Do you make money on a Tuesday?
Not whether Tuesday is busy. Whether the 50 Tuesdays you worked this year paid for themselves once rent, wages, and the power for the dryers came out. Nobody running a two-chair salon can answer that from memory.
The raw material is all there: every appointment, every price, every no-show, every dog who quietly stopped rebooking. What has never existed is a way to ask a question in that shape and get an answer before tomorrow's first dog.
The direct answer. There are two honest ways to analyse business data with AI today, good at different things. Inside Petboost, ask Grazi AI: it reads your live data and answers in a panel beside whatever screen you are on. Outside Petboost, download your data as a CSV, point a tool on your own computer at the file, and ask in plain English.
The first is quicker and covers most of what you wonder about day to day. The second answers the hard ones, because it can work with numbers Petboost never sees, starting with what a Tuesday costs you to open.
The choice between them is not about which tool is cleverer. It is about where the number you need lives.
This is the top of the ladder in this series, the level we call Running on rails, and it assumes you know what Claude Code is. The examples follow Steph, a two-groomer salon in Sydney's Inner West, a worked example rather than a real client's books.
Route one: analyse your business data with AI inside Petboost
Grazi AI is the panel that slides in on the right of every page once you are logged in: press Cmd + / on a Mac or Ctrl + / on Windows. It is included at no extra cost on every one of our plans, Calendar Lite too.
It does not answer from memory. Every reply is built by querying your live data through 27 tools covering appointments, clients and pets, services, packages, team, memberships, and availability. Ask how the software works and it searches the published help centre rather than guessing at a menu path. It runs on Claude Sonnet 5.
Four questions from our own Grazi AI page, mirroring what the panel suggests:
Read the second one carefully, because it is where our own copy runs ahead of the truth. Revenue is the only half of that sum your Petboost data holds, so revenue is what it can rank.
Petboost knows what you charged. It does not know your rent, your wages, or how long the job took when Moose turned up matted again. Profit needs both halves, which is why route two exists.
Type a question into the Quick Command Centre with Cmd + K and it offers to hand it over, wording filled in.
What Grazi AI will not do, said plainly
Naming the limits is what makes the rest believable.
- It is read-only. It cannot book, reschedule, cancel, refund, charge, or change a setting. It drafts client messages; you read them and you send them.
- It is not instant. Around 13 seconds is normal and a harder question can take most of a minute, because it is running real queries rather than finishing a sentence.
- It only sees current Petboost data. No archived records, no email inbox, no accounting system, no Google Calendar.
- It will not give professional advice. Financial, legal, veterinary, and tax questions get declined, correctly.
- You cannot configure it. No tone setting, no knowledge upload, no model picker. It is designated experimental in our terms, and the panel says permanently that it can make mistakes and to double-check anything important.
Why you cannot connect Claude to your Petboost account
If you have used these tools elsewhere, you have connected one piece of software to another and watched an assistant read straight from it. So why not point Claude at your Petboost account and skip the file?
There is no connector: you cannot link Claude, ChatGPT, or any other external AI client to your Petboost account, because there is no token, no key, and no setting anywhere in the product that would let you try. Our own integrations FAQ gives the underlying reason in one line: there is no public API.
What exists is internal. We use MCP, the standard plug socket between an AI tool and other software, to give Grazi AI live access to your data on your behalf. That socket is wired into our own systems and holds our own infrastructure key, so it is not a door with your name on it.
Here is what we think, and you can hold us to the framing rather than a date. The right version of this is you holding a key to your own business: point whichever tool you like at your account, let it read only what you allowed, and revoke it on a Tuesday because you changed your mind. We have not built that, and this is not a promise that we will. We keep what we do not do and how we handle your data up in public for questions like this.
So the export is not a workaround. It is the same idea with the file in your hands.
Route two: export your data, then analyse it with AI on your own computer
Two export paths matter, for different jobs.
Export Your Data, under Billing, is the complete one. The screen is headed "Take your data with you" and holds 10 cards, each with its own Download CSV button: pets, pet owners, services, team members, resources, packages, package credit balances, all past appointments, future appointments, and future recurring bookings. Files land with names you can predict, like past-appointments-export.csv.
Two honest limits. It is 10 separate downloads rather than one button, so take only what your question needs. And it sits behind billing access: Owner Admin and Admin only, which is a control rather than an inconvenience.
The Customers and Pets page is the everyday one, behind the same two roles. Its action menu carries Smart Export, a segmentation dialog with filters, plus Quick Export for pet owners, for pets, or combined: one row per owner with their pets alongside.
