AI for Business
Course 2026

Let's start honestly, because most AI advice aimed at small business owners is written by people selling something. I use AI every day across my marketing work and my real estate business. It has genuinely changed how much I get done. It has also cost me time on things it was never going to do well.

Here's the mental model that matters: today's AI is very good at producing a first draft of something a competent person could produce — and very bad at knowing whether that draft is true. Everything else follows from that one sentence.

What it genuinely does well

The honest limits

AI models generate plausible text. Plausible is not the same as correct. It will confidently give you a wrong GST rate, a court case that doesn't exist, or a statistic it made up — in a fluent, confident tone. This isn't a bug you can prompt your way out of; it's how the technology works. We cover it properly in Module 9.

It also doesn't know your business. It has never met your customers, doesn't know your pricing, and can't see your calendar unless you connect it. Most disappointing AI results come from a person expecting mind-reading and providing three words of context.

What it costs

You can do everything in Modules 2, 3, 6, 7 and 8 on free tiers. Paid plans for the main assistants sit around NZ$30–40/month per person and are worth it once you're using AI daily. Image and video generation is where costs appear — that's Module 5, and there's a cheaper path than you think.

• Write down the 3 tasks that eat the most of your week
• Mark which of them are "producing a first draft of something"
• Mark which involve facts that must be exactly right
• Start with the first group — leave the second for later modules

Skip the strategy for a moment. These five tasks are low-risk, high-frequency, and you can do all of them today on a free account. Nothing here is customer-facing without you reading it first.

Win 1 — The email you've been putting off

Everyone has one: the awkward reply, the follow-up to a quote, the chase for an overdue invoice. Give the AI the situation and the tone you want, not just "write an email".

You're helping me write a short, professional email. I run [type of business] in [city].

Situation: [what happened, in plain words]
Who I'm writing to: [customer / supplier / staff member]
What I need from them: [the specific outcome]
Tone: polite but direct, no corporate waffle
Length: under 150 words, NZ English

Write two versions — one warmer, one firmer.

Ask for two versions every time. Choosing between options is faster than editing one draft, and you'll quickly learn which register suits you.

Win 2 — Summarise the thing you don't want to read

Paste in a long email chain, a supplier contract, a council document, a 20-page report. Ask for: a three-sentence summary, the decisions required of you, the deadlines, and anything that looks unusual. Then still skim the original for anything money-related — the summary is a map, not a substitute.

Win 3 — Job ads and role descriptions

Hiring copy is a genuine AI sweet spot: formulaic, time-consuming, and low risk. Give it the role, the must-haves, the nice-to-haves, your pay range and something honest about what the job's actually like. The result will be better than most agency-written ads because you supplied the specifics.

Win 4 — Your FAQ page, finally written

Open your inbox and copy out the last 20 customer questions. Paste them in and ask the AI to group them into themes and draft an answer for each in your voice. You'll have an FAQ page in 30 minutes that would otherwise never get written. Fix every price, timeframe and policy detail yourself — those are the exact things it will guess at.

Win 5 — Meeting notes that survive the week

Record the meeting (with everyone's agreement — say so out loud at the start), or just type rough notes as you go. Then:

Turn these rough notes into a clean summary.

Output exactly these sections:
1. Decisions made
2. Action items — who, what, by when
3. Open questions still unresolved
4. Anything mentioned that needs a follow-up email

Flag with [UNCLEAR] anything you had to guess at. Don't invent detail.

Notes:
[paste your notes]

That "[UNCLEAR]" instruction is the single most useful line in this whole course. Use it everywhere.

• Pick one AI assistant and stick with it for two weeks
• Do the difficult email today
• Draft your FAQ from real inbox questions
• Use [UNCLEAR] flagging on your next meeting notes
• Note roughly how long each task took before and after

"Prompt engineering" sounds technical. It isn't. It's briefing — the same skill as briefing a new staff member who is fast, well-read, eager to please, and has no idea what your business does.

