exactly what the next workshop adds. --- ## Capstone exercise - Your business, end to end Do this on a real task from your own business. Allow 30-45 minutes. 1. **Pick it.** From your Moan List, choose your top-right task (high-frequency, high-annoyance, low-risk). 2. **Stopwatch the old way.** Do it your normal way once and write down the minutes. 3. **Write the ask.** Use Role-Context-Task-Format. Paste your real (redacted) specifics. 4. **Check it.** Run the 10-second check: any invented number, name, date, price or promise? Fix or cut. 5. **Make it yours.** Two follow-ups to match your voice. Stopwatch this way too. 6. **Save the asset.** Turn it into a reusable template with `[BLANKS]`. 7. **Decide.** Hours saved per year (minutes x times/week x 50)? Keep, tweak, or drop - and add it to your one-page house rules. **You finish with:** one completed real task, a before/after time, a reusable template, and a decision. That's a genuine business win, not a demo. --- ## Knowledge check Answer these to confirm you have the essentials. Aim for 9/11 or better. The answer key follows, so learners can self-mark without a facilitator. 1. **Multiple choice:** When generative AI answers you, what is it actually doing? A) Looking up verified facts in a database B) Predicting likely next words from patterns it learned C) Searching your business records D) Asking a human to write the answer 2. **True/False:** The AI already knows your business's prices, customers, bookings, and policies. 3. **Short answer:** What is a hallucination, and what should you do whenever AI gives you a fact, figure, source, price, date, or quote? 4. **Multiple choice:** Which is the safest first task to try with AI? A) Deciding how to handle a tax question B) Drafting five social media captions from your notes C) Signing a supplier contract D) Setting your final prices for the year 5. **True/False:** It is fine to let AI invent a price on a customer quote as long as it sounds reasonable. 6. **Multiple choice:** Which is safe to paste into a free consumer AI tool? A) A customer's full name and credit card number B) A patient's national health ID number C) A de-identified email with personal details removed D) Your staff payroll file 7. **Short answer:** Name the four parts of the RCTF prompting formula, and say why Context matters. 8. **Short answer:** Before you start using AI for a task, what one thing should you measure so you can later prove it saved time? 9. **Short answer:** Sarah pays $40/month for an AI tool but only opens it once a month. Has she achieved adoption? Why or why not? 10. **Practical task:** Using the RCTF formula, write one prompt asking AI to draft a friendly thank-you message to a customer for your business. Then list two things you would check before sending it. 11. **Short answer:** What five things belong in your one-page AI house rules? ### Answer key 1. **B.** It predicts likely next words. It is not automatically looking up verified facts. 2. **False.** It only knows what you tell it in the chat. It has no access to your prices, customer list, bookings, or files unless you provide them. 3. A hallucination is a confident but made-up or incorrect answer, such as a fake stat, wrong price, invented source, or false policy. Verify important claims against a trusted source before acting. 4. **B.** Drafting captions is frequent, low-risk, and easy to check. The others are high-stakes and harder to verify. 5. **False.** You must supply real prices and verify all numbers before anything reaches a customer. 6. **C.** Use the de-identified version. Never paste names plus card numbers, national health ID numbers, health records, contracts, or payroll files into a free consumer tool. 7. **Role, Context, Task, Format.** Context matters because AI cannot know your specific business, customer, tone, facts, or constraints unless you provide them. 8. Take a baseline: measure how long the task takes today, or how many times you do it, before using AI. 9. No. She has bought the tool, not adopted it. Adoption means it is used regularly in real work without being forced. 10. Self-marked. A strong answer includes all four RCTF parts, uses real but safe business context, and checks at least two of: customer name/details, tone, invented facts, price/promise accuracy, and whether private information was pasted unnecessarily. 11. Approved tools, approved uses, the never-paste list, the always-check-before-sending rule, and a named person to ask when unsure. --- ## Your first 7 days A small, low-pressure plan. Allow about 15 minutes a day. - **Day 1: Sign up.** Pick one free tool, create your account, and type one real task. Just get past the empty box. - **Day 2: Feel the limits.** Run the invented-opening-hours or industry-stats test. Write your first never-paste list. - **Day 3: Learn to ask.** Do the bad-prompt vs RCTF-prompt exercise on a real task. Start your prompt swipe file. - **Day 4: Pick your first win.** List three boring weekly tasks, circle one, and write down the baseline. - **Day 5: Run the win.** Do that task with AI. Time the whole job, including your check and edit. Use the result for real if it passes your checks. - **Day 6: Do the maths.** Compare before vs after. Use the ROI sum and write a kill rule. - **Day 7: Make it repeatable.