Six of the seven Reporting and Intelligence tabs have an Export CSV button too, giving that tab's aggregated figures rather than individual appointments. The Map tab is the one without a button. The appointments tab breaks the pattern the other way, carrying a full raw appointments table.
Everything is CSV, whatever our older marketing copy implies about PDF, and every path is a manual download that a person clicks. There is no overnight export to schedule. How a system lets you take your data out is worth seeing before you commit, so book a demo if you are weighing us up.
The question in the first line, answered
Put past-appointments-export.csv in a folder you chose, ~/Salon/exports/ in this series, then ask this.
Three things in there do the real work. Status words get printed before anything is counted, because a system that writes a finished groom as Closed rather than Completed will otherwise lose you a day's takings. Blank prices are reported, not averaged. And it asks for the one number the file does not hold instead of inventing it.
Say Tuesdays average six dogs where a Saturday does 10. At a $95 casual full groom that is $570 against $950, out of a day costing you the same in rent, wages, and power. Whether $570 clears that is the calculation in why a pricing problem is really a capacity problem, and you can now run it on real Tuesdays instead of on a feeling.
The answer is almost never "close Tuesdays". It is usually a different price or a different roster.
Four more questions worth asking of the same files
Which clients have quietly drifted.
What a no-show actually costs you a month.
Which service earns least per hour of chair time.
Whether the price rise stuck.
That last line matters more than the rest of the prompt. The file shows what happened. Why it happened is your judgement, and a confident paragraph guessing at it is the part to throw away.
An export is the most sensitive file your business will ever make
Read what our own export screen says is inside those files. Pets: names, breeds, medical info, grooming preferences, and notes. Pet owners: contact details, addresses, vet info, and communication preferences. Your whole client book, in one file, in your Downloads folder.
Never paste into an AI chat:
And never let it:
The full list, and the Australian law behind each line, is in what you must never paste into an AI chatbot. The OAIC recommends, as best practice, that organisations do not enter personal information into publicly available generative AI tools. Its reason is plain: once it has gone in, tracking or controlling it is very difficult and removing it may be impossible.
Three habits, in the order they matter.
- Cut the columns you do not need before you ask. A money question needs the date, the status, the price, the service, and the duration. Not a phone number, an address, or a vet's name. Delete those from your working copy and most of the risk leaves with them.
- Keep the file in one folder you chose. Not in Downloads, not emailed to yourself, not in a shared drive the Saturday casual can open.
- Delete it once the question is answered. business.gov.au is blunt: when you no longer need your clients' personal information, you must destroy or de-identify it.
Check one number by hand, and check the date on the file
An analysis you cannot verify is worse than no analysis, because you will act on it. So do the check every time. It takes a minute.
Pick the busiest day in whatever came back, find that day's rows in the export, and add the prices on the completed ones with a calculator. It must match to the cent.
Then the check that catches the failure that actually happens. Ask for the counts: worked, missed, cancelled, and not counted. Those four must add up to the number of rows in the file. If they do not, a status word has been silently dropped, and why it gets things wrong explains why the answer still looked confident.
Last, the date on the export. An answer from a stale file is worse than no answer, because you act on it: you chase a client who paid last Thursday, or you write off a Tuesday that has been fine since May. More than a few days old, export it again.
When the same question comes back every month, write it down once
The first time you ask one of these you will type a long prompt and fiddle with it. The third time, stop typing it. A skill is that job written into a file once, so you ask for it by name forever after, and two of the 12 in our free skill library are exactly these: weekly-numbers and month-end.
Then the only question left is when they run, which is the Monday morning routine. Leave your tooling in the setting it arrives in, where it asks before touching a file; what to allow and what never to comes first.
What this actually reclaims for you
Every question in this post is one you have carried for a year, because answering it properly meant an evening, a spreadsheet, and a calculator.
Say you finally ask the no-show one and the file comes back with five a month. At a $95 casual full groom that is $475 a month, and $5,700 a year of chair time you had never put a number on. That is the gross figure, before your policy recovers anything and before you refill a slot, and the small tool that nets it off is its own post.
Knowing it does not recover it. You recover it by changing one thing: a deposit, a reminder, or a policy you hold the line on. What you have now is the worth of that change before you make it.
Every export here comes out of a Petboost account, and Grazi AI is on every plan. Start your free trial, pull the appointments export, and ask the Tuesday question first.