AI AUTOMATION STRATEGY
Automation, in the right order.
Find the work worth automating, build it in tools you own, and keep a person in charge of anything that matters. That is the whole strategy, and the rest of this page is how we do it.
What AI automation actually is
The phrase covers two different things, and most of the confusion about it comes from treating them as one.
The first is ordinary workflow automation. Something happens, and a defined set of steps follows. A form is submitted, so a contact is created in your CRM, a task lands with the right person and a confirmation email goes out. Nothing is guessed. The same input produces the same result every time, which is exactly what you want from anything touching money or clients.
The second is the AI part: a language model reading something messy and making sense of it. It can sort a free-text enquiry into the right job type, pull the supplier name and amount off a PDF invoice, or draft a reply in your tone. These are tasks rules handle badly, because the input is never quite the same twice.
Useful workflows are mostly the first kind with a little of the second. A typical one might take a website enquiry, use AI to work out what the person is actually asking for, then hand everything else to plain rules: create the record, assign it, draft the reply, wait for a person to approve it. The AI does one narrow job inside a structure that does not depend on it being clever.
What it isn't
It isn't a robot employee. A model can read and write, but it has no stake in your business and no sense of when it is wrong. It will state an incorrect figure with the same confidence as a correct one.
It isn't a fix for a messy process. If nobody agrees on how a job should be handled, automating it just gets you the disagreement faster. Sometimes the most useful first step is agreeing on the process before anything gets built.
And quite often it isn't AI at all. A surprising amount of what gets pitched as an AI project is a simple rule that a basic workflow handles more cheaply and more reliably. We will tell you when that is the case.
How we decide what to automate first
Everyone has a list of things they would love to automate. The trick is ordering it. We score each candidate on three things, and we have written up the full method in what a service business should automate first.
Volume. How often does it happen? A task done every day beats one done once a year, however annoying the annual one is.
Whose time it uses. Work that pulls your most senior person away from billable work is worth more to fix than work that fills a quiet hour.
How much judgement it needs. If two competent people given the same inputs would produce the same output, it is a good candidate. If they would not, it is judgement, and it stays with a person for now.
Before scoring anything, we check what you already pay for. Xero already sends invoice reminders and handles repeating invoices. Most CRMs already have sequences. Practice management software often has workflows nobody has switched on. Rebuilding a feature you already own is the most common way automation budgets get wasted, and our guide to Xero invoice processing shows how much Xero covers before anything needs building.
Our process, step by step
These are the five steps on our process page, as they run on an automation build.
- Discovery & Audit. We map the work as it really happens, watching the person who does the task walk through it on a screen share. The written procedure, if there is one, is usually not what happens.
- Strategy & Roadmap. We choose one workflow on the scoring above, with a named person in your team who owns it, and design it on paper: the trigger, every step, what happens when something is missing or wrong, and which steps need a person to approve them. You sign this off before we build.
- Build & Integrate. We build it in a safe copy, against test accounts or a sandbox where the tools allow it, using real examples from your business rather than tidy invented ones.
- Launch & Deploy. For a period the workflow runs alongside the manual process, so differences show up before anyone relies on it. Then it goes live with alerts on: every workflow tells a named person when it fails. Silent failure is the risk we design hardest against.
- Optimise & Scale. Once it has been running, we look at what moved and score the list again. The first automation usually changes what should come second.
The tools we build in
Builds are custom code, connected to the tools you already run on: Xero, your CRM, email, calendars and the rest. Before anything is built we check what you already pay for, because rebuilding a feature you already own is the most common way automation budgets get wasted.
Custom code has one advantage over a rented workflow platform: it does exactly what your process needs and nothing else, with no per-task meter running in the background as volume grows. It has one honest cost as well: somebody has to maintain it. Every build is documented in plain English so that person does not have to be us, and ongoing maintenance is part of what a retainer covers.
AI agents
An agent is a language model given instructions and a set of tools it may use, such as reading an inbox, searching your files or creating a draft. It decides which tool to use next. That flexibility is useful for triaging email, extracting details from documents and drafting replies. It is also why we fence agents in tightly. An agent gets only the tools and access its job needs, and a person approves anything that leaves the building. Our guide to AI agents for service firms covers where they work and where they don't.
Risks, and how we manage them
AI gets things wrong. Anything sent to a client, anything financial, and anything legally significant goes through a person before it happens. Where we can, we check AI output against a rule as well, such as matching an amount read from a PDF against the bill in Xero.
