What should a small business's AI budget actually look like?

There's a strange gap in how AI spending gets discussed. On one side, you've got enterprise consultancies talking about seven-figure transformation programmes. On the other, you've got LinkedIn posts claiming you can run a business on free ChatGPT and a bit of imagination. Neither is particularly useful if you're running a 15-person accountancy firm or a 40-person marketing agency trying to work out what to budget for next year.

For most SMEs in 2026, a sensible annual AI budget sits somewhere between £3,000 and £40,000. That's a wide range. The reason it's wide is the point of this whole article: what you should spend depends far more on what you're trying to achieve than on how big you are.

Here's how to think about it properly.

What's the typical range for SME AI spending in 2026?

Recent benchmarking suggests the typical SME spends about 3 to 5 per cent of its technology budget on AI, though that figure is rising fast. For a 20-person business spending maybe £30,000 to £60,000 a year on software and IT, that's roughly £1,000 to £3,000 at the conservative end. Which, if you're reading the headlines, probably feels low.

It is low. It's also where most SMEs actually are. The gap between what the press talks about and what businesses your size are really spending is enormous, and that's worth remembering next time a vendor quotes you a number that makes your eyes water.

In my experience working with businesses in this bracket, spending falls into three rough tiers:

Tier 1. Testing the water (£3,000 to £8,000 a year). You're activating AI features inside software you already pay for, adding one or two standalone tools at the team level, and running small experiments to see what works. No dedicated headcount. No custom builds. Your main cost is time, not money.

Tier 2. Making it part of how you work (£8,000 to £20,000 a year). You've identified two or three real use cases, you're paying for proper business-tier licences across the team, you've invested in some training, and you've possibly brought in external help for a specific project. This is where most serious SMEs sit within eighteen months of starting.

Tier 3. Treating it as a real capability (£20,000 to £40,000 a year). You have a small number of AI-enabled workflows running in production, you're paying for more sophisticated tools or integrations, you've got someone internally who owns it even part time, and you may be doing some light custom work with a partner. This is where a business genuinely using AI as a commercial advantage ends up.

Anything above that bracket and you're either a much larger business, doing genuinely sophisticated custom work, or being oversold. All three happen.

What does an AI budget actually cover?

This is the part most people get wrong. AI budgets aren't just the monthly cost of ChatGPT Plus and a Copilot licence. If you only account for the software, you'll underspend on the stuff that actually determines whether any of it works.

A realistic AI budget has six cost categories. I'll walk through each one with honest numbers for a 20-person business.

1. Software and licences. The most obvious category. Expect £15 to £30 per user per month for the main business-tier AI tools (Microsoft 365 Copilot, Google Workspace with Gemini, ChatGPT Team, Claude for Work). For a 20-person business that's £3,600 to £7,200 a year if you licence everyone, less if you licence selectively. Most businesses don't need everyone on everything.

2. Training and upskilling. Widely underspent and usually the highest-ROI line in the budget. A proper AI literacy programme for the whole team costs £2,000 to £5,000. The alternative is paying for software nobody uses well, which costs far more in the long run. I'd rather see a business spend 40 per cent of its AI budget on training in year one than buy another tool.

3. Specific tools for specific jobs. Once you're past generic assistants, you'll start finding purpose-built tools for particular functions. A decent CRM-integrated AI add-on might be £100 to £400 a month. A meeting transcription tool that actually works might be £20 per user per month. A niche tool for your sector could be anything from £50 a month to several thousand. Budget £2,000 to £8,000 a year for this category once you know what you need.

4. Data preparation. The quiet cost. Most AI tools work better when your data is in decent shape, and most SMEs have messy data. Tidying up your customer records, your product data, or your document storage so AI can actually use it is usually a one-off project costing £2,000 to £10,000 depending on how much mess there is. Skip this and you'll wonder why the tools seem to work in demos but not in your business.

5. External help. At some point you'll want a second opinion, a strategy sense-check, or help implementing something specific. Budget £3,000 to £10,000 a year for this, or plan to do it as a one-off project. You don't need it every year, but you will need it at least once.

