I resisted this for longer than I'd like to admit, and you should know why before I ask you to read 30 more posts.
We build software for pet businesses, so anything with AI on it gets pitched at us, and almost all of it was a chatbot bolted to a booking page that could not answer whether Thursday was free. The demos ran on a tidy imaginary salon where no dog arrived matted. And when I tried the tools on my own work, they wrote in a voice that was not mine.
The short version. AI for pet businesses is worth your time on four levels, and they go in this order. Getting started: you get real work out of a chat window, and you learn what you must never paste into one. Making it yours: it stops guessing at your prices, because you wrote them down once. Letting it work: it stops being a chat window and does jobs on your actual files, with you deciding what it may touch. Running on rails: the repeatable half of your back office runs from files you own, and you never explain those jobs again.
Most people who read this will never leave the first level. That's a completely fine outcome, and I will come back to it.
Take Steph, a two-groomer salon in Sydney's Inner West. She is a worked example we follow through this series, and her numbers are the sort you would recognise rather than a real client's books.
What we resisted, and what changed it
Two things kept me out. The first was that it was confidently wrong. Not hedging, not obviously guessing: wrong in exactly the tone it uses when it is right, which is the one failure you cannot catch by reading carefully. Why it lies to you, and the check that catches it takes that apart, and the story in it is ours.
The second was that every pitch aimed at the interesting half of the job. Talking to clients. Setting a price. Reading a coat. I would not hand those over even if the software were perfect, and it is not.
What changed my mind was not a demo. It was giving one of these tools a job so boring that I could check its answer with a calculator. Files in, one plain report out, and a total at the bottom I could verify by hand in under a minute. It got it right. Then it got the same job right again the following week, which is a far more useful fact.
The second thing that changed my mind was working out why it had been so bad. It was not stupid. It was empty.
It didn't know what I charge, what my notice period is, or how I talk, because nobody had ever told it, and I had been marking it down for guessing. Write those three things down once and most of the disappointment goes away. That is an hour of genuinely dull work, and it is the hinge of this series.
So, my position now. It does not run a pet business. It does the half of the job I have never wanted to do, in a shape I can check, and only while I am the one reading what came out. Everything below is arranged around that sentence.
The four levels of AI for pet businesses, and why the order matters
| Level | What you can do by the end of it | What it costs you |
|---|---|---|
| Getting started | Get a usable first draft of anything you already know how to say, and know the eight lines you must never paste | An evening of reading. Nothing to install |
| Making it yours | Stop retyping yourself. It knows your prices, your policies, and how you talk | About an hour of boring writing, once |
| Letting it work | It works on your actual files, and you decide what it may touch | A paid plan, 20 minutes to install, an hour of small jobs, and a backup |
| Running on rails | The repeatable half of your back office runs from files you own | Evenings of writing things down before any of it pays |
The order is not a difficulty ranking; each rung removes the reason the next one fails.
Level one teaches you to read the output before you trust it, which is the habit everything above depends on. Level two hands over your facts, so there is less to catch. Level three gives it hands, which is only safe once you are checking its work and it knows your business. Level four writes the jobs down, which is only worth doing once you have watched it work.
Now the part most guides leave out. Level one is a complete destination. A reader who gets good at a chat window and never installs anything has taken most of the ordinary wins here. Levels three and four are for repetition, so if you're not doing the same job by hand more than once a fortnight, they are not for you yet. And if you'd rather have reminders and payment chasing handled with no AI in it at all, the plain automation guide is a fine place to spend the evening instead.
Level one, Getting started: real work out of a chat window
By the end of this level you can describe a job in enough detail to get something usable back, and tell a good answer from a plausible one. It costs no money if you don't want it to, and nothing to install.
Start by sorting the tools rather than the brands. Which AI to use for which job is one chart that stops you forcing a photo problem, a spreadsheet problem, and a caption problem through the same box. Then, if you would rather paste a prompt than compose one, 20 of them written for a groomer's week are ready to lift straight out.
Two posts here are not optional. What you must never paste into an AI chatbot comes down to eight lines you can read in two minutes, and it is the one post I would ask you to open first. Why it lies to you gives you the check that catches nearly all of the invention, in about 10 seconds a time.
Then two tools you may already be paying for without noticing. Gemini is the button in your Google account that you have looked at for a year and never pressed, and its argument is proximity: it already sits beside your client emails and your quotes. Canva AI turns a blank month on the planner into 30 filled squares in one sitting, with a photograph of a real dog doing the actual work. And before you subscribe to anything, what AI actually costs sets out what will really leave your account.
Level two, Making it yours: stop retyping your own business
By the end of this level the output stops sounding like a stranger and stops inventing your numbers. The cost is about an hour of writing that nobody enjoys, done once, plus the discipline to paste it in at the top of a new conversation.
That hour is teaching it your prices, your policies, and your voice, and everything else in this level goes faster afterwards. If you could not write your policy lines down without hesitating, that is the tell: write the policies properly first, then have them checked by a solicitor, which is the step everybody skips.
