AI Engineering 8 min read

AI consultancy, AI automation agency or software house: which one do you need?

What each kind of firm hands you, how it charges, where it goes wrong, and five questions to ask any of them before you sign.

Founder and Head of AI Engineering

Hire an AI consultancy when you do not yet know which jobs AI should take on. Hire an AI automation agency when the job is clear, simple and fits tools like Zapier, Make or n8n. Hire a software house when you already have a written spec and someone in-house to own it. If you need the thinking and the build, look for one team that does both, with the same person on each.

The three labels overlap more every month, because consultancies now build, automation agencies now advise and software houses now sell AI. What still separates them is what each one hands you at the end, how it charges and what happens when the thing it built breaks. This guide covers all three, the questions to put to each, and where we fit. For what AI work costs in Australian dollars, see our guide to AI consulting costs in Australia.

What does each one actually do?

AI consultancyAI automation agencySoftware house
Starts fromYour business: which jobs to give AI, and which to leave aloneA workflow you can already describeA written spec
Hands youAn assessment and a roadmap, sometimes a pilot or a buildWorking automations on a no-code platform, with a model step insideCustom software built to the spec
Usually chargesDay rates or a fixed-fee assessment, then a project priceA fee per workflow plus the platform's own subscription, often with a monthly care planFixed scope, or time and materials
Where it goes wrongThe roadmap is never builtLong chains of steps that break without anyone noticing, per-task fees that grow, accounts held in the agency's nameIt builds what the spec says, and nobody tested whether the AI is right often enough
Best forA business that knows AI should help and not yet whereLow-volume, linear jobs between tools you already pay forA product or platform you will own and run yourself

None of these is the wrong kind of firm. Each is the wrong pick for some jobs, and most of the expensive mistakes we see come from hiring the right firm for the wrong job.

When is an AI consultancy the right call?

When you can name the problem and not the fix. "Our sales team spends Monday building the pipeline report" is a consultancy question, because the answer might be an AI agent, a HubSpot workflow, a cleaner pipeline or nothing at all. A good consultant works through which jobs AI should take on, what happens when it is wrong, who gets to see which data and what each job costs to run every month at your real volume.

The risk is paying for thinking that never ships. Ask any consultancy to show you a system it built that is still running, and to name who built it. If the people who write the roadmap hand it to someone else to build, or to nobody, the roadmap is the product. Our guide to choosing an AI implementation partner in Australia has the full question list.

When does an AI automation agency make sense?

When the job is a short, predictable chain: a form comes in, a model tidies or classifies it, a record is created, someone is told. Platforms like Zapier, Make and n8n do that well and quickly, and an agency that knows them can have it running in days. For a small team with a few hundred runs a month, that is often the cheapest good answer.

Two things change the maths as you grow. The first is billing. Zapier counts every successful action step as a task against your plan's limit, and n8n's cloud plans are priced on monthly workflow executions, so the bill rises with volume. The second is how many steps it takes. On a project this year, syncing a client's project management tool with HubSpot, our team worked out that the same two-way sync would have needed 71 triggers in Zapier. We built a direct integration between the two systems through their APIs instead, with an alert when it fails.

The quieter problem is what happens when an automation stalls. One client told us at kick-off that their old zaps would bank up when nobody was watching, then release all at once, so every record landed with the same timestamp and their reporting went wrong. Whoever builds the automation should tell you who notices when it stops and how. If the platform account sits in the agency's name, ask how you get it back. Our post on choosing between a HubSpot AI agent and a simple workflow covers the cases where no AI is needed at all.

When do you need a software house?

When you know exactly what you want built and you have someone in-house who will own it afterwards: a customer-facing app, a portal, a product you sell. A software house brings engineering depth that most consultancies and automation agencies do not have, and a spec protects both sides on scope.

The gap is usually the AI part. A spec can say "the system reads the uploaded documents and fills in the loan record", but it rarely says how often the reading has to be right, what happens with the document it has never seen, or who checks the output. Unless those are in scope, they are nobody's job. Ask a software house how it will test the AI on your real cases before launch, and what it will cost to run each month at your volume.

Which kind of firm fits the job
Choosing between an AI consultancy, an automation agency and a software houseThree questions in order. First: do you know which job AI should do? If no, start with an AI consultancy and an assessment. If yes, second: is it a short chain between tools you already use, at low volume? If yes, an AI automation agency. If no, third: do you have a written spec and someone in-house to own the system? If yes, a software house. If no, a team that scopes and builds, with the same person on both. Do you know whichjob AI should do? A short chain betweentools you already use?at low volume A written spec and anin-house owner? yes no no yes yes no AI consultancyan assessment first,with running costs Automation agencyZapier, Make or n8n,in your own accounts Softwarehouse One teamthat scopesand builds

Simplified. Many firms now sell more than one of these, so judge each by what it has built and still runs, whatever it calls itself.

