https://www.noxia.co.uk/field-notes/the-two-hundred-pound-floor · printed from noxia.co.uk · sources checked 20 September 2026
Field note · Buying AI
The £200 floor: below this, buying the tool costs more than the tool.
The cheapest software is not the cheapest decision. There is a floor below which a tool cannot pay for itself no matter what it costs, because the cost is not the subscription — and most of the AI being sold to small professional firms right now sits underneath it.
- Buying AI
- Pricing
- Getting started
- Total cost of ownership
The short answer
Buying a tool costs more than its price: somebody must evaluate it, onboard the firm onto it, and supervise its output for as long as it runs. Put realistic hours against those three and they come to roughly £2,400 in the first year for a small professional firm — about £200 a month. A tool saving less than that loses money at any subscription price, including free. The floor is our arithmetic, not a published figure, and the inputs are shown so you can move it.
On this page · 6 sections
A £29-a-month tool and a £290-a-month tool are usually compared on price, and the comparison is almost always wrong, because the subscription is the smallest number in the decision.
Three costs arrive with any tool and none of them appear on the invoice.
Choosing. Somebody shortlists, sits through demos, reads the terms, and carries the risk of being the person who picked it. In a ten-person firm that is a partner or an office manager, and it is rarely less than a day spread over a fortnight.
Onboarding. Accounts, data, permissions, the first week of everyone doing it wrong. This is the cost everybody underestimates, because the vendor's onboarding is free and the firm's is not.
Supervising. The one that never ends. Any tool producing output a client or a regulator might see needs somebody reading that output. An hour a week is modest — it is ten minutes a day — and over a year it is more than both other costs combined.
What the floor is actually for
It is not an argument against cheap software. It is an argument against small software — tools whose best case is saving twenty minutes a week.
Run it the other way. A tool has to save more than 80 hours' worth of value in year one to break even before you have paid a penny of subscription. At the illustrative £30 an hour that is £2,400, or about £200 a month of genuine saving. Twenty minutes a week saved is about 17 hours a year — roughly £520. It loses £1,880 in year one and it would still lose money if the vendor paid you to take it.
This is why "it's only £29 a month, just try it" is the most expensive sentence in small-firm software procurement. The £29 was never the cost.
Change the inputs, move the floor
Every number above is ours and none of them is a law. The floor moves, sometimes a long way:
- It falls when the tool needs no data migration, when one person uses it rather than the firm, and when its output is internal and nobody has to check it. A transcription tool that one person uses on their own notes has almost no onboarding and no supervision. Its floor might be £40.
- It rises sharply when the output is client-facing or regulated, because supervision stops being an hour a week and becomes a process with a record. For anything touching advice, the new automated-decision rules put a floor under supervision that is not negotiable.
- It rises again when the tool is priced per use rather than per month, because the number you modelled is not the number you will pay. That is the per-minute pricing problem, and it is the same trap from the other end.
The practical version
Before buying anything, write one sentence: "This saves [who] [how long] [how often], and the person who checks its output is [name]." If you cannot complete it, the tool is below the floor whatever it costs. If you can, multiply it out and compare it with 80 hours, not with the subscription.
The triage tool asks six questions that amount to this, and it tells people not to automate more often than it tells them to. The missed-call calculator does the same arithmetic for one specific case, and it can output "buy nothing". Both exist because the honest answer to "should we buy this" is usually no, and a tool that cannot say no is a brochure with input fields.
What this does not tell you
It is a model, not a measurement. We have not surveyed a hundred firms and derived 80 hours; we have put defensible hours against three real activities and shown the multiplication, which is a different and weaker kind of claim. If you think choosing takes you two hours rather than eight, change it — the structure survives, the number moves, and that is the point of publishing the inputs rather than the conclusion.
It also says nothing about tools bought for reasons other than saving time: winning work, satisfying an insurer, or being able to answer a question you currently cannot. Those are bought against a different number, and the floor does not apply. Software that sells rather than saves is the other half of this argument, and what our own infrastructure costs is the reminder that the invoice was never the expensive part.
Questions people actually ask
What is the real cost of buying business software?
The subscription plus the hours: evaluating and choosing it, onboarding the firm onto it, and supervising its output for as long as it runs. On our model those come to about 80 hours in the first year for a small professional firm — roughly £2,400 at £30 an hour of internal cost, or about £200 a month, before any subscription is paid.
When is software too cheap to be worth buying?
When the saving it produces is smaller than the cost of adopting it. A tool saving twenty minutes a week saves about 17 hours a year; against roughly 80 hours of adoption cost it loses money in year one even if the software is free. Price is not the test — size of saving is.
Does the £200 floor apply to every tool?
No. It falls for a tool one person uses on their own work with no data migration and no output anyone has to check — that floor might be £40. It rises sharply where output is client-facing or regulated, because supervision becomes a process with a record rather than an hour a week. The inputs are published so the floor can be recalculated for a specific case.
Where does the £200 figure come from?
It is our own arithmetic, not a published statistic: roughly 8 hours choosing, 12 onboarding and 52 supervising, at an illustrative £30 an hour of internal cost. Every input is stated so it can be disagreed with. We would rather publish a model you can argue with than a number you cannot check.
Sources
- Noxia’s own model. Roughly 8 hours to choose, 12 to onboard and 52 to supervise in the first year — 80 hours — at an illustrative £30 per hour of internal cost, giving £2,400 in year one, about £200 a month. None of these hours is measured from a survey. They are stated so they can be disputed, and the conclusion moves with them. — our own arithmetic, labelled as such. This is an opinion with its working shown, not a finding
- The comparison case — twenty minutes a week saved is about 17 hours a year, roughly £520 at the same rate — is arithmetic from the figures above. — arithmetic from our own stated inputs
- The observation that supervision becomes non-negotiable where output is regulated rests on the safeguards in Article 22C of the UK GDPR as substituted by section 80 of the Data (Use and Access) Act 2025, in force 5 February 2026, which require human intervention to be available on significant automated decisions. legislation.gov.uk ↗ — primary for the legal point; the costing of it is still ours re-checked every 6 months
Checked 20 September 2026. Next scheduled check 19 March 2027. Numbers that move — leaderboards, live indices — are re-checked every 30 days; annual datasets and rules in force every six months; dated research once a year. If something here has gone stale before we got to it, tell us and we will correct it and say what changed.
Cite this note
Noxia, “The £200 floor: below this, buying the tool costs more than the tool”, Field notes, 20 September 2026; sources checked 20 September 2026. https://www.noxia.co.uk/field-notes/the-two-hundred-pound-floor
Tell us the saving. We will tell you if it clears the floor.
One sentence is enough: who it saves, how long, how often, and who checks the output. We will do the arithmetic in front of you and say whether it is worth building — and if it is not, that conversation cost you an email and saved you a year of a subscription nobody uses.
Talk to us about thisRead next
Buying AI
Software that sells, and software that just costs.
The difference is not the technology. It is whether anyone can name the number it moves before it is bought — and whether that number was already being counted.
Buying AI
What this website costs to run. The whole bill, line by line.
Three days of our own Vercel bill, pulled from the billing API. Fifty lines, forty-one of them zero. Serving the site: $0.0000. The biggest line is observability.
Buying AI
Per-minute AI pricing charges you for the thing working.
A price that rises with volume charges you for success. Worse, it makes the quote uncomparable against the only alternative you were seriously considering — a person.