AI Marketing Workflows for Small Business: What to Build When You're the Whole Marketing Department

Start with the marketing job that already has a customer waiting at the end of it. The reply to an enquiry, the follow-up after a quote, the same answer you type out three times a week. That work has a deadline, you'll spot a wrong draft immediately, and it's the only marketing you never skip. Social posts and blog articles come second. Most of what ranks for this search hands you a stack of tools to buy, which quietly assumes somebody on staff has time to run them.

TL;DR

• Your marketing loses to every other job in the business because nothing breaks when you skip it. That absence of a deadline, not a shortage of tools, is the problem an AI workflow has to solve for you.
• The audit method the big guides publish will rank your marketing last, because one of its three tests asks whether the job delays anyone when done by hand. Marketing delays nobody. Swap that test for a better one: what happens to this job in a bad week?
• Handle the enquiry reply and the quote follow-up first. They carry a real customer, real consequences and source material you already wrote. The published content layer comes after that.
• The standard advice for making AI output sound like you is to feed it your brand guidelines. You don't have any. What you do have is a decade of your own emails, quotes and phone answers, and that material is more specific than any brand book.
• With nobody else reading the draft, the review has to be a written list rather than a feeling, because you'll be running it at 9pm. Published marketing also carries an obligation an internal document doesn't: under Australian Consumer Law your claims have to be accurate and provable, and intent makes no difference.
• Keep the personal reply personal. A national competitor can match your prices and outspend your ads, and it still can't have the owner answer the message.
• One seat in our full-day workshop suits an owner doing this for themselves. A team needs a private onsite session. If you'd rather the system was built and connected for you, that's the other rung of the same ladder.

Why does your marketing always lose to the rest of the business?

Because nothing goes wrong when you skip it.

Look at what else your week contains. A quote has a customer who rings on Thursday if it hasn't arrived. An invoice has money attached. A BAS has a date and a penalty. A complaint escalates. Every one of those jobs has someone outside your business applying pressure, so they get done even in a terrible week.

Marketing has nobody applying pressure. Skip the newsletter and no phone rings. Skip a month of posts and the only consequence arrives in about six months, disguised as a quiet January. So marketing loses every negotiation with your own calendar, and it loses quietly, which is the part that makes it hard to fix.

This matters for choosing a workflow, because the standard method for picking one will send you the wrong way. GoDaddy's guide runs a three-question audit before automating anything: does this happen every day or every week, does it follow the same steps every time, and does it cause a delay when done manually, with a yes to all three marking a good candidate 1. Now answer those three for writing your own marketing. Yes. Roughly. And no, it delays nobody at all.

That test is sound for the work that moves through your business, which is what it was written for. Applied to marketing it recommends deprioritising the one job that's already starving.

So change the third question. Instead of asking who gets delayed, ask what happens to this job when the week goes bad. If the honest answer is that it silently doesn't happen, you've found a candidate, and what an AI workflow buys you there is the job existing at all rather than the job going faster. A workflow that turns a 40-minute task into 15 minutes saves you 25 minutes. A workflow that turns an hour of dread into a 15-minute job you'll actually do on a Wednesday night changes whether the work happens, and those are different sizes of win even though a stopwatch reports them the same way.

If you want the general version of this decision, covering every repeated job in the business rather than the marketing ones, how to choose your first AI workflow for a small business sets out the full test.

Which marketing job should you hand over first?

The one with a specific customer waiting at the end of it.

Most owners don't file the enquiry reply under marketing, which is why it never gets improved. It's the single highest-value piece of writing in a small business. Someone has decided to consider you, they've sent a short message with half the information missing, and what you send back decides whether you get the job. Then you write it from scratch again the next day, at a different standard, depending on how the morning went.

It's also the safest place to learn. You already own the source material, because your own past replies are sitting in your sent folder. You'd notice a wrong price or an invented inclusion instantly, because it's your price list. Nothing gets published, so a bad draft costs you a delete rather than a screenshot. And the job has a real deadline, so you'll run the workflow often enough to find out where it breaks. The AI workflow for customer follow-up covers the follow-up half of that in detail.

This is roughly the shape of the work I did for Glen House Homoeopathy, where we used AI to automate SEO, enquiries and admin. A single-practitioner practice has no marketing department and no spare afternoon, so the enquiry and admin side is where the pressure actually sits, and the search work sat alongside it rather than instead of it. I'm describing the scope of that engagement rather than promising you a number from it.

