AI consultant, in-house hire or ongoing support: which fits a business of 20 to 200 people?

For an Australian business of 20 to 200 people, the right choice depends on how much AI work you have and who will keep it running once it is built. As at November 2026, a one-off consultant suits one clear project your team can run afterwards. An in-house AI engineer suits a business with a full-time pipeline of technical work. Ongoing support suits a business that wants a regular flow of improvements, reviewed and kept working as AI tools change.

Published Australian consultant rates in 2026 run from about A$150 to A$600 an hour, and some firms quote fixed prices instead. An in-house AI engineer costs about A$180,000 to A$220,000 a year all in at Melbourne pay levels; Robert Half's 2026 median salary for the role is A$175,100 before super. Ongoing support comes as a monthly retainer of hours, or as a dedicated specialist backed by a wider team.

What are the three ways to get AI work done?

There are three routes.

1. Bring in a consultant. An outside firm or freelancer scopes one defined project, builds it and hands it over. You pay by the hour or a fixed price.
2. Hire in-house. You employ an AI or automation engineer who builds and looks after your tools as a member of staff.
3. Pay for ongoing support. A monthly fee covers continuing work, through a retainer of hours or a dedicated specialist with a wider team reviewing each build.

Coverleo sells the third kind, so it has a stake in the answer. Every cost below is a published figure with its source named, or working you can check.

What does each cost in Australia in 2026?

A consultant

Team 400's April 2026 guide to Microsoft AI consulting rates in Australia puts the Big Four firms at A$350 to A$600 an hour, mid-tier specialist AI firms at A$250 to A$400 and freelancers at A$150 to A$300, all ex GST. Lumeio's Melbourne price guide puts independents and boutiques at A$200 to A$400 an hour. Flowtivity's 2026 Australian pricing guide gives A$150 to A$400, with most experienced consultants at A$200 to A$300.

Some firms quote a fixed price instead. Source Digital's guide to AI consulting costs recommends fixed-price quotes and says its own small-business projects average A$8,000 to A$15,000. Lumeio sells a fixed-price audit at A$1,950 plus GST.

To compare with a salary: at A$200 to A$300 an hour, A$180,000 buys 600 to 900 hours of consultant time. That is about 15 to 22 weeks of full-time work.

An in-house hire

Robert Half's 2026 salary guide for AI engineers in Melbourne puts the role at A$154,500 at the 25th percentile, A$175,100 at the median and A$200,850 at the 75th, before super. Robert Half describes its 25th percentile as someone new to the role or with limited experience. In Sydney, the median is A$185,300.

Employer costs sit on top of salary. Working from the 25th percentile to the median in Melbourne:

• The salary is A$154,500 to A$175,100.
• Super at 12%, the guarantee rate since 1 July 2025 under the Superannuation Guarantee (Administration) Act, adds A$18,540 to A$21,012.
• Workers compensation at WorkSafe Victoria's average premium of 1.8% for 2026-27 adds about A$3,100 to A$3,500. Your own rate depends on your industry.
• A laptop, software, AI subscriptions and training add about A$5,000 to A$10,000, an estimate made for this article.

That comes to about A$181,000 to A$210,000 a year.

Payroll tax applies only once your total Australian wages pass your state's threshold. In Victoria, the State Revenue Office sets it at A$1 million a year, with a rate of 4.85%. If your payroll is over that, this role adds about A$8,400 to A$9,500, for a total of about A$190,000 to A$219,000. Larger payrolls lose some or all of the threshold and can pay a surcharge, so ask your accountant for your own figure.

Year one costs more if an agency fills the role. Certus's published Australian terms set a fee of 25% of the salary package, including super, for packages under A$200,000. At the 25th percentile (A$154,500 plus super, or A$173,040) that is about A$43,000. At the median (A$175,100 plus super, or A$196,112) it is about A$49,000. Year one then comes to about A$224,000 to A$259,000 before payroll tax, or about A$233,000 to A$268,000 with it.

A mid-level automation specialist costs less. Robert Half's nearest title, Automation Engineer, runs from A$113,300 at the 25th percentile to A$133,900 at the median in Melbourne. On the same working, that is about A$134,000 to A$163,000 a year, or about A$140,000 to A$170,000 with payroll tax.

Ongoing support

Synap's 2026 pricing guide lists retainers for its fractional AI adviser service from A$1,200 a month for businesses with up to 25 staff to A$15,500 a month for businesses with 100 or more, plus GST. Its retainers run at 8, 16 or 24 hours a month.

For a dedicated specialist with a team behind them, ask for the full monthly cost in writing and what it includes. Ask, too, whether larger builds are quoted separately.

When is a one-off consultant the better choice?