Bad results almost always trace back to a thin brief. Compare:

The four ingredients

Context — who you are, what you sell, who the audience is, any facts it must use. Role — "you're an experienced NZ hiring manager", "you're a sceptical customer". Examples — paste two things you've written before; this is the highest-leverage thing you can do, and almost nobody does it. Iteration — the first output is a starting point, not a deliverable. Say what's wrong and ask again.

ROLE
You are [an experienced NZ copywriter / hiring manager / customer service lead].

CONTEXT
My business: [name, what you do, where, who you serve]
The audience: [who's reading, what they already know, what they worry about]
What I want them to do: [one specific action]
Facts you must use (do not add others): [prices, dates, names, guarantees]

TASK
[What you want written.]

STYLE
- NZ English, [tone]
- [length]
- Avoid: [buzzwords you hate — "unlock", "elevate", "in today's fast-paced world"]
- Match the voice in the example below.

EXAMPLE OF MY VOICE
[paste 100–200 words you've actually written]

OUTPUT
Give me 3 options. After each, one line on who it's aimed at.
Mark anything you were unsure about with [UNCLEAR].

Iterating well

Don't restart. Tell it what to change: "Too formal — rewrite like I'm talking to a regular." "Option 2, but cut the last paragraph and make the opening a question." "That statistic — where's it from? If you're not certain, remove it."

Here are three things I've written: [paste].

First, describe my writing voice back to me in 5 bullet points — sentence length, vocabulary, what I do and don't do.

Then rewrite the draft below to match it. Keep my quirks. Don't smooth them out.

Draft: [paste]

• Build one master prompt for your most repeated task
• Include a real writing sample of your own
• List the buzzwords you never want to see
• Practise iterating instead of restarting
• Save your winners somewhere you can copy from on your phone

Most small businesses don't have a content problem, they have a starting problem. You know your trade cold; you just don't want to sit down and write about it seven times a week. AI is genuinely excellent here, with one hard rule: it drafts, you keep the voice.

The one-idea-to-a-week method

Don't ask for "a week of posts" — you'll get seven interchangeable, forgettable paragraphs. Instead, take one real thing that happened this week (a job you finished, a question a customer asked, a mistake you fixed) and angle it seven ways.

Here's one real thing from my week: [describe it in 3–4 sentences, including the specific detail — the actual number, the actual objection, what surprised you].

My business: [what you do, who for, where]
My audience: [who follows you and why]

Turn this into 7 short social posts, each a DIFFERENT angle:
1. The story, told simply
2. The lesson for the customer
3. A common myth this disproves
4. A behind-the-scenes / how-it-actually-works angle
5. A question that invites replies
6. A short list of practical tips
7. A straight offer, low-pressure

Rules:
- Keep my specific details in. Never replace them with generic claims.
- No hashtag spam, max 3.
- NZ English. No "unlock", "elevate", "game-changer".
- Under 90 words each.

Keeping your own voice

Generic AI writing has a recognisable smell and audiences have learned it: the tricolon, the em-dash flourish, "it's not just X, it's Y", relentless positivity. Three defences that work:

• Always paste a real sample of your own writing into the prompt
• Keep the specific detail — "$340 and two hours" beats "affordable and fast"
• Change the first and last line yourself, every single time
• Read it aloud. If you'd never say it, rewrite it
• Keep a banned-phrase list and paste it into every prompt

Turning posts into visuals

Words are half the job. Once you have the week's captions, you need something to attach to them — and that's where most people stall, because a designer costs money and Canva templates all look the same.

If you want the marketing-specific walkthrough, start at Paji Studio for marketing.

Going deeper on the platforms themselves? The free Social Media Course covers posting strategy per channel, and the Small Business Marketing Course covers the wider funnel.

This is the area where AI has changed small business marketing the most, and the area with the widest gap between what people think it costs and what it actually costs. A product photoshoot in Auckland runs into four figures. The same set of images, generated and edited, costs a few dollars and an afternoon.