The second tier is content you've already checked once. Take something you published, or a quote you wrote, or a job you finished, and turn it into the short pieces you'd otherwise never get around to. An Australian guide describes the pattern plainly: after publishing a blog article, AI can draft the LinkedIn, Facebook and Google Business Profile posts from it, and a person reviews and publishes them 2. The reason this is second rather than first is that the facts have already survived one review, so the drafting step can't introduce a claim you never made. Getting the posts themselves right has its own set of problems, and the AI workflow for social content deals with them.

Third is the material written from scratch: the article, the service page, the thing with no existing source. It's the obvious target because there's the most of it, and it's the worst place to start, because you have to go and verify every statement rather than recognise it. When you get there, the AI workflow for blog creation is the long version, and the research side of choosing what to write at all belongs with AI workflows for SEO.

Paid ads sit further out, not because AI is bad at ad copy, but because you're spending money while you're still learning where the draft goes wrong. Learn on the work that costs you nothing to get wrong.

And the content calendar you've never run doesn't belong on the list. You can't tell whether a draft is right for a job you've never done, and building a workflow for a habit you don't have produces a folder of drafts rather than a marketing routine.

You already have a brand guide, it's just sitting in your sent folder

The standard fix for generic AI output assumes a document you've never written.

Salesforce's small business guide sets the mistake out in a table: losing brand authenticity, with the recommended solution being to "inject specific brand guidelines and unique perspectives into every generated draft" 3. That's correct advice for a marketing team with a brand book, a tone-of-voice section and someone whose job it is to maintain them. You have a business to run and no such document exists.

What you have instead is better raw material, because it's real. Ten years of emails where you explained the same thing to a customer. Quotes where you worded a scope so it wouldn't be misread. The answer you give on the phone when someone asks why you're dearer than the bloke down the road. Nobody wrote that down as guidelines, and it's far more specific than guidelines ever are.

Harvesting it takes about an hour. Find ten replies you were happy with and keep them as examples. Write one page of facts: your prices or price ranges, your service area, what you do and the jobs you turn down, the three things customers always ask. Then write a short list of what you'd never say, which for most owners is the fastest part, because you know exactly which industry phrases make you wince.

That's the same thing a marketer who does this for a living ends up doing by hand. Omid Ghiam describes uploading his own writing samples to train a model on how he writes, plus "a document of words and styling not to use" 4, and his blunt rule is worth carrying into your own work: "you can't automate what you can't teach someone else to do" 5. If you can't say what makes your replies yours, the draft will average out to everybody else's.

Averaging out is the actual risk now, not embarrassment. Salesforce cites IDC research finding that 97% of marketers use generative AI to support content marketing, including creating derivatives of content and localising it 6. Take that at face value and the default output of an unprompted model is roughly the middle of your market. Your competitors are drafting with the same tools from the same blank page, so the sentences that could only have come from your business are the part that still separates you, and they're the part sitting in your sent folder.

Once you've harvested the material, it goes somewhere specific. We teach five inputs: Context, Task, Source, Rules and Output. Your harvested facts become Source, and your never-say list becomes Rules. The mechanics of writing those five down are set out in the five-input prompt.

If you run an agency rather than a business that does its own marketing, your problem is genuinely different: you have brand guidelines, several people drafting for the same client and an approval step to defend, and AI workshops for marketing agencies is written for that.

How do you check a marketing draft when nobody else will ever read it?

With a written list, in the same order every time, because you'll be doing it at 9pm and your judgement will be worse than you think.

Agencies and marketing teams have a second pair of eyes built into the process. You don't. The person who wrote it, the person who approves it and the person whose name is on the business are all you, and self-review is the weakest check there is, because you read what you meant rather than what's on the page.

The stakes are also higher than they are for your internal work, and this is the part most guides skip. A wrong figure in a quote is one awkward phone call with one customer. A wrong figure in a published post is public, and it's a representation about your business. The ACCC states that any information or claim a business provides about its products or services must be accurate, truthful and based on reasonable grounds, covering prices and claims about benefits or performance. It also states that the rule applies to advertising, quotations, social media, testimonials and websites alike 7. Two lines from the same guidance matter more once a model is drafting for you. "It makes no difference whether a business intends to mislead or not." And the first item on its list of what a business shouldn't do: "guess the facts" 7. Guessing the facts is precisely what an assistant does when you hand it a task with no source material. I'm describing the obligation rather than giving you legal advice, and the ACCC says plainly that it doesn't give legal advice either, so get proper advice if a specific claim looks borderline.

None of that requires a compliance process. It requires five lines you run in order:

1. Every number in the draft appears in the material you supplied, including prices and any figure with a dollar sign or a percentage.
2. Every claim about your work is one you could prove if somebody asked you to.
3. Names, dates, suburbs and job details match the source, not your memory of the source.
4. The offer described is the offer you actually sell, with the same inclusions and exclusions.
5. Read the first two sentences out loud. If they don't sound like you, the source material was too thin, not the prompt.

