Work that shouldn't be done by hand.
We build AI systems that take over a defined part of your business end-to-end: the enquiries, the documents, the handoffs between your systems. Then we keep them running. You own the code, your data stays in Australia, and a human approves anything that matters.
Using AI isn't the same as running on it.
Most Australian businesses now use ChatGPT for emails and drafts. Almost none have changed how an enquiry gets answered, how a document becomes a filed record, or how two systems stay in sync. The admin is still admin; it's just done with better prose. The gap between "we use AI" and "this runs itself" is where the hours are, and closing it is an engineering job, not a prompt.
Five jobs we automate.
Each one is a pattern we've shipped, shown with the system that runs it. If your version looks different, that's exactly what the assessment is for.
Enquiries & bookings
For tour operators, clinics, education providers, and anyone else whose inbox is a to-do list.
An enquiry lands at 9:40pm. The system reads it, answers in plain English, checks real availability, holds the spot, writes the booking into the calendar and the CRM, and sends the confirmation. Anything unusual waits in a queue for a person, with a drafted reply attached.
Running at · Cheeky Kiwi Travel A booking platform on the Rezdy API with an AI admin behind it that drafts the daily supplier emails from live bookings and answers booking questions in plain English, from the live data only.The 3pm supplier run took Cheeky Kiwi staff one to two hours a day for fifteen-plus suppliers. Now Claude drafts every email from tomorrow's bookings and a person reviews and sends.
Documents & reports
For brokers, advisers, agents, legal, licensing, and any service that runs on paperwork.
A client report that someone would assemble by hand gets built by the system instead: it pulls the data from the source, writes the document, finds the client's record, and either sends it or leaves it for one person to approve. The same thing works in reverse. What arrives gets read, checked against your rules, extracted into the file, and the missing pieces chased.
Running at · Joris "Send Sarah the report on 12 Smith Street" is one sentence: the report is generated on the property data service, matched to Sarah's record in HubSpot, built and emailed.Joris asks in the Claude and ChatGPT he already pays for, and our MCP server does the work behind it. Nothing new for him to log into, and switching on a new ability doesn't mean an update he has to install.
Systems that don't talk
For businesses running a booking tool + accounting + CRM + spreadsheets.
A booking changes in one place. The invoice updates in Xero, the CRM record follows, the ops spreadsheet stops being wrong, and nobody re-keys anything. "Which one is right?" stops being a question anyone asks.
Running at · POP Local Every sale, in store or online, flows from Shopify into the maker's fortnightly payout and the bank file. Nothing re-keyed.Reporting & feedback loops
For owner-led teams who want Monday's numbers without building them.
Reviews get requested at the right moment and routed: praise to the website, problems to you first. The Monday report writes itself from the systems you already run, with the three numbers you actually watch on top.
Running at · Cheeky Kiwi Travel One question after every tour: five stars goes straight to TripAdvisor, anything less to a private form first. The latest reviews sync back to the site every day.Internal tools & dashboards
For when the team needs to see and steer all of the above.
None of this is a black box. A dashboard shows what the system handled, what it flagged, and where a human stepped in, and it gives your team the controls to steer it without calling us.
Running at · Edtripper and Nilkick Operations dashboards on top of the systems doing the work.Edtripper's cockpit shows every phone's heartbeat, kiosk state and incidents, and drafts questions with Claude for an educator to approve. Nilkick meters every model call and reports the spend per run.
Not on the list?
The five above cover most of what we're asked to build, not most of what can be built. If it's repeatable, rules-based and eating hours, it's a candidate. The assessment exists to find your version: two weeks mapping the manual work in your business, ranked by what it actually costs you.
It also looks like this.
- A missed call gets an SMS with the answer and a booking link before the caller tries your competitor.
- Supplier invoices matched to purchase orders and pushed into Xero; only the exceptions reach a person.
- Quote follow-ups that go out on day 3 and day 10, and stop the second the client replies.
- New-client intake checked, chased and filed before the first appointment.
- Compliance documents (insurances, certifications, IDs) tracked, chased before they expire, filed.
- Website enquiries scored and entered into the CRM with a drafted reply waiting.
- Your own ChatGPT or Claude wired to your systems, so "send Sarah the report on 12 Smith Street" is one sentence and the report goes out.
- The day's footage read, cut into short edits and posted to your social accounts on a schedule.
- Stock levels kept in sync between the store, the warehouse sheet and the accounting file.