What's realistic right now

The rule that keeps you out of trouble

Start from a real photo whenever the image represents something a customer is buying. Generating a product from scratch that doesn't match what arrives in the box isn't a style choice, it's a misrepresentation — and under the Fair Trading Act it's a real risk. Use AI to improve, restage and extend genuine photos, not to invent the product.

Same applies to people. Don't generate a fake "team" or fake customer testimonials with faces. It's the fastest way to lose trust you spent years building.

Prompting for images

Image prompts reward specificity about photography, not adjectives. Say the shot, the lens, the light, the surface, the mood — in that order.

[Product, described plainly] on [surface/setting],
[time of day] natural light from the [left/right], soft shadows,
shot on 50mm lens, shallow depth of field,
[colour palette], clean uncluttered background,
photorealistic commercial product photography, 4:5 vertical

Avoid: text, logos, watermarks, extra objects, oversaturated colour

Then iterate on one variable at a time — change the light, keep everything else. Changing five things at once teaches you nothing about what worked.

Selling products online? The Paji Studio for ecommerce guide covers the product-shot workflow specifically. Running campaigns? See Paji Studio for marketing.

Video, realistically

Short generated clips are best used as seasoning — a three-second opener, a transition, a product hero shot — cut into footage you actually filmed on your phone. Fully generated 60-second videos still look like fully generated videos. Your phone plus good light plus a decent hook will outperform them nearly every time.

• Collect your existing product/premises photos in one folder
• Try one background replacement on a real photo
• Generate three lifestyle variations, changing one thing at a time
• Check every image for garbled text and wrong detail before publishing
• Never generate fake people, premises or testimonials

Customer service is where AI can either save you an hour a day or quietly damage your reputation, depending on one decision: whether a human reads it before the customer does.

Start with drafting, not automating

The highest-return use isn't a chatbot. It's drafting replies that you approve. You keep full control, the customer gets a considered response in two minutes instead of two days, and nothing goes out that you haven't read.

Draft a reply to this customer enquiry.

About us: [business, what we sell, our policies on refunds/delivery/lead times]
Our tone: [friendly and straightforward / formal / warm]
Facts you may use: [only these — prices, timeframes, policies]

Rules:
- If the enquiry asks something not covered by the facts above, DO NOT guess.
  Write [NEEDS AMIT] and note what you'd need to answer it.
- Answer the actual question first, in the first sentence.
- Under 120 words. NZ English.

Enquiry:
[paste]

The "[NEEDS AMIT]" instruction turns the AI from a liability into an assistant. It will flag the hard ones instead of inventing an answer.

Building the answer bank

Before any of this scales, write down your real answers once: pricing, lead times, refund policy, delivery areas, warranty, what you don't do. Feed that same block into every prompt. Most bad AI customer service is just AI with no access to the truth.

• Export or copy your last 50 customer enquiries
• Group them — you'll find 8–12 questions cover 80%
• Write your definitive answer to each, with real numbers
• Keep that as one document you paste into prompts
• Publish the same answers as a public FAQ page
• Review and update it quarterly — stale facts are worse than none

When a human must take over — no exceptions

If you do run a chatbot, give it a visible, obvious route to a real person on every screen — not buried after four failed attempts. And be upfront that it's automated. People forgive a bot; they don't forgive being tricked by one.

Admin is where the hours quietly disappear. It's also lower-risk than customer-facing work, so it's a good place to build confidence.

Quotes and proposals

You already know the price — the slow part is writing it up. Give the AI your line items, your standard terms, and the customer's situation in a sentence, and ask for a clean quote with a short scope description and clear exclusions. Never let it calculate the numbers. Language models are unreliable at arithmetic; do the maths in a spreadsheet and paste in the totals.

Write a clear, professional quote document.

Client situation: [one or two sentences]
Line items and prices (use EXACTLY these, do not recalculate):
[paste your items and totals]
Standard terms: [deposit, payment terms, validity period]
What's NOT included: [be specific]

Structure: brief scope summary, itemised table, exclusions, terms, next step.
Tone: confident and plain. No sales fluff. NZ English.
Do not add any figure I haven't given you.

