Short answer: Use off-the-shelf tools for ad-hoc tasks and common workflows. Build an in-house team when AI is core to your product and you can keep engineers busy for years. Hire an AI agency, such as Hyperfocus Studio in Melbourne, when one specific workflow needs a custom system connected to your own data and tools. Most businesses end up with a mix.
We are an agency, so we have an obvious bias. To balance it, every option gets a fair hearing, every figure is sourced, and there is a section near the end on when you should not hire us.
In this guide
The three routes compared in one decision matrix
When off-the-shelf tools, an in-house team or an agency is the right call
What each route costs in Australia, with the arithmetic shown
What the evidence says about build, buy and partner
The 4-signal test for choosing a route
The hybrid route and what a good handover includes
When not to hire us
Eight common situations and the route that fits each
The three routes at a glance
There are three ways to get AI working inside a business: buy a tool, hire a team, or bring in a partner to build it. Each one wins in a different situation. The table below is the whole post on one screen.
Option | Upfront cost | Time to first value | Control and risk | Best when |
|---|---|---|---|---|
Off-the-shelf tools | Low. Monthly subscription | Days | Low control. Little spend at risk | Tasks are ad-hoc or the workflow is common |
AI agency | Medium to high. Priced per project | Weeks to months | High control if you own the code. Risk depends on the partner | One specific workflow needs a custom, integrated system |
In-house team | High. Salaries start before output | Months, after hiring | Highest control. Risk sits in hiring and idle time | AI is core to your product and the work is continuous |
A mix | Spread across all three | Days for tools, months for builds | High where it matters. Needs one clear owner | You are past your first AI project |
If you want the background on what custom AI is first, start with our guide to custom AI solutions for business.
When are off-the-shelf tools the right answer?
Off-the-shelf tools are the right answer for ad-hoc tasks and for workflows that look the same in most businesses. Drafting, summarising, research, meeting notes, first-pass analysis and standard customer support all fit. For that work, a subscription is faster, cheaper and lower risk than anything an agency or an in-house team can build.
MIT Project NANDA’s State of AI in Business report (July 2025) found that more than 80% of organisations had explored or piloted general-purpose tools such as ChatGPT and Copilot, and nearly 40% reported deployment. Task-specific tools did far worse: 60% of organisations evaluated them, 20% reached a pilot and 5% reached production. The report calls its own figures directional, because they come from interviews with 52 organisations and a survey of 153 senior leaders.
The same report notes that a general tool costing US$20 a month (about A$29 at 1 USD = 1.43 AUD, September 2026) often beats a bespoke enterprise system on immediate usability. It also found that for quick tasks 70% of users preferred AI, while for complex work running over several weeks 90% preferred a human.
Australian guides put the price low as well. Valenor’s 2026 pricing guide puts a small business AI software stack at A$150 to A$500 a month. Remap.AI’s 2026 guide puts off-the-shelf AI tools at A$25 to A$250 a month. Both are published agency guides, so treat them as market claims.
Where tools run out
Tools run out when the work is specific to you. Users in the MIT Project NANDA study gave four reasons for keeping AI tools away from mission-critical work: the tool does not learn from feedback, it needs too much context typed in each time, it cannot be customised to their workflow, and it breaks on edge cases.
About 40% of companies in that study had bought an official AI subscription, yet workers at more than 90% of them reported using personal AI tools for work. Your staff are probably on this route already. Proper licences and clear rules about what data goes in are often the cheapest AI project available.
When is an in-house team the right call?
An in-house team is the right call when AI is core to what you sell, the work is continuous, and you can hire and keep good engineers. If the system is your product or your main advantage, you want the people who understand it on your payroll.
Building internally is also getting easier. In McKinsey’s 2026 State of AI survey, as reported by Banking 4.0 in September 2026, 32% of respondents said their organisation had decided against buying at least one software product or feature because it could be built internally with agentic coding tools.
The counterweight: in the MIT Project NANDA interviews, internally built tools reached deployment about 33% of the time. That figure has limits, covered in the evidence section below.
The risks of this route are mostly about people:
Hiring: recruitment comes before any building. ABS data for the 2024-25 financial year, released in 2026, shows 35% of Australian businesses reported skill shortages across all skill types.
Key-person risk: with a team of one or two, a single resignation can stall the system.
Utilisation: salaries run whether or not there is a project to work on.
