Let the robots do the boring bits.

// What we build

AI automation for Sydney business: the three builds.

AI work for small business falls into three shapes. Each has a published starting price, and each starts with the same question: how many hours does it give back, and is that worth more than it costs to build and run?

• Quote and invoice extraction, from $2,200. Supplier bills, quotes, purchase orders and forms arrive as PDFs and email attachments, and someone retypes the numbers into the accounting package or the job system. The build reads the document, pulls out the fields that matter and puts them where they belong, with a person checking anything that looks wrong. A born-digital PDF is read exactly, character for character. Only true scans and photos go through vision OCR, for the reason explained further down this page.
• Email and enquiry triage, from $1,650. An assistant that sorts what comes into the inbox, drafts replies to the routine questions, and pulls its answers from your own documents rather than the open internet. A person still presses send. The win is that the inbox is sorted before anyone opens it, and the questions you answer the same way every week arrive with the reply already drafted.
• Report drafting from your own data, from $2,750. The weekly or monthly numbers pulled out of the systems that hold them, assembled and written up as a first draft you edit rather than write. This sits inside the workflow automation build, because most of the work is the wiring between systems and only the last step is the writing.

All three are scoped fixed-price with a payback estimate first, and all three are documented so the next person can see how they work and change them without me.

// Systems that talk to each other

Workflow automation between the systems you already run.

Most of the repetitive work in a small business is not hard. It is copying. A figure leaves one system and gets typed into another, a status changes in one place and someone has to remember to update it somewhere else. That is the work automation is best at, and most of it needs no AI at all.

The systems are the ones you already have: the POS, the CRM, Xero or MYOB, the phone system and the job or booking system. Where a ready-made connector exists I use it. Where it does not, I build the bridge, the same custom API and middleware work described on the system integrations page. AI comes in only for the step a plain rule cannot handle, such as reading a free-text email to work out which job it belongs to.

• Sales to accounting. A sale in the POS or the web store raises the invoice and the stock movement in Xero or MYOB. No double entry, and no end-of-month reconciliation of two systems that disagree.
• Phones to CRM. Caller details on screen before you pick up, and the call notes filed against the right customer afterwards.
• Enquiry to job. A web form or an email creates the job, the calendar entry and the acknowledgement, and chases the details the customer forgot to include.
• Job to invoice. A job marked complete raises the invoice, attaches the photos and sends the follow-up, without anyone remembering to do it on Friday afternoon.

The tooling underneath is deliberately boring. Where a workflow can run on your own hardware it runs as containers on a small server you own, the same self-hosted pattern I use for my own business tools, which keeps the monthly cost close to zero and keeps the logic yours. Where the systems are cloud services, the automation talks to them through their APIs and runs wherever makes sense.

Automation is not only paperwork. The same thinking applies to hardware, and our Rural IoT site covers on-farm sensors and monitoring. There too, the sensors are the cheap part. The value is the alert that reaches your phone the moment a reading crosses a line you set, rather than a dashboard you have to remember to open.

// Use AI only where it earns its keep

Document processing, done the right way round.

Here's the short version: most "AI document processing" reaches for the fanciest tool first, and that's backwards. Before I point any AI at a PDF, I check whether it already has a text layer. A born-digital PDF, one exported straight from software rather than scanned, carries its text inside it. I pull that text out exactly, character for character, with zero AI in the loop and zero chance of a hallucinated number. AI vision OCR is held in reserve for what actually needs it: true scans and photos, where there's no text to read and a machine has to interpret the pixels.

Two more rules I stick to. Sensitive documents get processed on a local model, so the contents never leave the machine, no third-party cloud, no copy of your paperwork sitting on someone else's server. And when the digits have to be right, financial figures, medical records, I pick precision over speed every time. A fast answer that's wrong on a decimal point is worse than useless. That's the whole philosophy: use AI where it genuinely earns its keep, and use the boring, exact method everywhere else.

// Your data stays put

Local models: your documents stay on your machine.

The rules above have a practical consequence. Sensitive work runs on a local model, on hardware in your office or on a server you own, and the document never leaves the building. No third-party cloud, no copy of your paperwork on someone else's server, and no terms of service to read before you can process a client file.

• Runs locally. Anything with financial figures, medical or legal records, staff details, customer lists or contracts. The model reads the document on the machine, produces the fields or the summary, and nothing is sent anywhere. This is the default for document processing.
• Fine in the cloud. Drafting a marketing email, summarising a public document, answering a question about your own published website. If the input is not sensitive and the output is a draft a person will edit, a cloud model is cheaper to run and easier to keep current.
• How we decide. Two questions per task: would it matter if this text was on someone else's server, and do the digits have to be right? A yes to either sends it local, and the exact, non-AI method is used wherever it can be.

The hardware is smaller than people expect. A local model that reads a small business's daily invoices and forms runs on one quiet box in the office, sized to the workload, and it can be the same box that carries the self-hosted tools and the backup target. If you already have a server, I check whether it can carry the job before anything is bought.

// How it runs

How an AI project runs: workshop, roadmap, payback check, build.

There is a fixed order, and it starts with a conversation rather than a purchase.

• Free consultation. A conversation about where the hours go. If nothing in your week is worth automating, that is the answer, and it costs nothing.
• Discovery workshop, $990, half a day. I sit with the people who actually do the work and map the repetitive jobs: what comes in, what gets typed, where it goes, what breaks. You leave with a written roadmap ranked by payback.
• Payback estimate before any build. Every automation on the roadmap gets an estimate: the hours it saves, the errors it prevents, and what it costs to build and run. If the number does not work, I say so and it comes off the list.
• Fixed-price build. One process at a time, scoped and quoted before work starts. Small first, measured against the estimate, then the next one.
• Tuning, monitoring and support, from $165 a month. Automations drift as the business changes: a supplier changes their invoice layout, a form gains a field, a system update moves a button. Monitoring catches it and tuning fixes it before it costs you anything.

// Honest advice

What we will tell you not to automate.

Part of the job is saying no. Three kinds of work go straight to the bottom of the list. This is the expensive sparkle the top of the page warns about.

• Customer-facing decisions with money attached. Discounts, credits, refunds, and quotes for anything non-standard. An AI can draft the number; a person decides it. One wrong automated decision reaching a customer costs more than the hours it saved.
• Anything that needs a signature or an approval. Contracts, purchase orders over a limit, payroll changes, compliance sign-offs. Automate the paperwork up to the point of the decision and stop there. The approval is the control, and removing it is not efficiency.
• Anything where the data is not clean yet. If the customer list has three spellings of every name and the job codes mean different things to different staff, automation will process the mess faster. Fix the data first, usually with the data migration and cleanup work on the integrations page, then automate.

There is also the plain case where the maths does not work: a task that takes twenty minutes a month is not worth a build, however clever the tool. Those get a quicker fix, a template or a checklist, or nothing at all.

// Indicative pricing

What it roughly costs.

Real numbers, because "contact us for pricing" is code for "brace yourself." These are honest ballparks, your fixed quote comes after a free consult.

Indicative pricing ex GST at market rates +10%. Every automation is scoped with a payback estimate first, if it won't pay for itself, I'll tell you.

// Questions

AI automation, your questions answered.

// Pairs well with