Most small businesses can run useful AI for 20 to 60 a month. The cost that hurts is not the subscription, it is the five subscriptions nobody cancelled and the hours spent checking output that was wrong.
Ask what AI costs and you get a list of plan prices, which is the least useful answer. The subscription is the visible part. What actually decides whether AI is cheap or expensive for a small business is the ratio between what you pay and what it replaces, and that ratio varies wildly between two businesses paying the same 40 a month.
Prices move, and every figure below is a shape rather than a quote. Check current pricing before you commit to anything, and treat the arithmetic as the durable part.
The four things you are actually paying for
- The assistant itself. A general model at consumer tier, usually the cheapest useful thing you will buy. Business tiers cost more per seat and add administration you probably do not need under five people.
- Task tools. Transcription, meeting notes, image generation, writing assistants built on top of the same models. These are where subscriptions quietly stack up, because each one is small enough to justify.
- AI inside software you already pay for. Your accounting package, booking system or invoicing app will add an AI tier. This is often the best value, because it works on your real data, and the easiest to miss on a bank statement.
- Usage-based costs. Anything billed per minute, per document or per API call. Cheap at low volume, surprising at high volume, and invisible until the first invoice.
A typical one-person service business ends up with two of these, not four. A business that has been “experimenting” for a year often has six.
Three realistic monthly budgets
The scenarios below assume a service business with one to three people, billing between 60 and 120 per hour. Adjust the rates to your own market.
Lean: 0 to 25 a month
One consumer-tier assistant, used for drafting quotes, replies and chasing. Everything else done in the tools you already own. This covers most of the value for most small businesses, because writing is the task where the time saving is real and the risk is manageable.
Standard: 40 to 80 a month
The assistant, plus one task tool at the point where a specific job is genuinely painful, plus an AI feature inside software you already pay for. Usually transcription for anyone doing site visits, or the AI tier of a booking or accounting system.
Generous: 150 to 300 a month
Multiple seats, a transcription or meeting tool, an automation layer connecting systems, and image or video generation for marketing. Defensible for a business publishing content weekly or running several staff who all need access. Indefensible for a two-person trades business that writes four quotes a week.
The pattern to notice: the jump from lean to generous is rarely justified by the volume of work. It is justified by having several people who all need the same tool, which is a different problem.
Prompt for building a budget
Recommend a monthly AI budget for my business and name what to buy. Business: [what you do], [number of people], [turnover band] Billable hour value: [rate] Tasks I do weekly that take too long: [list 3-5, with rough minutes each] Tools I already pay for: [accounting, booking, invoicing software] Budget I am willing to spend: [amount per month] Rules: recommend one of three tiers (lean, standard, generous) and justify it against the tasks above. For each recommendation, state the tool category, an approximate monthly cost range, and the specific task it replaces. Flag anything on my list that AI will not fix. Do not invent pricing for named products; give ranges and tell me what to check.
The costs that never appear on the invoice
This is the part the pricing pages leave out, and it is usually larger than the subscription.
- Verification time. Every output about money, regulation, safety or a client relationship needs checking. If checking takes as long as writing, you have bought nothing.
- Rework. Drafts that miss the brief, tone that has to be stripped out, structure that has to be rebuilt. Cheap per attempt, expensive per useful result.
- Learning time. The first month with any new tool is slower than the old way. Most abandoned subscriptions are abandoned in that month.
- Tool sprawl. The subscription you stopped using in March and pay for until December. Three of those is a real cost and none of them feel like a decision.
- Data risk. Not a fee, but a cost: client details pasted into a tool whose terms you have not read is a liability you cannot put a number on until it goes wrong.
The honest test is not “is this tool worth 20 a month”. It is “does this tool save me more than an hour of billable work a month”, because at 60 an hour, an hour pays for three tools.
Working out whether it earns its place
Two numbers make the decision for you. What an hour of your time must earn across a year, from the hourly rate calculator, and what your overhead costs per billable hour, from the overhead calculator. A tool that costs less than one billable hour a month is a rounding error. A tool that costs more needs to be saving whole hours, not minutes.
Then watch it for a month with a crude log: what you used it for, and how long the same task took before. Two lines per use is enough. The result is usually surprising in one direction or the other, and either way it settles the argument about renewing.
Prompt for auditing your AI spend
Help me decide which AI subscriptions to keep. For each tool, I will give: name, monthly cost, what I use it for, how often I use it, and roughly how long the task took before. TOOLS: [Tool 1]: cost [amount], used for [task], [frequency], previously took [time] [Tool 2]: ... [Tool 3]: ... My billable hour value is [rate]. Rules: produce a table with monthly cost, estimated hours saved per month, value of hours saved, and net position. Then list, in order, the one tool to cancel first and the reason. Do not invent usage figures I have not given. If a tool has no clear hours saved, say so plainly.
Prompt for a staff AI policy
Draft a one-page AI use policy for a small business with [number] staff. What staff currently use AI for: [e.g. drafting replies, site notes, estimates] What must never be pasted into a public AI tool: [client names, addresses, payment details, contracts, anything under NDA] Approved tools: [list] Person responsible for questions: [name or role] Rules: plain English, under 400 words, no legalese, written as rules staff can follow rather than principles they can interpret. Include what to do if unsure and what happens if the rule is broken, stated calmly. British English. No invented legal obligations; describe our own rules only.
Where AI is a bad buy
Some work should stay away from paid tools entirely, not because of ethics but because of arithmetic:
- Low-volume, high-stakes writing. One client proposal a month. The setup time exceeds the saving, and the risk is concentrated.
- Anything you have already templated. A saved quote template beats a prompt for a job you do weekly. Templates are faster, cheaper and consistent.
- Compliance and safety content. Electrical certification, medical claims, legal wording. The verification cost is the whole task.
- Tasks where being wrong is invisible. Summaries of long documents nobody re-reads, for instance. You pay, you save time, and you never find out what you missed.
None of this argues against AI. It argues for buying less of it than the tool directories suggest, and for being specific about what each subscription replaces. The businesses getting real value from this are not the ones with the biggest stack. They are the ones who can name, in one sentence, the job each tool does.
If you are starting from nothing, the prompt library is a better first purchase than a subscription, because it is free and it tells you which tasks you actually repeat. And the toolkit covers quotes, rates, job profit and invoices without a monthly fee at all.
Frequently asked questions
Is the free tier enough for a small business?
Often, yes, for a business doing a handful of writing tasks a week. The paid tier becomes worth it when you hit usage limits during real work, or when you need a feature like file uploads or longer context. Paying on day one to feel committed is how subscriptions become clutter.
Should I pay annually to save money?
Not in the first month with any tool. Annual pricing usually saves around 15 to 20 percent, which is a poor trade against twelve months of a tool you might stop using in March. Switch to annual once you have two months of consistent use.
What is a reasonable AI budget as a percentage of revenue?
For a service business under 100k turnover, one to two percent of revenue is a sensible ceiling, and most land well below that. Anything above five percent should be replacing paid hours or you are buying novelty.
Do I need to tell clients I use AI?
You need to be honest if asked, and you should never put client data into a tool whose terms you have not read. For most writing support, disclosure is not required. For anything that materially shapes advice a client relies on, say so, because trust is harder to rebuild than a subscription is to cancel.
What is the biggest waste in a typical AI stack?
Duplicate capability: two writing assistants, two transcription tools, or an AI feature inside software that overlaps with a standalone tool. It is rarely a bad product, just the second one bought before the first was cancelled.
Last reviewed and updated on November 10, 2026.
