The short answer
SouthEast IT uses an AI-assisted triage system that reads new support tickets, suggests routing and priority, and drafts an initial response — but every action still requires a technician’s approval before anything is sent or changed. That is what AI IT support Melbourne businesses get from us in practice: faster sorting, human decisions.
The AI never closes a ticket, never emails a client, and never touches a system on its own. It removes the sorting work so a human gets to the actual problem sooner.
0Actions the AI takes without human approval
5Stages in the triage pipeline
AUData residency for ticket processing
650+Businesses supported across Melbourne and Australia
Every managed service provider has the same unglamorous bottleneck. A ticket arrives, and before anyone can fix anything, somebody has to read it, work out how urgent it really is, decide who should handle it, and check whether it is related to the three other tickets that came in this morning.
AI ticket management for an MSP is mostly about that first step, not about answering clients. That sorting work is genuinely valuable and genuinely repetitive. It is also where time quietly disappears: the ticket that sat in a general queue for forty minutes because nobody had read far enough to notice the phrase “nobody in the office can log in”.
So we built something to do the reading. Here is exactly what it does, what it deliberately does not do, and why a technician still approves every single action.
Behind the scenes
The AI does the sorting. A technician still makes every decision that reaches you.
SouthEast IT · Lumeio · Melbourne
Who built it, and why
SouthEast IT partnered with Lumeio, an AI decision and automation consultancy based in Melbourne, to design the system. Lumeio built the triage layer that sits in front of the SouthEast IT helpdesk, handling sentiment detection and technician scoring before a human ever touches the ticket.
The brief was narrow on purpose. We were not trying to build a chatbot that answers clients, and we were not trying to reduce headcount. We were trying to shorten the gap between a ticket arriving and the right technician reading it — which is the part of the process a client actually experiences as “slow”.
The system integrates with ConnectWise, the PSA platform where SouthEast IT’s tickets already live, and surfaces its suggestions for approval through Microsoft Teams. No new tool for technicians to learn, and no second place to check.
How a ticket actually moves through the system
Five stages. Two of them are the AI doing preparation; three of them require a person.
The triage pipeline
AI
The ticket is read and summarised
A new ticket arrives in ConnectWise. The system reads the full text and any thread history, then produces a short plain-language summary: what appears to be broken, who is affected, and whether this looks related to an existing open ticket.
AI
Sentiment and urgency are scored
Two signals are extracted. Sentiment picks up frustration and escalation language — “this is the third time” matters more than a polite subject line suggests. Urgency separates “one person cannot print” from “nobody can log in”, which are very different problems wearing similar words.
AI
Routing and a draft reply are suggested
The system scores which technician is the best fit based on the skills the ticket needs and current workload, then drafts an opening response. Both are suggestions with reasoning attached, not decisions.
Human
A technician approves, edits or overrides
The suggestion appears in Microsoft Teams. The technician can accept it, change the priority, reassign it, rewrite the draft, or throw the whole thing out. Nothing moves until someone chooses. This is the step that makes the rest of it safe.
Human
The technician does the actual work
From here it is an ordinary support ticket handled by a person, exactly as it always was. The AI has no further involvement — it does not follow up, close the ticket, or contact the client.
What the AI does, and what it never does
The second list matters more than the first. Most nervousness about AI in a support context is really a question about what it is allowed to do unsupervised. The answer here is: nothing.
What the AI does
- Reads and summarises the ticket text
- Detects sentiment and escalation language
- Scores urgency against defined criteria
- Suggests the best-fit technician
- Flags likely duplicates and related tickets
- Drafts an opening response for review
What the AI never does
- Send anything to a client
- Close or resolve a ticket
- Change a system, account or setting
- Access a client environment
- Make a final priority call
- Act without a named technician approving
What “human-in-the-loop” actually means here
Human-in-the-loop means the AI produces a recommendation and a person decides whether to act on it. The human is not reviewing a sample afterwards or auditing it monthly — approval is required before any individual action takes effect.
The phrase gets used loosely, so it is worth being precise about the version we run. There are three common arrangements, and they are not equivalent:
- Human-out-of-the-loop. The system acts autonomously and a person reviews samples later. Fine for sorting photographs. Not appropriate for someone’s email system.
- Human-on-the-loop. The system acts, a person monitors and can intervene. Faster, but it means something has already happened before anyone looked.
- Human-in-the-loop. The system recommends, a person approves, then it happens. Slower per ticket by a few seconds, and the only version we were willing to run against live client environments.
Keeping a person accountable for each decision also lines up with Australia’s AI Ethics Principles, which name human-centred values and human oversight among the principles for responsible AI use.
The practical consequence: on a quiet afternoon, the approval step costs a technician about five seconds. On the morning when four tickets arrive at once and one of them is genuinely an outage, those five seconds are what stop the wrong ticket going to the wrong person.
Where the data goes
Ticket processing happens with Australian data residency. Ticket content is processed for triage only, is not used to train any third-party model, and stays inside the SouthEast IT and Lumeio systems that already handle it.
This is the question SMB clients ask first, and they are right to. When you email a support ticket describing a problem, that text can contain a great deal: staff names, internal system names, occasionally a screenshot with something sensitive in the corner.
- Residency. Ticket processing is kept within Australian data residency rather than being sent to whichever region is cheapest.
- Purpose limitation. Ticket content is used for triage and nothing else. It is not used to train third-party models.
- No new data store. The system reads tickets already in ConnectWise; it does not create a parallel copy of your support history somewhere new.
- Access unchanged. Technicians see what they always saw. The AI layer does not widen who can read a ticket.
