Paying for AI: what small providers should budget for
Licences are the small part. Real prices for the main tools, plus the costs nobody quotes you — staff time, the DPIA, training and the checking that never stops.
Most articles about the cost of AI quote you a licence fee. The licence fee is rarely the thing that decides whether it was worth it.
This page has two parts: what the tools actually cost, and what else you will spend.
What the licences cost
Prices read in September 2026. Where a supplier would not show us sterling, we have converted and said so — always confirm against your own invoice.
| Plan | Cost per user per month | Notes |
|---|---|---|
| Microsoft Copilot (add-on) | from £13.80 | Paid yearly, excluding VAT. Needs a qualifying Microsoft 365 plan underneath |
| Microsoft 365 Business Standard with Copilot | £18.10 | Paid yearly, ex VAT. Bundle, not add-on |
| Microsoft 365 Business Premium with Copilot | £24.60 | Paid yearly, ex VAT |
| Claude Team | about £15 | Billed annually; about £19 monthly. Converted from US dollars |
| ChatGPT Business | about £18 | Billed annually; about £22 monthly. Teams of 2 to 200. Converted from euros |
| Otter Business | about £15 | Billed annually. Converted from US dollars |
Two traps in that table.
Microsoft’s prices exclude VAT. Every other figure here is the headline price. If you budget £13.80 and pay £16.56, that is a 20% surprise across every seat.
Microsoft Copilot is an add-on. The £13.80 sits on top of a Microsoft 365 licence you must already hold. The real number is both.
For a service with 12 office and management staff, the licence line lands somewhere between £2,000 and £3,200 a year. That is genuinely affordable, and it is also not the number that matters.
What else you will spend
Before you start: the compliance work
You need a DPIA before AI touches anything about a real person. Not after the pilot — the law says “prior to the processing”.
Done properly for the first time, budget two to four days of a manager’s or data lead’s time. It is faster the second time, because you are editing rather than starting. Add a policy on top: half a day using our one-page template, longer if it needs board sign-off.
If you buy that in, consultancy for a DPIA and policy typically runs to low four figures. If you do it yourself, the cost is real but it is time, not cash — which for most small providers is the harder currency.
Before you start: choosing
Reading supplier terms properly takes longer than anyone expects. We spent most of a day on four tools, and still could not confirm whether Otter offers a data processing agreement. Budget a day for shortlisting and supplier questions, using our question sheet.
At the start: training
Staff need to know which tools are approved, what must never be typed in, and how to check output. That is not a course; it is a team meeting and a printed reminder. Budget an hour per member of staff, plus whoever prepares it.
For 30 staff at roughly £14 an hour including on-costs, that is about £420 of paid time — probably more than your first month of licences.
Forever: the checking
This is the cost that is always missed, and the one that decides whether AI saved you anything.
Every AI draft needs a person to read it before it is used. CQC’s principles call this human oversight, and its AI page asks for “monitoring outputs” under Regulation 17. That time does not go away after the pilot.
The honest arithmetic: if a care worker saves six minutes writing a visit note and spends four minutes checking the draft, you saved two minutes, not six. If the draft is wrong often enough that they re-read the whole thing carefully, you may have saved nothing at all — and you are still paying the licence.
Measure it rather than assuming it. Our admin time tracker and measuring a baseline exist for this.
Sometimes: integration
If a tool has to put text into your care record system, ask early whether that is a supported integration, a copy and paste, or a piece of paid development work. The answer changes the cost by an order of magnitude, and suppliers are often vague about it until you sign.
Where the money might come from
Nonprofit discounts. OpenAI’s pricing FAQ says nonprofits can get “up to a 75% discount on ChatGPT Business or ChatGPT Enterprise”. That would take Business from roughly £18 to under £5 per user per month, which changes the picture entirely. The FAQ does not say whether UK charities qualify, so confirm before you budget on it. Microsoft and Google both run nonprofit programmes; ask.
What you already pay for. If you hold Microsoft 365 Business Premium, you are paying for capability you may not be using. Before buying anything new, find out what your existing licences already include. This is the cheapest AI budget available and it is routinely overlooked.
Framework buying. Some care-specific suppliers sell through the government’s G-Cloud framework, which can be simpler than a direct procurement for a small organisation.
A rule of thumb
For a first AI tool in a small care organisation, expect the first year to cost roughly three times the licence fee once you count the DPIA, the policy, the training and the checking. After year one, the ratio improves sharply, because the governance work is done and only the checking continues.
If that sounds like a lot for a chatbot, that is the right reaction, and it is the argument for starting with one task rather than a strategy.
Before you sign anything
- Confirm whether the price includes VAT, and whether it needs another licence underneath.
- Check the minimum term and what happens if you stop. Annual billing is usually the cheaper price and the longer commitment.
- Ask about the nonprofit discount, in writing.
- Ask what integration actually costs.
- Work out how you will know whether it paid for itself, and write that down before you start — otherwise you will be asked in a year and have nothing but impressions.
If you cannot say how a tool pays for itself, that is not a reason to refuse it. It is a reason to trial it small, measure it, and decide afterwards.