If you are an engineer reading this to work out where you belong, there is a fourth option. You can join the studio.
When is an AI agency the right call?
An agency is the right call when one specific workflow needs a custom system that connects to your own data and tools, you want it live in months, and you have no spare engineers to build it. You are buying a finished system, plus the judgement of people who build these for a living.
This is the route the partnership evidence supports. In the MIT Project NANDA interviews, external partnerships reached deployment about 67% of the time. The buyers who did well shared four habits. They demanded deep customisation to their own processes and data. They judged the work on operational outcomes. They treated deployment as something that evolves with the vendor. And they sourced projects from frontline managers, who know where the pain is.
The route has real costs. A day of agency time costs more than a day of salary. The agency does not know your business on day one, so the first weeks go on diagnosis. And if the handover is poor, you swap a staffing problem for a dependency problem. A good agency plans the handover before it writes any code.
For the step-by-step version, read what an AI solutions agency does. If you are comparing firms, we keep an honest shortlist of custom AI agencies in Australia. It includes plenty of studios that are not us.
What does each route cost in Australia?
Tools cost tens to hundreds of dollars a month. A two-person in-house AI team costs roughly A$330,000 to A$570,000 a year on published Australian rates. Agency projects run from about A$8,500 for a small agent project to several hundred thousand dollars for a production system. The working is below, and all figures are in AUD excluding GST.
An in-house team: the arithmetic
Every rate below is published and every assumption is stated, so you can swap in your own numbers.
The team is two people: one senior AI engineer and one mid-level AI engineer.
Permanent salaries come from AI Talent On Demand’s AI engineer salary guide (March 2026): A$165,000 to A$200,000 for senior and A$130,000 to A$165,000 for mid-level.
Super is 12%, the rate used in Bluebird Talent’s 2026 hiring budget guide.
Contract day rates come from the same AI Talent On Demand guide: A$1,100 to A$1,500 for senior and A$850 to A$1,100 for mid-level.
A contractor bills 220 days a year. This is our assumption, so adjust it to suit.
Bonuses, recruiter fees, software, cloud costs and management time are left out.
Team | Working | Cost per year |
|---|---|---|
Permanent pair, base salary | A$165,000 + A$130,000 to A$200,000 + A$165,000 | A$295,000 to A$365,000 |
Permanent pair, with 12% super | Base x 1.12 | A$330,400 to A$408,800 |
Senior contractor | A$1,100 to A$1,500 x 220 days | A$242,000 to A$330,000 |
Mid-level contractor | A$850 to A$1,100 x 220 days | A$187,000 to A$242,000 |
Contract pair | Senior + mid-level | A$429,000 to A$572,000 |
The real first-year figure for permanent staff is higher. Bluebird Talent’s guide (July 2026) puts the fully loaded first-year cost of one A$145,000 mid-level hire at A$203,650 to A$223,750 once super, bonus and recruiter fee are counted. That is 40% to 54% above base salary.
As a cross-check, Latitude IT’s April 2026 salary guide reports contract AI and ML engineers billing A$150 to A$220 an hour.
An agency project: our ranges
Hyperfocus Studio is a Melbourne agency that builds custom AI systems into businesses: a plan to solve the problem, an AI solution that solves it, or both. Our small custom AI agent projects start at A$8,500 (about US$6,000) and medium-size projects at A$15,000 (about US$10,000). These are our published ranges.
Engagement | Range (AUD, ex GST) | Typical timeline |
|---|---|---|
The Plan | A$15,000 to A$45,000 | 2 to 4 weeks |
Small AI agent project, plan included | A$8,500 to A$15,000 | 1 to 3 weeks |
Medium AI solution, plan included | A$15,000 to A$30,000 | 2 to 4 weeks |
Advanced custom AI agent | A$30,000 to A$75,000 | 4 to 8 weeks |
Production AI solution | A$90,000 to A$280,000 | 8 to 16 weeks |
Enterprise AI system | A$280,000 to A$900,000+ | 4 to 9 months |
Care plan | 15% to 25% of build cost per year, billed monthly | Ongoing |
Every project is quoted individually after the Clarity Session. These are ranges and should not be read as a rate card. Our small and medium projects sit within the market’s entry bands. Production and enterprise work sits above the Australian market midpoint, because senior people do the work and the system ships working.