If you want the broader context for how we handle client data, our minimum standards for business IT set the baseline for every supported environment, and the Australian Privacy Principles govern how Australian businesses must handle personal information.
The honest framing
AI that is allowed to act alone is a risk. AI that is only allowed to suggest is a tool.
Human-in-the-loop by design
What actually changed, and what did not
The improvement is not that tickets get fixed faster by a machine. It is that the right technician starts reading the right ticket sooner, and that fewer tickets sit in a general queue waiting to be sorted.
What improved
Time from ticket arriving to the right technician reading it. Duplicate and related tickets spotted earlier. Escalation language caught on the first read rather than the third reply.
What did not change
Who fixes your problem, who you speak to, and who is accountable. Every resolution is still a technician’s work and a technician’s decision.
We have deliberately not published a percentage. Anyone quoting a precise efficiency figure for an AI deployment this young is estimating, and the honest position is that the system has removed a specific bottleneck rather than transformed the business. If you want the wider argument for why response time is the metric worth measuring at all, our guide to choosing a managed IT support provider covers what to ask any provider about their SLA.
If you are considering something similar
A few things we would tell any business owner weighing up an AI automation project, based on building this one:
- Pick the boring bottleneck. The best first candidate is repetitive, high-volume, low-judgement work — sorting, routing, summarising. Not the part of the job that requires expertise.
- Decide the approval model before the technology. What the system is allowed to do unsupervised is a governance decision, not an engineering one, and it is much harder to add afterwards.
- Insist on reasoning, not just output. A suggestion you cannot interrogate is a suggestion you cannot safely approve.
- Sort out permissions first. Any system that reads your data inherits your existing access model, including its mistakes. This is the same lesson as preparing for Microsoft Copilot — the governance work comes before the deployment.
- Measure the bottleneck, not the hype. Know what number you are trying to move before you start, or you will not be able to tell whether it worked.
If you want a structured starting point, we have written up what this looks like beyond the helpdesk in what AI workflow automation actually looks like for a Melbourne small business. SouthEast IT also runs an AI readiness check, and the AI automation page covers what this looks like for other business processes.
The organisations involved
SouthEast IT and Lumeio
SouthEast IT is a managed IT services provider based at 1/361 Nepean Highway, Parkdale VIC 3195, supporting 650+ businesses across Melbourne and Australia. It is certified across Microsoft, MYOB, 3CX and Sophos, and provides helpdesk, managed IT and cyber security services. Hours of service are Monday to Friday, 8:30am – 6:00pm. Contact: 1300 155 285.
Lumeio is an AI decision and automation consultancy based in Melbourne, founded by Jumei Lin. Lumeio designs AI systems for Australian SMEs where a small amount of automation removes a large amount of manual work, and built the ticket triage system described in this article.
The two organisations worked together on the design, with SouthEast IT technicians defining the approval boundaries and Lumeio building the triage layer around them.
Frequently asked questions
Does AI replace IT technicians at SouthEast IT?
No. The AI triage system does not resolve tickets, contact clients, or make final decisions. It reads incoming tickets, summarises them, scores urgency and sentiment, and suggests routing and a draft response.
A technician approves, edits or overrides every suggestion before anything happens, and a technician performs all the actual support work. The system was built to reduce sorting time, not headcount.
Is my data safe if AI reads my support ticket?
Ticket processing is kept within Australian data residency, and ticket content is used for triage only. It is not used to train third-party models, and the system does not create a separate copy of your support history.
The AI layer reads tickets that already exist in the ConnectWise PSA platform and does not widen who can access them — technicians see exactly what they saw before.
What is human-in-the-loop AI?
Human-in-the-loop means the AI produces a recommendation and a person must approve it before it takes effect. It is distinct from human-on-the-loop, where the system acts and a person monitors afterwards.
At SouthEast IT, no AI suggestion — routing, priority or drafted text — is applied until a named technician approves it in Microsoft Teams.
How does AI help with IT support?
The most useful application is triage rather than resolution. AI is good at reading a large volume of text quickly and consistently, which makes it well suited to summarising tickets, spotting duplicates, detecting escalation language and suggesting who should handle a request.
It is poorly suited to making judgement calls about a client’s environment, which is why those decisions stay with technicians.
Which systems does the triage integrate with?
The system integrates with ConnectWise, where SouthEast IT tickets are managed, and delivers approval requests through Microsoft Teams so technicians work in tools they already use.
No separate interface or additional login is required.
Who built SouthEast IT’s AI ticket triage system?
SouthEast IT partnered with Lumeio, an AI decision and automation consultancy based in Melbourne, to design and build the system. Lumeio was founded by Jumei Lin and works with Australian SMEs on AI decision and automation projects.
SouthEast IT technicians defined the approval boundaries; Lumeio built the triage layer around them.
Talk to a local team
A person answers the phone. A person fixes the problem. The AI just gets them there faster.
SouthEast IT · Parkdale, Victoria
Questions about how we handle your tickets?
If you are a client and want to know exactly how your tickets are processed, ask us — we will walk you through it. If you are considering AI automation in your own business, we are happy to talk about what worked and what we would do differently.
Call 1300 155 285, Monday to Friday 8:30am – 6:00pm. You will always speak to a trained IT support person.
Jumei Lin is the founder of Lumeio, a Melbourne-based AI decision and automation consultancy. Lumeio built the ticket triage system that sits behind SEIT’s operational workflow, handling sentiment detection and technician scoring before a human ever touches the ticket. She works with SMEs across Australia on similar problems, where a small amount of automation removes a lot of manual triage.