- The Monday work-in-progress report written from your job-management system before you're at your desk.
Assessment. Build. Managed.
You don't have to arrive knowing what to buy. That's what step one is for.
- Step 1 · 2 weeks
Assessment: decide what's actually worth building.
Book an assessmentTwo weeks, fixed price. We sit with the people doing the work, map every manual step, and cost each one in hours and risk. You get a ranked list with build prices attached. If the honest answer on something is "don't automate this", the document says that too. The fee is credited against a build started within 90 days.
- Step 2 · 3 to 8 weeks
Build: one process, end-to-end.
Three to eight weeks. Integrated with what you already run, staged alongside the real thing, your team testing on real data before anything goes live. 30 days of hypercare after.
- Step 3 · monthly
Managed: keep it true.
These systems drift: models change, APIs change, your business changes. We watch it and fix it. Improvements are scoped and quoted as they come up. Every month you get a report of what it handled, what it saved, and what it got wrong. Required for the first 12 months.
Built so you're never stuck with us.
-
You own the code and the accounts.
Delivered into your repos and your cloud. No per-task fees. Leave any time.
-
Data stays where you say.
Australian-region cloud by default; fully self-hosted when it needs to be. We'll tell you plainly which AI providers see what.
-
Human in the loop.
Approvals on anything customer-facing or money-related until you say otherwise. Every action logged; every decision explainable.
-
Built for regulated environments.
We've run production systems for an ASX-listed financial services company since 2021. Access control, change review, audit trails aren't add-ons.
-
Integrates, doesn't replace.
Your CRM, accounting, booking tool and inbox stay. We connect them.
Works with what you run.
If it has an API, we'll use it. If it doesn't, we've dealt with worse.
- Xero
- MYOB
- HubSpot
- Pipedrive
- Google Workspace
- Microsoft 365
- Calendly
- Shopify
- Airtable
- Slack
Plus the in-house system nobody wants to touch.
What it costs.
- Automation Opportunity Assessment 2 weeks · credited to your build
- $1,500 fixed
- Automation build typically $10–30k per process
- from $8,000
- Managed monitoring, fixes, monthly report · changes quoted
- from $250/month
- Internal tool / dashboard on top of the systems doing the work
- typically $8–20k
What moves the price within those ranges:
- how many systems we're integrating
- whether steps need human approval
- how messy the source data is
- volume
Why not cheaper?
A system that removes 15 hours a week of admin displaces $60–80k a year of loaded wages. We price to that, not to our hours, and the managed fee is what keeps it true in month eleven.
Indicative. Fixed price after the assessment.
Five of them, running.
We already use ChatGPT. What's different?
ChatGPT helps a person do a task. We build a system that does the task itself, connected to your calendar, CRM, inbox and files, and it only asks a person when it should. When it makes sense, we plug our tools into the ChatGPT or Claude you already pay for (an MCP server we host), so you ask in plain English and the system does the work.
Is this workflow automation?
Partly. Workflow tools like Zapier, Make and n8n move data between apps when something triggers, and we use them where they fit. What we build sits above that: a system that reads the enquiry, document or booking, decides what to do under rules you set, acts across your tools, and asks a person when it should. The workflow is one piece. The system is the thing that runs.
What if the AI gets it wrong?
It will, sometimes. That's why anything that matters waits for approval, every action is logged, and the managed plan exists. Error rates are measured and reported, not assumed.
Where does our data go?
Australian-region cloud by default, self-hosted if you need it. We document exactly which providers process what, and we don't train models on your data.
Do we have to replace our systems?
No. We integrate with what you run.
What does this mean for our staff?
The system takes the copying, checking and chasing: the parts nobody was hired for. Judgment calls, exceptions and customers stay with your team. In businesses our size the pattern is stopping the next admin hire, not losing people.
How long until something's live?
Assessment two weeks; first process typically 3–8 weeks after.
What does it cost to run?
Model/API usage is usually tens to a few hundred dollars a month for an SMB process; the managed fee covers our time. You'll see both on the monthly report.
Can we get funding for this?
Sometimes. Federal and state programs exist and change often. We keep a current list on the site and will tell you what applies.
What if we outgrow it / want to leave?
You own everything. We hand over documented, and it keeps running.
Do you do the "AI strategy" thing?
We do the assessment, which is that with prices attached. Then we build it.
Who does the work?
Johnny, Alex and Joris. Nobody else.