Gantry is a small, done-for-you build studio. We design, build and switch on AI agents and automations inside the systems your Sydney business already runs — then hand you everything needed to own them.
Almost every useful AI project for a small or mid-sized business falls into one of these. We don’t sell anything else.
AI agents that do a job
An agent is software that takes a defined piece of work from start to finish: it reads the incoming request, looks things up in your systems, prepares the output and stops for a person to approve. Typical jobs: turning enquiries into quotes, chasing missing paperwork, preparing a weekly report from three different systems, or triaging a shared inbox.
Best when the job has clear inputs and a clear finished state.
Assistants on your own documents
A chat assistant that answers from your policies, procedures, product sheets, contracts or past jobs — and shows which document each answer came from. Built for staff first: the new starter who needs the leave policy, the estimator who wants the last three quotes for a similar job, the account manager checking what was agreed.
Best when the answers already exist in writing, just scattered.
Email and document automation
The quiet workhorse. Supplier invoices read and coded, forms extracted into your system, attachments renamed and filed in the right SharePoint folder, contracts compared against your standard terms, first-draft replies prepared from templates. Less glamorous than an agent, and often the fastest payback.
Best when someone is re-typing the same fields every day.
A first build runs six to eight weeks from kick-off to measured result. Here is the shape; the timeline page has each week in detail.
Weeks 0–2 Map the process, collect real samples, agree what “done” and “better” mean.
Weeks 3–4 Build: connect systems, write the prompts and rules, add approval steps and logging.
Weeks 5–6 Run on real work with your team checking every output. Fix what’s wrong. Train.
Weeks 6–8 Measure against the before number. Decide: keep, adjust or stop.
Two things make this pace possible. First, we build inside tools you already pay for — usually Microsoft 365 or Google Workspace, plus your accounting package, CRM or job system — so there is nothing to migrate. Second, we build one thing at a time. A business that asks for five agents at once gets one agent first, and a better second one because of it.
03 What lands in your hands
An AI build you can’t see inside is a liability. At handover you receive:
workflow The running automation, in accounts that belong to your business, not ours.
prompts/ Every instruction the AI follows, in plain text, with notes on why each rule is there.
runbook.pdf What it does, what it doesn’t, how to pause it, who to call, and what to check when something looks off.
access.md A list of every permission the build uses and how to revoke each one.
tests/ The sample inputs we tested with and the outputs we accepted, so changes can be re-checked.
report Before-and-after measurement from the live run.
training Short training material for the people who approve its work.
If you later want someone else to maintain it, or your own IT provider, they start from that package. The handover page explains each item and how maintenance works.
04 Good fit
Sydney businesses of roughly 5 to 200 people with a repetitive, document- or email-heavy process.
Teams already on Microsoft 365 or Google Workspace with cloud accounting or a CRM.
Owners who want the thing built, not a strategy deck.
Someone on your side who can spend an hour or two a week with us during the build.
05 Not a fit
You want AI to make decisions with no person checking: payments, legal positions, clinical calls.
The process isn’t settled yet and changes weekly.
You need a public-facing chatbot live in a fortnight with no review stage.
You’re after a large custom software product. We build focused automations, not platforms.
06 How we decide if something is worth building
Before we quote, every candidate process goes through the same five checks. You can run them yourself before the call.
Volume. Does it happen at least a few times a day, or in a heavy batch every week? A process that runs twice a month rarely repays a custom build, however annoying it is.
Shape. Are the inputs similar enough that a person could write instructions for a new starter? If every case is unique, an assistant to help a person is a better build than an agent to do the work.
Reachability. Can software reach the inputs and the destination? Email, SharePoint, Google Drive, Xero and most modern CRMs are easy. A desktop program with no API, or paper, makes the build longer and dearer.
Checkability. Can a person tell in under a minute whether the output is right? If checking takes as long as doing, the build saves nothing.
Consequence. What’s the worst a wrong output could do if the check missed it? We design the approval step around that answer, and some processes stay manual because of it.
Processes that pass all five are usually good first builds. Those that fail one can often be reshaped — for example, by having the build prepare a summary for a person rather than act on its own. And if a process fails several, we’ll say so on the call, which costs you fifteen minutes rather than a project budget.
One more thing we ask: who is going to approve the output for the first month? A build that nobody reviews never earns trust, and one that nobody trusts gets switched off. Naming that person early is the single best predictor we know of whether a build sticks.
07 Keeping it running
AI systems drift in ways ordinary software doesn’t. The model behind a build gets updated; a supplier changes their invoice layout; your team starts sending a new kind of request. None of this breaks the build outright — it quietly lowers quality. So maintenance is mostly watching: a monthly look at approval rates, corrections and failures, then small adjustments to prompts and rules. Most months it’s brief. Occasionally it means rebuilding one step.
You can keep us on for that, hand it to your IT provider with the runbook, or do it yourself. We’d rather you chose us because the work is good than because you can’t leave.
08 Pricing
Scoping call
Free
15 minutes. Your process, whether AI suits it, and what a first build might be.
Opportunity audit
from $2,500
About two weeks. Process map, 3–5 ranked opportunities, expected return, 90-day plan. Yours to keep.
Pilot build
from $5,000
30–60 days. One or two automations, integrated, with approvals, training and measured savings. Typically $5,000–$15,000.
Scale & maintain
to scope
Expand what worked, add the next process, monthly reviews.
AUD. Running costs for the AI services and integrations your build uses are listed separately in the proposal, so you can see the monthly cost once the build is done.
09 Straight answers
Is a chatbot on our website a good first build?
Usually not. Public-facing answers carry the most risk and the least room for review. An assistant for your own staff, or an intake agent whose replies a person approves, teaches you far more for less risk. A public chatbot can come later.
Which AI model do you use?
Whichever suits the job and your data obligations. We avoid locking a build to one model where we can, so it can be moved if prices or quality change.
Will it work with our software?
If your systems have a usable API — Microsoft 365, Google Workspace, Xero, MYOB, HubSpot, most modern job and practice systems do — yes. If not, we’ll say so in the scoping call rather than discover it in week three.
Who sees our data?
The build runs in your accounts with access scoped to what it needs. Our team sees what it must to build and test it, and that access is removed at handover unless you keep us on for maintenance.
Do you only work in Sydney?
Our work is set up for Sydney businesses, but almost all of it happens remotely, so location isn’t a barrier.
Can we start smaller than a pilot?
Yes: the audit on its own, or just the free call. Some businesses find a ready-made tool covers what they need, and we’ll tell you if so.
Got a process you’re tired of doing by hand?
Fifteen minutes is enough to tell whether it’s a good build.