Spreadsheets without the formula headache

Describe what you want in plain English and ask for the formula: "In Google Sheets, I need column D to show the number of days between the date in B and today, but blank if B is empty." You'll get a working formula and an explanation. Spot-check it on three rows you can verify by hand before trusting it on 500.

Documents and policies

Health and safety procedures, onboarding checklists, terms of service, standard operating procedures — all fine as AI first drafts, all requiring your review. For anything with legal or compliance weight, treat the draft as a starting point to take to your accountant or lawyer, not as the finished article. It's still a big saving: you're paying a professional to review rather than to write.

The free toolkit that isn't AI but saves as much time

Half of what people want AI for is really just a tool they don't have. I keep a set of 60+ free browser tools on this site — no signup, nothing uploaded to a server for most of them.

Browse all of them at amit.marketing/tools — PDF editing and signing, image and video conversion, OCR, invoicing and accounting. All free.

• Template your most-repeated document as a saved prompt
• Never let AI do arithmetic — paste in totals you calculated
• Spot-check every generated formula on rows you can verify
• Have a professional review anything legal or tax-related
• Bookmark the three free tools above

Most deals are lost to silence, not rejection. The follow-up doesn't happen because it's tedious. AI fixes the tedium — but there's a line between personalised and creepy, and it's worth being deliberate about where you stand.

Research before you write

Before a meeting or a pitch, ask your assistant to summarise a company's public website and recent news, and list five sensible questions to ask them. Verify anything factual yourself — this is exactly the situation where an invented detail gets said out loud in a meeting. Treat it as a briefing to check, not a briefing to trust.

Personalised outreach that isn't mail-merge

The AI can write the structure. You supply the one specific observation that proves you actually looked. Without that observation it's just a nicer-sounding template, and people can tell.

Write a short outreach email.

Me: [what I do, who I help, one proof point]
Them: [company, what they do]
The specific thing I noticed: [something real — a job ad, a new location,
  a post they made, something in their process I can genuinely help with]
The ask: a 15-minute call, low pressure

Rules:
- Lead with the specific observation, not with me.
- Under 100 words. NZ English.
- No flattery, no "I hope this email finds you well".
- One clear question at the end.
- If the observation above isn't strong enough to justify the email, say so
  instead of writing it.

That last rule saves you from sending mediocre emails. If the AI tells you the reason is thin, it usually is.

CRM hygiene

Paste your rough notes after a call and have AI produce a structured record: what they need, budget signals, timeline, objections raised, agreed next step and date. Consistent records make your pipeline actually usable. Ask for the same fields every time — the format matters more than the prose.

• Define 6 fields you capture after every sales conversation
• Build one prompt that outputs exactly those fields
• Set a follow-up date on every single record, no exceptions
• Draft the follow-up sequence once, personalise the first line each time
• Review no-response leads monthly — one good re-approach beats ten new cold ones

Privacy and consent — the NZ basics

This deserves care. In New Zealand, customer information you hold is personal information under the Privacy Act 2020, and the Unsolicited Electronic Messages Act 2007 governs marketing emails and texts: you need consent, you must identify yourself clearly, and every message needs a working unsubscribe.

Two practical implications for AI-assisted sales:

This is general information, not legal advice. The Office of the Privacy Commissioner (privacy.org.nz) and the Department of Internal Affairs publish plain-English guidance for businesses — worth 20 minutes of your time.

This is the module people skip and later wish they hadn't. None of it is complicated, and none of it should put you off using AI — it's just the set of habits that separates people who use AI well from people who get burnt by it.

Everything below is general information for business owners, not legal advice. For anything specific to your situation, talk to your lawyer or accountant.

Hallucinations — the thing everyone underestimates

AI models produce text that sounds right. When they don't know something, they don't stop — they fill the gap fluently. I've seen invented statistics, invented product specs, invented council regulations and invented case law, all delivered with total confidence.