Other published Australian ranges vary widely. Team 400’s 2026 guide lists A$150,000 to A$500,000 or more for a production system. Digital One Agency’s 2026 guide gives A$50,000 to A$300,000. Valenor’s 2026 guide lists light custom work at A$5,000 to A$15,000. These are agency guides, not audited surveys.
If we got something wrong about your studio anywhere in this guide, tell us at hello@hyperfocus.studio. We will fix it fast and owe you a coffee.
The fair comparison is utilisation
On these numbers, one year of a permanent pair costs more than the top of our production build range, and many times the price of a small project. So the question is how much work you have. If you can keep two engineers busy on valuable AI work for three years, a team costs less per project. If you have one or two projects, an agency costs less in total, because you stop paying when the work stops.
Idle engineers rarely stay idle. They build things nobody asked for, which is its own line item.
Every route has running costs. Published guides, including Team 400 and Lanex (both 2026), put maintenance of a custom system at 15% to 25% of build cost per year. With an in-house team, the team is the maintenance. With a tool, it is in the subscription. For a full breakdown, see how much a custom AI solution costs in Australia.
What does the evidence say about build, buy and partner?
Three sources point the same way. Most organisations now buy more AI than they build. In one directional study, partnered builds reached deployment about twice as often as internal ones. And the organisations getting value put most of their effort into people and process.
Source | What it measured | Finding |
|---|---|---|
MIT Project NANDA, July 2025 | Deployment rates for custom AI tools, from interviews with 52 organisations | About 67% with external partners, about 33% for internal builds |
Menlo Ventures, December 2025 | Share of enterprise AI use cases purchased or built internally | 76% purchased in 2025. In 2024 it was 53% purchased, 47% built |
BCG, October 2024 | Where AI leaders direct their resources, from a survey of 1,000 executives | About 70% people and processes, 20% technology and data, 10% algorithms |
How to read these figures
MIT Project NANDA: this is a Media Lab project and is separate from MIT Sloan. The report calls its figures directional. They are self-reported, and the partnership finding is a correlation. Organisations that choose partners may simply be more capable to begin with. The report also found employee usage was nearly double for externally built tools.
Menlo Ventures: the figure covers software and infrastructure spending. Purchased includes off-the-shelf products, so it says nothing about how much work goes to agencies.
BCG: the 10-20-70 split comes from its Where’s the Value in AI report, a survey of 1,000 senior executives across 59 countries. It describes what leaders do, and it suggests whoever builds your system should spend most of their time on your workflow.
One thing the evidence does not say: that custom AI beats off-the-shelf tools. We could not find a rigorous head-to-head study, so we will not claim one.
Generic tools suit ad-hoc tasks. Workflow-specific, integrated systems built with a partner are the ones reaching production.
The 4-signal test: which route fits your problem?
The 4-signal test is four questions about one problem, asked in order. Each answer points to a route. Run it per problem, because a business can have a tool problem and an agency problem in the same week.
Signal 1, specificity: would this workflow look the same in the business next door? If yes, a tool probably exists. If it depends on your own rules, documents and exceptions, it needs custom work.
Signal 2, integration: does the system need to read from and write to your own systems, such as your CRM, ERP or document store? Copy and paste is fine for ad-hoc work. Anything that acts inside your systems points to a build.
Signal 3, continuity: is this one project or a permanent stream of work? One to three defined projects suit a partner. A pipeline that keeps engineers busy for years suits a team.
Signal 4, ownership: is there a named person inside your business who will own the system after launch? If there is nobody, no route works, and the first job is to appoint one.
Your answers | Route |
|---|---|
Common workflow, no integration needed | Off-the-shelf tool |
Specific workflow, integration needed, one to three projects | AI agency |
Specific workflow, integration needed, permanent pipeline | In-house team, or agency first and team second |
No named owner | Pause and appoint an owner |
The hybrid route: start with a partner, hand over to an in-house owner
The hybrid route is an agency build followed by a planned handover to a person on your payroll. You get a partner’s speed for the build and in-house control for the long run. For many mid-sized businesses it is the most sensible path.
Name the in-house owner before the build starts.
The partner diagnoses the problem and builds the system with the owner involved throughout.
The owner runs the system alongside the partner for an agreed period.
The partner hands over everything on the list below.
The partner steps back to a care plan, or steps away entirely.