The rest of the level is the same move pointed at four jobs you already do badly at 9pm. Your camera roll becomes a feed in phone photos to a booked-out Instagram. The three facts you type into a DM every morning become 20 questions answered once, sitting where a client meets them before it occurs to her to message you. The voice memo your groomer records in the car park becomes a written handover note before she drives off. And every review on your profile gets an answer in your own voice, including the four-star one with nothing written in it.
One more, the reference desk for the series. What all the AI words actually mean explains nine of them in 10 minutes, and nothing else in these 31 posts stops to define them, on purpose.
Level three, Letting it work: jobs on your actual files
This is where the character of the thing changes. Below here it is a window you talk into. From here it reads, writes, and creates real files on your own computer, which is more useful and more uncomfortable at once.
Be clear about the cost, because this is the first level asking for something real. It needs a paid plan; the free one does not include the tool. Installing asks about 20 minutes, and there is an honest chance your machine will not let you. Call it an evening, spent once, with no guarantee at the end.
Start with what Claude Code actually is if you have never written code, which also answers the question people think and rarely type: is it cheating? Then the honest install guide asks for its 20 minutes and carries a way back out if it goes wrong. Then three real jobs for your first hour, in an order that matters more than the jobs do.
Before you point it at anything you care about, two posts. What to allow, and what never to covers the setting that decides what it may do without asking you first. Where AI must never go in pet care is written by Annika, who runs the business we started, and it is the least comfortable post in the series.
Then the payoff. Give it a memory is one file it reads at the start of every session, so your prices and hard rules are in the room before you type. Build a skill is a job written down once in plain English, and it is the post this series walks towards. Between them sit the two jobs everyone recognises: 4,000 photos, 200 receipts, and one spreadsheet nobody maintains, and making it repeat itself on a Monday morning so that the Monday is a list rather than a decision.
Level four, Running on rails: the repeatable half, written down
By the end of this level you never explain the countable jobs again, because the explanation is sitting in a file. What it costs is the least glamorous thing in this guide: hours of writing down what you already know, before a single one of them pays you back.
If you would rather not start from a blank page, 12 skills written out in full covers the back office half of a salon, and you can read every line. A week in the life of an AI back office is the same 12 on a wall grid, Monday to Sunday, with three days where nothing runs at all.
Build your own little tools is the calculator you have wanted for three years and nobody sells, because nobody can sell you your own numbers. Ask your business a question is honest about a split: Grazi AI inside Petboost reads your live data, while an export on your own computer answers the harder ones, because it can work with costs the software never sees.
The last three are about other people. Your AI operating manual is for a 10-person daycare rather than a two-chair salon: four short documents, each with a named owner. What we actually run is Mo showing you the files we build Petboost with, because the argument only holds if it is specific. And AI rules for your team is one side of A4 a Saturday casual can read in two minutes, which is this whole ladder reduced to a page by the kettle.
Where AI for pet businesses can hurt you
Three places, none hypothetical, and all cheap to avoid.
Client data. Your client book is the most sensitive thing your business holds, and a chat window is a disclosure to whoever owns it. Australia's privacy regulator recommends, as best practice, that organisations do not enter personal information into publicly available generative AI tools at all, because once it is in you cannot reliably get it back out. The eight lines, and the law behind each, are in what you must never paste into an AI chatbot. The first catches people:
Never paste into an AI chat:
Anything clinical. A lump, a limp, a coat that has changed, a dog who will not settle. Ask, and an answer arrives, and a reasonable-sounding answer is enough to make you wait when what was needed was a call to the vet that afternoon. That is the one where the cost lands on the dog rather than on you, and where AI must never go in pet care holds that line.
Anything that sends or deletes. Once a tool can act rather than answer, these two lines are most of the safety story, and what to allow and what never to is where they turn into a setting:
And never let it:
Those three can cost you a client or a dog. One more can only cost you money: two of the four tools in this series will not price themselves in the currency you bank in, so the monthly charge is a moving number rather than a fixed one. What AI actually costs names which two, and what it does to your statement.
Where to start this week
- Read the eight lines. Two minutes, and the only item here I would call compulsory.
- Pick one job you did badly last Sunday. A caption, a message you have been avoiding, one review reply. One, not four.
- Write down your prices, your policies, and a paragraph on how you talk. This is the hour that makes everything above it work.
- Read what came back out loud before it goes anywhere. If a number or a rule is in it, check it against your own list.
If nothing survives past the second week, stop. That is a real answer, and it cost about two hours.
What this actually reclaims for you
Say Steph never leaves the first level. Her admin sits in two evenings a week, two hours each, and the part of those four hours that is actually writing is maybe 90 minutes: a caption, two messages she has been putting off, and a review reply.
She still has to read the draft, fix it, and send it. What a draft takes away is starting from nothing, and that is about half of the 90 minutes. Call it 45 minutes a week, across 48 trading weeks, which is 36 hours a year against about two hours spent once.
Thirty-six hours is 18 of those two-hour evenings. That is the floor of this series rather than the ceiling, from the level that costs nothing, and plenty of readers should stop there.
The ceiling is further up and it is paid for in evenings before it gives anything back. That trade is yours. The one thing I would not do again is what I did for too long: assume the whole field was a pitch because the pitches were bad.
If the part working against you is the software holding your client book, Book a demo and we will start with the job you least want to do on a Sunday.