In production, not in a pitch deck
The report someone builds by hand every month. The document rewritten for the twentieth client. The tender that swallows a fortnight. This is precisely the work AI is good at — and we did it to ourselves first, before we offered it to anybody else.
Give us ninety minutes →The work we take off people — 01 / 07
Right now
Someone loses a day every month pulling numbers out of three systems into last month's template.
With us
It assembles itself overnight from live data. A person reads it, adds the judgement, sends it.
We run this. Our own MSA and board reports, generated from live ticket, SLA and project data and reconciling to source within one percentage point.
The work we take off people — 02 / 07
Right now
Policies, plans and proposals that are ninety per cent identical, retyped by hand, quietly inconsistent, versions everywhere.
With us
Answer a short set of questions and a finished, branded, correctly versioned document comes out in minutes.
We run this. Our own policy and proposal generation, with house style enforced by the pipeline rather than by whoever is proofreading.
The work we take off people — 03 / 07
Right now
A fortnight of your best person's time. Requirements get missed. Last year's answers are in someone's old folder.
With us
Every requirement is pulled out and listed, matched to what you actually offer, with a draft response and a list of what is still missing. Your answers get reused next time instead of rewritten.
We run this. A five-stage pipeline that reads the tender, maps each requirement to our service catalogue, assigns a compliance position, drafts the submission, then harvests the result back into the catalogue after award.
The work we take off people — 04 / 07
Right now
Research, draft, edit, format, chase approval, repeat. It slips the moment anyone gets busy.
With us
The research and the first draft are waiting for you. You make the editorial calls and approve. Ours has gone out on time four months running.
We run this. Cyberinsights, our monthly client security newsletter, is produced by a six-agent pipeline with three fixed human approval gates. Four consecutive editions, all on schedule.
The work we take off people — 05 / 07
Right now
Insurance certificates, supplier questionnaires, expiring accreditations. Tracked in a spreadsheet until the day it matters.
With us
Tracked and chased automatically at ninety, sixty and thirty days, with a live picture of who is compliant and who is not.
We run this. Our supplier risk platform runs nightly workers that chase evidence expiry and re-score every supplier, and it blocks approval until an assessment and an agreement both exist.
The work we take off people — 06 / 07
Right now
Commitments buried in email, remembered by accident, apologised for later.
With us
A short daily list of what you actually owe people, in priority order, with replies already drafted for you to check.
We run this. Scheduled agents sweep a rolling thirty-day mailbox window, tier every outstanding commitment by urgency and produce a daily brief. Nothing is ever sent without a person approving it.
The work we take off people — 07 / 07
Right now
Finance says one thing, the operations system says another, so every meeting starts by arguing about the data.
With us
One set of figures, refreshed on a schedule, that everyone can see and that reconciles back to where it came from.
We run this. A containerised job aggregates contract, ticket and SLA data monthly into a governed analytics store — one source of truth, reproducible figures, documented methodology.
A day of assembly becomes twenty minutes of review.
The same standard every time, whoever is on leave.
Renewals, expiries and follow-ups stop depending on memory.
Judgement, relationships and the client conversation, not retyping.
Built, deployed, in daily use
Everything below runs inside Mercury IT right now. Our client reports, our monthly security newsletter, our supplier register, our tender responses and our own inboxes are all handled this way. We are not experimenting on you.
01
Three production integration servers give our AI live read and write access to our PSA, IT documentation and security operations platforms — roughly 150 tools in all. Hosted in Azure Australia East, authenticated with Microsoft Entra ID, containerised and CI/CD deployed. The AI answers from live operational data, not exported spreadsheets.
02
Monthly managed-services reports and board-level security reports are produced from live ticket, SLA, project and threat data with charts embedded. Days of analyst effort become minutes, and the numbers reconcile to the source system within one percentage point.
03
A six-agent research → draft → edit → publish pipeline produces our monthly client security-awareness newsletter, with three fixed human approval gates and machine-enforced house style. Four consecutive editions shipped on schedule.
04
A five-stage pipeline reads a tender document, extracts every discrete requirement, maps each to our service catalogue, assigns a compliance position, drafts the proposal and submission checklist, then harvests the result back into the catalogue after award. Every bid measurably improves the next one.