The risk is highest exactly where it matters most: numbers, dates, names, legal requirements, prices, technical specifications, anything with "according to" in front of it.

• Verify every number, date, name and statistic against a primary source
• Ask "what are you uncertain about here?" and read the answer
• Never publish a statistic AI gave you without finding the original
• Be extra sceptical when it agrees with you enthusiastically
• Use the [UNCLEAR] flagging instruction from Module 2 everywhere

Confidentiality — what not to paste

Assume anything you type into a free consumer AI tool may be stored, and may be used to improve the service unless you've turned that off. Check the settings of whatever you use — most major assistants now let you opt out of training, and business plans usually exclude it by default.

The simple habit: strip identifiers before you paste. "Customer in Papakura, invoice overdue 45 days" gets you exactly the same quality of draft as the real name and address.

NZ Privacy Act 2020 — the parts that matter day to day

The Act sets 13 information privacy principles. Four of them come up constantly once AI is in the picture:

• Only collect personal information you actually need for a lawful purpose
• Be transparent — people should know what you collect and why
• Keep it secure, including when it passes through third-party tools
• Use it only for the purpose you collected it for
• Note that many AI tools store data offshore — that engages cross-border rules

The Office of the Privacy Commissioner has published guidance specifically on generative AI for organisations. If you handle customer data at any scale, read it directly at privacy.org.nz rather than relying on a summary — including mine.

Disclosure — when to say it's AI

There's no blanket NZ law requiring you to label AI-assisted marketing copy, and nobody expects a disclaimer on a Facebook post you drafted with help. But misleading customers about what they're buying or who they're talking to is a different matter, and the Fair Trading Act applies regardless of what tool produced the claim.

My own line, for what it's worth: be upfront when someone might reasonably assume they're dealing with a human (chatbots, phone systems) or when the image represents the actual product or people. Don't fake reviews, testimonials, staff or premises — ever. Beyond that, AI is a tool like a camera or a word processor, and nobody discloses those.

Keeping a human in the loop

The single rule that covers almost everything: a person who is accountable reads it before a customer does. Not a person who skims it. A person who would be comfortable defending every sentence.

Everything up to here is useless if you close this page and carry on as normal. So: three workflows, thirty days, one honest measurement. That's the whole plan.

Pick three — not ten. The businesses that get real value from AI are the ones that did three things properly, not the ones that trialled fifteen tools.

Week 1 — Choose and set up

• List your 10 most repeated weekly tasks and the hours each takes
• Pick the 3 that are "drafting something", not "getting facts exactly right"
• Pick ONE AI assistant and set it up properly
• Write your business context block — what you do, who for, your tone, your key facts
• Check your privacy settings and turn off training on your data
• Do the five quick wins from Module 2

Week 2 — Build your templates

• Build one master prompt for each of your three workflows
• Paste in real samples of your own writing
• Add your banned-words list and the [UNCLEAR] flagging rule
• Save them somewhere you can copy from on your phone
• Run each one three times and refine what came out weak

Week 3 — Content and visuals

• Take one real story from your week and run the seven-angles prompt
• Make the matching visuals in Paji Studio
• Schedule the week of posts
• Draft or update your FAQ page from real customer questions
• Set up your reply-drafting prompt with your answer bank

Week 4 — Measure and decide

• Log actual hours saved per workflow — be honest, not optimistic
• Note any output you had to redo entirely, and why
• Check quality: has anything gone out that shouldn't have?
• Keep what saved real time. Drop what didn't. Don't keep a tool out of guilt
• Write your one-page AI policy from Module 9
• Only then consider adding a fourth workflow

What good looks like at day 30

You're not "using AI" in any impressive way. You have three saved prompts, you use them without thinking, you've clawed back a few hours a week, and you've stopped worrying about it. That's the whole win. Anyone promising a transformed business in a month is selling something.