6. Governance and admin. Policies, security reviews, keeping track of what the team's using. Low in monetary terms (£500 to £2,000 a year) but high in attention cost if you ignore it. More on this in a future piece.

Add that up and you get somewhere between £13,100 and £42,200 a year for a 20-person business doing this properly, which maps reasonably well onto the Tier 2 and Tier 3 ranges above.

How should I sequence AI spending over a year?

Three budgets fail out of every four I see, and they mostly fail for the same reason: the business bought the tools before it worked out what it was trying to do. Here's a better sequence.

Months 1 to 3. Small commitment, high learning. Budget about 15 per cent of your annual AI spend here. Activate AI features in software you already pay for. Run a proper skills audit. Pick two or three specific problems worth solving. Don't buy anything substantial yet.

Months 4 to 6. First real investments. Budget about 30 per cent of your annual AI spend here. Now you know what you need, so you can buy it. Proper licences for the people who'll use them. The first specific-purpose tool. Your first training programme.

Months 7 to 9. Building the habits. Budget about 25 per cent. This is when you find out whether what you bought is actually getting used. If it isn't, you need to know why. This phase is less about new spending and more about embedding what you've already got.

Months 10 to 12. Scaling what works, cutting what doesn't. Budget about 30 per cent. Cancel the subscriptions nobody uses. Expand the ones that are working. Start thinking about whether anything needs a more permanent home in a workflow or integration.

This sequencing matters because the biggest AI budget mistake isn't overspending or underspending. It's spending at the wrong time. Committing heavily in month one means committing to choices you're not yet qualified to make. Holding back for twelve months means never actually starting.

How do I know if I'm spending too much or too little?

Three quick sanity checks.

The first is about usage. For every pound of AI software you're paying for, how many hours of actual use is the team getting back? If licences sit unused for more than six weeks, you're overspending. Cancel them and reallocate.

The second is about specificity. Ask someone on your team to explain in one sentence what a given AI tool is doing for the business. If they say something like "it helps us be more efficient", you probably haven't nailed the use case. Which usually means you're spending on hope.

The third is about outcomes. Six months in, can you point to something quantifiable? Hours saved on a specific task. Leads handled that would otherwise have sat. Mistakes caught before they went out. If nothing comes to mind, the problem isn't usually the budget. It's the use case or the adoption.

Pass all three and you're in the right zone. Fail two or more and a bigger budget won't help.

What should I absolutely not spend money on?

Three things, in order of frequency.

Custom AI development you don't need yet. Most SMEs don't need bespoke AI. The off-the-shelf tools are extraordinarily capable, and the temptation to build something custom is almost always premature. If a vendor is pitching you a six-figure custom build and you haven't yet got value from a £20-a-month tool, something's the wrong way round.

Enterprise contracts with unused seats. Buying 50 seats when you need 12 is one of the most common traps. AI vendors are aggressive on annual commitments. Start small, prove value, then scale.

Trend-chasing tools. There's a new "revolutionary" AI tool every week, and most of them won't exist in 18 months. Unless a tool solves a problem you've specifically named, wait. The fear of missing out on AI is real. It's also rarely expensive to be a month late on a tool that actually matters.

So what should I actually do?

If you're starting from nothing, here's a sensible 12-month budget for a typical 20-person business:

• Year-one total: £12,000 to £18,000
• Split roughly 35 per cent software, 30 per cent training, 20 per cent specific tools, 15 per cent external help and data work
• Commit to reviewing the whole budget at month six rather than month twelve
• Plan to spend more in year two than year one, not less

If you're already spending but not sure you're getting value, run the three checks above and be honest with the answers. The fix usually isn't a bigger budget. It's better use of the one you've got.

An AI budget isn't a way to keep up with what other businesses are doing. It's a way of turning specific business problems into specific outcomes, one line item at a time. Build it like that and the number at the bottom takes care of itself.