Then one stop rule, which does more work than the rest of the list combined. If the draft contains a fact you can't find in what you gave it, delete the fact. Don't park it to verify later, because later is when it goes out.

Once you've run that five times you'll know which two lines your own drafts keep failing, and the check drops to about a minute.

Keep the marketing that only works because it's you

The reason somebody picks a two-person business over a franchise is that the owner answers, so don't hand that part over.

This is the one structural advantage you have and it's easy to spend without noticing. A larger competitor can match your prices, outspend you on ads and publish five times your volume of content. What it can't do is have the person who'll be doing the job write the reply. Route that through a workflow and you've traded away the thing you were winning on to save four minutes.

In practice that means a few specific pieces of writing stay yours. The reply to the customer who mentioned their dog or their renovation. The response to an unhappy customer, where a competent generic draft reads as a brush-off and makes things worse. And any piece where your actual opinion is the value: why you won't use a cheaper product, what you've seen go wrong, the thing you'd tell a mate. A model can dress up an opinion you hand it, and it hasn't got one of its own.

Review replies sit right on the line. The published pattern for them is to draft a response in your preferred tone and then either publish automatically or send it to a person for approval 8. For a business with hundreds of reviews a week, automatic publishing is a reasonable trade. For a small business, draft and approve, always, and read the negative ones twice. A five-star review deserves a warm two lines. A one-star review is a piece of public writing that prospective customers will read more carefully than anything else on your profile.

One more thing to leave off the list: the weekly AI marketing report. It's the easiest thing in this whole subject to build and the least likely to change anything. In my own business, Searchmaxxed, what I want out of a search process is a recommendation about what to do next, built from evidence about the market, rather than a larger summary of every number available. Apply that to a report before you automate it, and decide which piece of information would change what you post or send next week. If nothing in it would change a decision, all you've added is a task with a nice format.

Do you need to buy marketing AI software to start?

No. Start with whatever your business already pays for, and let the job tell you what's missing.

The market answers this question differently, and you'll see that in the results for almost any search on the subject. Kimi's roundup of ten marketing automation tools states that "each tool supports different parts of the workflow" 9, which is an honest description of a stack and not a description of a workflow. A stack is a set of purchases. What you actually need is a written set of instructions, the source material they're allowed to use and the check you run before anything goes out, and all three fit inside a chat window you already have access to.

The published savings figures are where I'd be most careful. GoDaddy's guide lists marketing and content creation as saving an average of 2 to 5 hours per week with a typical time to return of 1 to 2 months 10. There's no method attached to those numbers and they're published by a company selling a website builder, so treat them as a general reference rather than a forecast for your business. Getting your own number is not hard. Time the job once by hand. Time it again with the draft, and include the checking, because a workflow that halves the writing and doubles the review has saved you nothing.

A subscription buys capacity, and capacity you haven't scoped goes unused. Run the job by hand for a month first, notice the specific thing that annoys you, then buy the thing that fixes that. Usually the wall you hit is dull and practical rather than intelligent: you need it to read a long document, or to remember your facts without you pasting them in every time, or to reach a system where your customer records live.

Where do you build this, and who builds it?

Four options, and the right one depends on whether this is your own workflow, your team's, or something you'd rather not build at all.

On your own it costs an evening. Harvest your material, write the five inputs, run the job twice and keep what you corrected. Plenty of owners get their first workflow to a decent standard that way, and if you enjoy the fiddling, do that before you pay anyone.

The public workshop is for getting it right the first time with someone next to you when it breaks. You bring the jobs your business repeats and work on it for a full day, on your own laptop, and you leave with the written workflow rather than notes about workflows. Marketing tasks suit it well, because the source material is yours and the failures show up fast.

The details of what to have ready are on the public workshop page, and people travel in from outside the region for it.

If more than one person in your business writes the marketing, the problem changes shape. Two people drafting from two different sets of facts produces two voices and two versions of your price list, and that doesn't get fixed by one person attending anything. Private onsite AI training runs at your workplace and gets scoped around your team's tasks, experience and the tools you already use. Sessions suit roughly 4 to 20 people, larger teams can be split across more than one, and a fixed quote follows a conversation about the team's tasks, location and experience, with team size, session length and travel determining the price.

The fourth option is the one some owners actually want. If you'd rather the system was built and connected for you than learn to build it yourself, we do that too. It's a separate engagement with its own scope and price, and it starts with the same conversation. Say so when you enquire and we'll scope that rung instead.

Frequently asked questions