Workflows fail quietly. Logins expire and apps change. Every workflow we build reports failures to a person rather than skipping a step without anyone noticing.
Client information ends up somewhere it shouldn't. Under the Privacy Act 2020, information you send to an overseas service to store or process for you is generally treated as still held by you, so you stay responsible for it. We list every service a workflow sends personal information to, send only what a step needs, and check the data settings of any AI provider before connecting it.
Running costs creep up. Platforms that charge per task or per run can become expensive as volume grows. We estimate running costs before building, not after the first invoice.
Only one person understands it. Every build is documented in plain English, and the accounts are yours.
What a first project looks like
A first project is deliberately small: one workflow, one owner and one clear sign that it is working.
Take a hypothetical example. An accounting practice spends every March chasing clients for year-end documents. The workflow sends each client a personalised checklist, records what arrives against the client in the practice management system, sends polite reminders for what is missing, and gives the partner a weekly list of who is still outstanding. AI reads the uploaded files to suggest which checklist item each one covers. A person confirms anything it is unsure about. Nothing is filed or lodged automatically.
A focused build like that starts at NZ$1,500, scoped and quoted before work begins. If you want us to keep developing and monitoring workflows afterwards, a retainer starts at NZ$800 a month.
How you will know it worked
Before we build, we agree what success looks like for this one workflow, in terms your team already uses. It might be how many document reminders went out without anyone sending them, how many enquiries were logged with their source, or how many errors a person caught at the approval step. The workflow records those numbers as it runs, so the review is a look at real data rather than a feeling.
We don't promise a figure in advance. Every business is different, and an estimate made before the workflow exists is a guess. What we promise is that it will be measured, and that you will see the results.
When automation is the wrong answer
When the process changes every month. When the task happens a handful of times a year. When the real problem is a decision nobody has made, or a permission nobody has granted. And when the work is mostly judgement: assessing a visa case or advising on a dispute is a professional's job, and the right automation there is the admin around it.
Related pages
PRICING
Build it once, then keep it running.
Automation is priced by the scope of the workflow, not by page count. The free automation audit gives you a fixed scope and estimate before you commit.
- Automation SprintStarts atNZ$1,500one-off build
One focused workflow, scoped in the free automation audit and built end to end.
- Mapping your current process
- Building the workflow in custom code
- Testing with your real data
- Handover documentation
- Automation RetainerStarts atNZ$800per month
Ongoing development, monitoring and support once your workflows are live.
- Ongoing workflow development
- Monitoring and error alerts
- Changes as your process changes
- Priority support
Larger multi-system builds and custom AI agents are quoted individually. Every engagement starts with a free automation audit, so you see the scope and the number before anything is built.
QUESTIONS
Frequently asked questions
What is the difference between automation and AI automation?
Automation follows fixed rules: when this happens, do that. AI automation adds a language model for the steps rules handle badly, such as reading a free-text enquiry or pulling details from a document. Most good workflows are mainly rules, with AI used in one or two narrow places.
How do you decide what to automate first?
We score each candidate on how often it happens, whose time it takes and how much judgement it needs, after checking whether a tool you already pay for does it. The highest score goes first, and we re-score once it is live.
Do you build in tools like Zapier or Make?
Builds are custom code, connected to the tools you already use. If you already run a workflow platform, we look at working with it first. And where a tool you already pay for covers the job on its own, we will say so rather than rebuild it.
Will AI be making decisions about our clients?
Not the ones that matter. AI reads and drafts. A person approves anything sent to a client, anything financial and anything with legal weight. Those approval steps are written into the design before anything is built.
How much does an automation project cost?
A focused one-workflow build starts at NZ$1,500. A monthly retainer for ongoing development, monitoring and support starts at NZ$800. Platform subscriptions are separate and billed to you directly.
What happens when a workflow breaks?
It tells someone. Every workflow we build sends an alert to a named person when a step fails, with enough detail to see what went wrong. On a retainer, that alert also comes to us and we fix it.
Get in touch
Ready to Transform Your Business?
Book a focused automation audit. We will identify the workflow most likely to save time, increase speed, or improve lead conversion first.
- Enquiries
- Use the audit form — it reaches us directly
- harpreet@auxa.co.nz
- Phone / WhatsApp
- +64 21 465 109
- Coverage
- New Zealand and Australia — remote delivery
- Response Time
- Within one business day
What Happens Next?
We review your current process, tools, and obvious revenue leaks.
We map the highest-ROI workflow to automate first, with risk and effort noted.
You get a clear roadmap: what to build, what to skip, and what it should return.