Not sure which kind of help your idea needs?

Talk to Matt about the job your team keeps working around, and our AI engineering services.

What should you ask each one before you sign?

Put the same five questions to every firm on your list, whatever it calls itself, us included.

AskA good answerA warning sign
Can you show me something you built that is still running?A named system, what it does, how long it has been live, and the person who built it on the callDemos, pilots and a case study deck
Is the person scoping this the person building it?Yes, or a named builder in every scoping session"Our delivery team will pick it up after you sign"
What will it cost to run each month at our volume?A figure worked out from your numbers, including platform fees and model usageA build price and silence on running cost
Whose accounts does it run in?Yours: your platform login, your API keys, your repositoryThe agency's account, "for convenience"
Who notices when it breaks, and how?An alert, a log of every run, a named person and a way to roll back"It hasn't broken yet"

The ownership question matters more in Australia than many overseas guides suggest. Under Australian law a contractor generally owns the copyright in code it writes unless the contract assigns it to you, so get that in writing before work starts. Our post on how we test an AI agent before it goes near a customer shows what a pass mark looks like when you ask about testing.

What does each one cost?

Every firm prices differently and few publish their numbers, so compare quotes on the same two figures: what it costs to build, and what it costs to run each month afterwards. An automation agency's build is often the cheapest of the three, but the running cost includes the platform subscription and grows with every task or execution. A consultancy's assessment can look expensive next to a quick automation, and it is the only one of the three whose job includes telling you not to build.

Ours are published. AI projects start from A$10,000 and ongoing AI partnerships start from A$8,500 a month, both ex GST. Claude usage runs on your own Claude Console account, so Anthropic bills you directly and the account stays yours. What a custom AI build costs an Australian business breaks down what moves both numbers.

Where does Neighbourhood fit?

Closest to a consultancy that builds. We work out which jobs AI should take on, put a monthly running cost on each, then build and run the ones that pay their way, on Claude. Your PM is your builder: the person who scopes the work in the room is the person who builds it, so nothing is lost in a handover.

Three builds show what that looks like. For Flint, a mortgage brokerage, a fact-find reads the documents a borrower uploads, pulls the details into the loan record and flags anything that does not match, and a broker signs off before anything moves. For Plungie, sales gets a short brief on each buyer, written from its own HubSpot and sales data, before the call. For a Sydney pool builder, Claude turns a quote into a branded proposal page and writes up each site visit from the recording.

We fit best when the job crosses your CRM and the systems around it: an agent or integration that reads and writes HubSpot, scoped, built and looked after by the same people. And if your HubSpot data is a mess, the first spend should be the clean-up, because any AI system inherits whatever is wrong in the CRM.

Trav on which steps in an AI system need a person and which do not, the question that separates a demo from a system you can run. Watch it on YouTube.

Frequently asked questions

What is an AI automation agency? A firm that builds automated workflows with an AI step inside them, usually on a no-code platform such as Zapier, Make or n8n. It suits short, predictable jobs between tools you already use.

What is the difference between an AI consultancy and an AI automation agency? A consultancy starts from your business and works out which jobs AI should take on, and what each would cost to run. An automation agency starts from a workflow you can already describe and builds it. Many firms now do some of both.

Should I hire a software house to build an AI system? If you have a written spec and someone in-house to own the system, yes. Make sure the scope includes testing the AI on your real cases and a monthly running cost, because a spec rarely covers either.

How much does an AI automation agency cost in Australia? Pricing varies by firm. Compare the build price and the monthly running cost, which includes the platform subscription and grows with volume. Neighbourhood's AI projects start from A$10,000 ex GST.

Who should own the accounts an AI system runs in? You should. Ask for the platform login, the API keys and the code repository to be in your business's name, and for the contract to assign copyright in any code to you.

Read next

Sources


Neighbourhood builds AI systems on Claude for Australian businesses, from the first assessment to running them in production, as part of our AI consulting work. We have worked with 300+ businesses since 2015.

Got a job your team keeps working around?

Talk to Matt about it. He can explain how we work and whether our AI engineering services fit what you need.

Neighbourhood

Neighbourhood is a HubSpot Diamond Partner in Brisbane. We build AI systems and the revenue operations they run on, for businesses across Australia and New Zealand.