** Turn the prompt into a saved template or one-page playbook. If you have a team, show one trusted person. After seven days you should have one real, measured win and one repeatable process. That is the bridge into the next hands-on workshop. --- ## Keep going - **Do, don't just read.** This week, run your week-one plan: one task a day you already do, one line in your win log each time. - **Keep your three artefacts handy:** your one-page house rules, your reusable template, and your personal cheat-sheet. - **Re-run the experiments that stuck:** the "invent my opening hours" hallucination test and the "5 stats with sources" verification hunt - they keep your checking reflex sharp. - **Explore the data setting** in whichever tool you use: find the "don't train on my data" control and decide your default. - **Your natural next step - the hands-on workshop "Put AI to Work in Your Business" ($600/seat).** You've hit the ceiling of solo chatbot use. The workshop is where you climb the ladder: working alongside AI on your own files (Cowork), training agents to run repeatable jobs, connecting tools so the boring work runs itself, real SMB ROI case studies, and a repeatable consulting method. Bring the use case you picked in Module 6 - you'll build it out there. --- ## Legal-sensitive notes for production This course intentionally teaches safe habits, but final production should get a focused legal review before publication. Do not position the notes below as legal advice. - **Privacy and personal information:** the course mentions customer details, health information, national health ID numbers, payroll files, tool data controls, and free consumer AI tools. Confirm the final wording against the target country and privacy regime before launch. - **Tax, legal, medical, HR, and regulated advice:** the course correctly keeps these human-reviewed, but examples such as tax filing, contracts, hiring, and health records should be checked for jurisdiction-specific wording. - **Copyright, ownership, testimonials, and disclosure:** keep the advice plain and conservative. Avoid implying that AI output is always owned by the learner or always safe to publish unchanged. - **Workshop pricing and commercial claims:** the $600/seat workshop bridge should be confirmed by Marketing before production. --- ## Facilitator / instructor notes **Live delivery (half-day, ~3 hours)** - **Everyone does, nobody just watches.** The course only transfers skill and confidence when each person completes a task on *their own* business. Protect hands-on time ruthlessly; cut your talking before you cut their doing. - **Pre-flight in Module 0.** Get every account created and a first message sent before you move on - a stranded learner at minute five is lost for the session. Have a backup tool ready in case one is down. - **Keep demos vendor-neutral in wording, live in Claude.** Say "this works in ChatGPT, Claude, Gemini or Copilot" so nobody feels locked out, even as you walk through Claude on screen. Never require logins, paid tiers, plugins or connected files mid-session. - **Run the two "shock" demos live and together:** the invented-opening-hours hallucination (Module 1) and finding a fabricated source (Module 5). Seeing it fail on something verifiable does more than any slide. - **Module 7 is a choose-your-own.** Don't run all six examples for everyone; let each table pick the one or two closest to their trade and circulate. - **Honesty is the trust-builder, not a weakness.** Lean into the limits. An owner who hears "it never makes mistakes, your data's fine" and then gets burned won't come back - or buy the workshop. - **Name the ceiling at the end.** The on-ramp only works if learners feel what solo chatbot use *can't* do. End on the ladder, with the workshop as the obvious next rung - a clear pointer, not a hard sell. **Self-paced delivery** - Lessons are short (5-10 min) and self-contained - no long dependency chains. Tell learners they can do one in a 15-minute gap and resume next week. - Put the "Try it" activity front and centre in each lesson; success is *completing* it, not watching the video. - Replace live shock demos with a "do this now and screenshot the result" prompt so the hallucination and verification lessons still land by experience. - Provide the three fill-in artefacts as downloadable one-pagers: house rules, reusable template, and cheat-sheet. **AU / local variant notes** - For Australian delivery, examples can use TFN, BAS, GST, Medicare, OAIC, and local small-business regulator wording. - For other markets, swap those examples for the local tax ID, health ID, privacy regulator, and business-advice body. - Keep the main course market-neutral; put local specifics in slides, facilitator notes, or downloadable variant worksheets. **Common pitfalls to coach against (both modes)** - Don't dismiss the fears to seem upbeat - make each fear concrete and give a containing rule. - Don't teach "prompt engineering" as a jargon discipline - one plain recipe, reused. - Don't pick flashy first use cases (AI runs your ads/website/live chat) - high blast-radius, high disappointment. - Don't bury safety at the end - it's threaded through Modules 1, 4 and 5 on purpose. - Don't overpromise the workshop's automation in the 101 - be explicit about what's solo-chatbot now vs what comes next.