05
A multi-tenant supplier risk platform aligned to ISO 27001:2022 A.5.19–5.23, CIS v8.1 Control 15, NIST CSF 2.0 GV.SC and the ISM procurement guidelines. AI triages external news signals into severity ratings; scheduled workers chase evidence expiry and re-score suppliers nightly. Approval is blocked until an assessment and an agreement exist.
06
A containerised job runs monthly in Azure to collect and aggregate contract, ticket and SLA data into a governed analytics store. Unglamorous, and the reason the AI can be trusted — one source of truth, reproducible figures, documented methodology.
07
A scored AI readiness self-assessment that emails a branded governance report. Live, and built by us.
08
Multi-year ICT roadmap planning with customer approval workflow, MFA and row-level tenant isolation. Live, and built by us.
09
Scheduled agents sweep a rolling thirty-day mailbox window, tier every outstanding commitment by urgency and produce a daily brief with suggested next steps. Nothing is ever sent or posted without a person approving it.
Nothing we build sends, posts or commits on its own. Drafts land for review; approval gates are designed in, not bolted on.
We connect to sources of truth rather than copying them. Australian hosting, enterprise identity, audit trail on every action.
We prove value on a single painful process in weeks, then widen. No twelve-month transformation programme.
Every capability here runs inside our own business first. You are not the pilot.
No email, post, invoice or client document leaves without a person approving it.
We connect to your systems rather than copying your data elsewhere, and it stays in Australia.
If the honest fix is a better process, we will say so. It is usually cheaper.
Done properly, it looks like the work simply arriving finished.
Where AI can safely help, where it cannot, and what must be true first. Scored against Australia's AI Ethics Principles.
We take one repetitive, document-heavy process and turn it into a governed agent pipeline with human sign-off built in.
Connectors that let AI read and write your line-of-business systems under Entra SSO, least privilege and full audit logging.
Board packs, compliance reports and client reporting generated from live data on a schedule, in your format.
Tenant configuration, data classification, acceptable use and monitoring, so your people use AI without leaking your IP.
Hands-on training that leaves your non-technical staff building and maintaining their own automations. Includes Working Smarter with AI, a two-hour session delivered on site for up to twenty people, covering Copilot, Claude and ChatGPT.
Most Copilot rollouts stall for the same reason: turn it on and it surfaces everything your staff could already technically reach. We fix that first — sensitivity labels, Purview configuration and oversharing remediation — so Copilot is useful on day one instead of quietly dangerous.
Automation is not a project you finish. We monitor what we build, watch for drift when your source systems change, keep the approval gates honest and report on what it actually saved you.
When the small wins are proven
Prove the small things first. Once a few workflows are running and you can see what they gave back, a larger question becomes worth asking.
Somewhere in your business is software that does not fit. A platform you use perhaps a fifth of and pay for all of. Or several that will not talk to each other, so your people spend their day moving between windows, carrying information across by hand. It works. It also shapes how you work, because it was built for somebody else's business and yours has quietly bent to fit.
Building your own used to be unthinkable: the cost, the timeline, the risk of a two-year project landing late and wrong. That has changed — not because software became easy, but because the slow part became fast.
What comes out is not the old tool with your logo on it. It is the job itself, reconsidered. Only the features you actually use. Your integrations in from the first sketch rather than bolted on afterwards. AI in the places where it earns its keep, and nowhere else.
And the people building it are the people whose job is to worry about how it could be attacked.
We run this
Helm is the tool our own engineers work in. The ticket system, the documentation platform, email and an AI chat window, pulled into one screen instead of four windows and the tax of moving between them. Six to eight weeks from idea to daily use.
Who's behind it
Mercury IT's AI practice is led by Chris Haigh, our CISO — CISSP, twenty-five years in cybersecurity, with training and advisory delivered across Australia, Europe, the United States and South Africa.
That matters because the thing standing between most businesses and useful AI is not enthusiasm. It is the reasonable fear of leaking your own information. Every system described on this page was designed by someone whose job is to worry about exactly that.
Give us ninety minutes and one process you dislike. We will map it end to end and tell you honestly whether AI is the right answer, what it would cost, and what it would return. If it is not, we will say so.
Not ready to talk to a person?
Take our scored AI readiness assessment instead. It takes a few minutes and emails you a governance report you can take to your board.
Take the assessment →