Comparisons like this are usually written by whoever benefits from the answer. So here is the method first: we're comparing the total cash out the door over three years, at list prices, for the same ten people doing the same work. No productivity multipliers, no "value of time saved", no assumed staff reductions. Those numbers exist but they're guesses, and a guess in a spreadsheet looks exactly like a fact.
The cloud side
Business AI seats in Australia sit between about $30 and $100 per person per month depending on the product — a general assistant at the lower end, a specialist medical scribe or legal drafting tool at the higher end. Take the middle: $75 per person per month.
Ten people, $75 each, twelve months: $9,000 a year. Over three years, $27,000, and the meter keeps running in year four. Hire two more people and it becomes $10,800 a year. Nothing is owned at the end.
Two costs commonly left out of the cloud column, in fairness to both sides: there's usually no hardware to buy, and no one to call at 4pm when it stops working — support is a web form.
The machine side
Ten people means The Practice: a 64 GB machine, installed, from $8,000 including setup on your network, transcription, document search, your templates, and configuration of workflows from our library.
Then the yearly service — roughly $300 a year — a site visit with a health check across every service, model upgrades as better open models are released, and rollbacks if you don't like one. It's optional. The machine runs without it.
Three-year total with the yearly service: $8,900. Without: $8,000, plus whatever your own IT provider charges for the occasional look.
Putting them side by side
- Cloud subscriptions, 10 people at $75/month
- $27,000
- The Practice, installed, plus the $300/year service
- $8,900
- The Practice, installed, no service plan at all
- $8,000
Read that honestly and the headline is not "on-premise is half the price". At ten people the machine costs less than a third of the subscriptions over three years. What differs is what you're holding at the end of it and what happens next.
Where the lines actually separate
Year four. The subscription bill repeats forever. The machine is bought. From year four the comparison is $9,000 a year against about $300 of optional service.
Growth. Every new hire adds $900 a year to the cloud column and $0 to ours. A practice that goes from ten to sixteen people over three years pays about $37,000 in subscriptions instead of $27,000. Our number doesn't move.
Usage. Cloud products increasingly meter heavy use — extra charges for long documents, more transcription minutes, higher tiers for the better model. On a machine you own, a busy week costs the same as a quiet one: nothing.
Ownership. At the end of three years the subscription leaves you with your files and a cancelled account. The machine is an asset on your books, running the workflows we built for you, and it keeps working if we vanish.
When a subscription is the better buy
Below about six or seven people, the arithmetic genuinely favours cloud. Four staff on $75 seats is $3,600 a year — a $5,000 machine plus support takes a long time to catch that, and if privacy isn't a binding constraint for your work, the honest recommendation is a subscription.
It also favours cloud if you want one narrow feature and nothing else, if your team is fully remote with no office to put a machine in, or if you're testing whether AI helps at all. Spend $200 a month for three months and find out. We'd rather you did that than bought hardware you didn't need.
Where the machine wins clearly: eight or more people; anything where the content is confidential by law or professional obligation; offices with poor internet; and any business that expects to grow its headcount faster than its software budget.
The number nobody quotes
There is one more line we can't fill in for you, and it's often the biggest: what the work costs today. If a practice manager spends ninety minutes a day retyping notes, that's roughly a day a week. Put your own hourly figure on it, multiply by three years, and compare that to both columns above. Whichever side of the fence you land on, that's the number that decides whether any of this is worth doing.
Want us to run these numbers for your actual headcount and quotes? Send us what you've been quoted and we'll do the arithmetic in front of you.