What a good handover includes
Source code in a repository you control
Accounts, keys and billing in your name
Documentation: how the system is built, plus a runbook for common failures
The evaluation set: the test cases and results, so you can recheck after any change
Monitoring and alerts that go to your people
Training sessions for the owner and the users
A written list of known limits and uses the system was never designed for
Clear support terms after handover, including how to end them
The owner does not need to be an AI engineer. In many businesses it is an operations lead or a product manager with time set aside for the job. Our care plan exists for teams that want us to keep monitoring and upgrading the system after handover. Handover is built into the final step of the Hyperfocus Method.
When not to hire us
Do not hire us if a cheaper route solves the problem. Size is not the test: our small projects start at A$8,500 (about US$6,000), medium-size projects at A$15,000 (about US$10,000), and we build enterprise systems too. Fit is the test. Here is the specific list.
The task is ad-hoc. If the job is drafting, summarising or research, buy a subscription.
A product already does it. If an existing tool covers most of the workflow, configure the tool.
The problem costs less than the fix. Team 400’s 2026 guide suggests problems costing under A$50,000 a year in manual effort rarely justify custom AI. Our small projects start at A$8,500, so check the payback before you call.
AI is your product. If the system is what you sell, build the team and own the capability.
Nobody will own it. A system without an owner decays. Appoint the owner first.
You need the lowest quote. Our production and enterprise work sits above the Australian market midpoint. Other studios publish lower entry points and may suit you better.
You want a demo and nothing after it. We build systems that go into production. A prototype tool or a freelancer will get you a demo for less.
The process changes every week. Settle the process first. Automating a moving target costs twice.
If a monthly subscription solves your problem, buy the subscription. We will be here when you find the problem it cannot solve.
Common situations and the route that fits
Every row below is an illustrative example, not a client. Find the one closest to your situation.
Situation (example) | Best route | Why |
|---|---|---|
A 12-person accounting firm wants help drafting client emails | Off-the-shelf tool | Ad-hoc task, and a person checks every output |
A marketing team wants meeting notes and summaries | Off-the-shelf tool | Common workflow with products built for it |
A logistics firm wants quotes built from its own rate cards and job history | AI agency | Specific rules, and it must connect to pricing data |
A software company is adding AI features to its own product | In-house team | AI is the product and the work never ends |
An insurer wants claims triage across several business units | Agency, then in-house owner | Large, compliance heavy, and needs long-term ownership |
A retailer wants a chatbot for order tracking and returns | Off-the-shelf tool first | Common workflow. Go custom only if the tool cannot reach your systems |
A professional services firm wants staff to search 20 years of internal documents | AI agency | Own data, where permissions and evaluation matter |
A manufacturer has one manual report that takes two hours a month | None. Leave it alone | The fix costs more than the problem |
What to do next
Run the 4-signal test on one real problem. If it points to a tool, buy the tool. If it points to a team, start hiring. If it points to a partner, or you cannot tell, drop your problem with us by text or voice note. A human replies within 24 hours, and a recommendation with scope, price and timeline lands within 72 hours. The first step is a free 15-minute Clarity Session. If the right answer is a subscription, we will tell you that.
Sources
MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (Jul 2025)
Menlo Ventures, 2025: The State of Generative AI in the Enterprise (9 Dec 2025)
Boston Consulting Group, Where’s the Value in AI? (24 Oct 2024)
Banking 4.0 (reporting McKinsey), The state of AI in 2026: there is a big gap between individual gains and enterprise impact (Sep 2026)
Australian Bureau of Statistics, Characteristics of Australian Business, 2024-25 (25 Jun 2026)
AI Talent On Demand, AI Engineer Salary Australia 2026 (25 Mar 2026)
Bluebird Talent, Machine Learning Engineer Salary in Australia 2026 Hiring Budget Guide (8 Jul 2026)
Latitude IT, Highest Paying IT Jobs in Australia 2026 (Apr 2026)
Team 400, How Much Does Custom AI Development Cost in Australia (10 Feb 2026)
Digital One Agency, The Real Cost Of Building A Custom AI Solution In Australia (2026) (6 May 2026)
Valenor, How Much Does AI Actually Cost? A 2026 Pricing Guide for Australian Businesses (22 Mar 2026)
Remap.AI, How Much Does AI Automation Cost in Australia? (2026 Honest Pricing Guide) (14 Apr 2026)
Lanex, AI Development Cost in 2026: Australia Budget Guide (7 Jan 2026)