Choose a consultant when the work is one bounded project and someone inside your business will own the result. A month-end report pack or a supplier bill workflow would fit. The owner needs time to run it every week and enough skill to notice when its output drifts.

A consultant is also a cheaper way to test an idea before you commit to a salary. At Source Digital's average of A$8,000 to A$15,000, a project costs less than a tenth of a year's in-house cost.

Ask what upkeep will cost before you sign. Flowtivity says ongoing maintenance "typically adds 15 to 25% of the initial build cost annually". On a build of A$15,000, that is A$2,250 to A$3,750 a year. Ask who will do that work, and at what rate.

If you haven't chosen the AI tool yet, Copilot, ChatGPT or Claude: which suits an Australian small business? compares the three.

When is an in-house hire the better choice?

Hire when AI is part of what you sell, or when you can see years of technical work ahead. A software business building AI into its product is the clearest case. So is a business with enough builds queued to keep an engineer busy every week.

An employee works only on your priorities. Their knowledge of your systems grows every month they stay, and they are on hand when something breaks.

You also need someone senior who can set priorities and judge the work. Without that, nobody checks what the engineer builds.

Expect the role to take time to fill. In Robert Half's July 2025 survey of 500 Australian hiring managers, 99% of employers hiring AI talent said they were running into hurdles. The biggest were high salary expectations (41%) and a scarcity of qualified candidates (36%).

Plan for the day the person leaves. Ask for each build to be documented as it is made, with logins held in your business's name, so a second person could pick it up.

When does ongoing support make more sense?

Ongoing support fits a business with many small finance and operations improvements and nobody to own them. Bank reconciliation rules and a weekly cash report are two examples. Each job is too small to justify a salary. Together they are more than one project. What AI can do with Xero and MYOB right now shows what is possible in your books today.

It also fits a business that wants someone to keep its tools working as the AI models underneath them change.

Ongoing support costs the same in a quiet month. On a retainer of 16 hours a month, a month where you use 4 still costs the full fee.

Read the contract for four things: who owns what is built, what happens to unused hours, the minimum term and how you leave. Ask who covers the work when your usual person is on leave.

Why do AI tools stall after handover?

In July 2024, Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. Three things can stop a tool once the builder has gone.

The first is a misunderstood problem. RAND's August 2024 report on the root causes of AI project failure notes that "by some estimates, more than 80 percent of AI projects fail". The first root cause it lists is misunderstanding the problem the project is meant to solve. Before anyone builds, write down who does each step, what they check and where it goes wrong.

The second is a tool that doesn't learn. MIT NANDA's July 2025 report, The GenAI Divide: State of AI in Business 2025, said "most GenAI systems do not retain feedback, adapt to context, or improve over time". Its findings are preliminary and come mostly from large organisations' own accounts. When your approval limits or chart of accounts change, someone has to update the tool.

The third is a retired model. Anthropic's model deprecations page says "applications relying on Anthropic models may need occasional updates to keep working", and requests to a retired model fail. Anthropic gives at least 60 days' notice. It retired Claude Sonnet 4 on 15 June 2026 and has set 30 November 2026 for Claude Sonnet 4.5. OpenAI's deprecations page says it gives at least six months' notice for generally available models, and it shut down chatgpt-4o-latest on 17 February 2026. Someone has to read those notices, move the tool to the new model and test that it still gives the right answers.

Why most Australian businesses aren't getting value from AI yet looks at what the 2026 data says.

Five questions to ask before you choose

Ask these of a consultant, a job candidate or a support provider.

1. What will it cost over a year, and how predictable is that? Include upkeep, agency fees and payroll tax. Ask what you pay in a quiet month.
2. Who does the work, and how well will they know your business? A named person who learns your processes needs less briefing each time.
3. Who reviews each build before it goes live? Someone other than the builder should test it on real cases. "Test and monitor" is one of the six practices in the National AI Centre's Guidance for AI Adoption, according to a November 2025 summary by law firm Allens.
4. Who keeps it running once it is built? Name the person who fixes faults, updates the tool when your process changes and moves it to a new model before the old one retires.
5. Who owns what is built? The code, prompts, workflows, documentation and accounts should sit in your business's name. Check what data each tool can see as well; Is it safe to put business data into ChatGPT? has a checklist.

Whichever route you choose, keep people on the decisions. AI tools prepare work: they draft, match, sort and flag. People send, approve and pay. If software will make or help make decisions about people, read about the Privacy Act's new rule on automated decisions. Take legal and tax questions to your own lawyer or accountant.

Where Coverleo fits

Coverleo sells ongoing support in two forms. An AI Business Specialist + works full time on your processes, with Coverleo's AI practice reviewing each build. The AI Expert Plan is a set number of hours each month for creating, updating and consulting on your AI tools. The AI Business Specialist + page explains how the specialist works, and the pricing page gives an indicative estimate for either.