How much does AI implementation cost?

When the UK government asked businesses what they spent on AI in 2024, among businesses using or planning AI that could provide a spending estimate, the median was about £2,000 and the mean about £19,000 (DSIT).

That gap is the most useful fact about AI implementation cost I know. Most businesses spend very little, a few spend a lot, and the average describes almost nobody. If someone quotes you an "average cost of AI", ask which business they averaged.

The short answer: AI implementation cost depends on what you're implementing. Off-the-shelf business assistants such as ChatGPT, Claude or Microsoft 365 Copilot cost about $20 to $30 a user a month at list price. A custom pilot is a different scale: vendor-published estimates run from roughly $8,000 to $60,000. Departmental deployments run into six figures. Then add running costs, which never stop.

Key numbers:

• Median 2024 AI spend among UK businesses using or planning AI that could estimate it: about £2,000; mean about £19,000 (DSIT)
• Business AI assistants: $20 to $30 per user per month at US list prices (see Copilot vs ChatGPT for business)
• Custom pilot or proof of concept: $8,000 to $60,000, according to vendors who sell them
• Gartner estimates agentic reasoning can cost providers at least five times as much in inference as a basic chatbot interaction (Gartner)

What does AI implementation cost, by type of project?

"AI" covers everything from a $20 subscription to a team of engineers for a year, so a single number is useless. Split it into tiers.

A health warning on the bottom four rows. They come from development firms that sell custom AI, in US dollars, and they don't publish how they were derived. Treat them as a sense of scale, and don't anchor a negotiation on them.

The bigger point is in the third row. For a lot of small businesses, the right first move is a business plan for an assistant, some automation between tools you already pay for, and the time to set it up properly. No build required. Build vs buy AI covers where that line sits.

What do UK businesses actually spend on AI?

Less than the vendor pages suggest. The Department for Science, Innovation and Technology surveyed 3,500 UK businesses with five or more employees between February and May 2025, and published the results in early 2026. Only 16% were using at least one AI technology.

Among the businesses using or planning to use AI:

• 35% didn't know what they'd spent on AI in 2024
• 31% reported spending nothing
• 11% spent under £1,000, 16% spent £1,000 to £9,999, and 7% spent more than £10,000
• Among respondents able to estimate spending, the mean was about £19,000 and the median about £2,000
• 65% expected their AI budget to rise

Two readings matter. First, most businesses doing AI are doing it cheaply, through subscriptions and features inside software they already own. Second, over a third couldn't say what they'd spent, which tells you AI costs are scattered across expense claims, software renewals and nobody's budget line.

Cost also shows up as a barrier. In the same survey, 23% of all businesses cited high cost as something stopping them, and 76% of those rated it significant.

What drives the cost of implementing AI?

Five things move the number far more than the choice of model.

1. Integration. An assistant on its own is cheap. The moment it has to read from your CRM, write to your finance system or sit in your phone line, each connection is its own piece of work.
2. Data condition. If the information the AI needs is scattered, duplicated or out of date, someone has to fix that first. That work is rarely glamorous and rarely in the first quote.
3. Autonomy. A tool that drafts something for a human to check is far cheaper to build and run than one that acts on its own. Approval steps, logging and fallbacks all cost money. See AI agents for small business.
4. Risk and regulation. Handling personal, health or financial data adds design, testing and documentation.
5. People. Training, process change and the hours your own staff spend on the project. These are real costs even when no invoice arrives.

Who does the work matters too. Day rates for outside help vary widely by market and seniority; how much an AI consultant costs breaks down the ranges.

What does AI cost to run after launch?

This is where the AI automation cost conversation usually goes quiet, and where it shouldn't.

Seats are predictable. Usage isn't. Many AI services charge by consumption: tokens, credits or API calls. Anthropic's Claude Enterprise, for example, is priced as a seat fee plus usage at API rates, and OpenAI lets Business and Enterprise customers buy extra credits for more access. A quiet month and a busy month can produce very different bills.

Agents make this sharper. In August 2026 Gartner predicted that AI inference costs per agentic workflow will rise more than fivefold through 2028, and noted that routing a task to an agentic reasoning model costs providers at least five times as much as a basic chatbot interaction. Its analyst warned that defaulting to generic autonomous intelligence leads to "unbounded costs" (Gartner).

Translated for a business owner: cheaper tokens don't mean a cheaper system if the system uses far more of them. Ask any supplier for a running-cost estimate at your expected volume, and at double it.

Then there's upkeep. Models get updated or retired, prompts drift, connected systems change their APIs. Somebody has to own that, either a supplier on a support contract or someone on your team with the time set aside.

What do most AI budgets leave out?

The build quote is usually honest. The omissions are around it.

• Your own team's time. Subject experts reviewing outputs, testing edge cases and answering questions. Often the largest hidden line.
• Training. Without it, licences go unused or get used badly. See AI training for employees.
• Access clean-up. Tools that read company data will surface whatever permissions currently allow. Fixing that first takes time.
• Policy and governance. Rules on what data goes in and who checks what comes out.
• Measurement. A baseline before you start, so you can tell whether it worked. See how to measure AI ROI.
• Exit. What it costs to move your data and workflows if you change supplier.

What I've seen

I've spent over 20 years turning technology into working businesses, and the cost pattern with AI is familiar. The licence or build fee is the visible bit. The expensive bit is the organisation around it.

The cheapest projects start small and specific. One process, one team, one measure of success. The common mistake is a broad "AI programme" with no defined first use case, which is how budgets dissolve into workshops.

Running costs surprise people more than build costs. I run my own lab of AI agents, Ascendancy IQ, doing real work under human direction, and the lesson there matches Gartner's warning. Usage climbs as agents take on more steps, so approval gates and limits are budget controls as much as safety controls.

Unknown spend is the real risk. When a third of businesses can't say what they spent on AI last year, the bigger risk is that nobody is deciding at all. Projects without an owner and a number are also high on the list of why AI projects fail.

How to budget for AI implementation

1. Find out what you already spend. Check expense claims and software renewals for AI subscriptions and add-ons. Most businesses are surprised.
2. Pick one use case with a clear owner and a measurable outcome.
3. Try the off-the-shelf route first. Price business seats for the people involved before you price a build.
4. If you need custom work, fund a fixed-price proof of concept with a pass mark agreed upfront.
5. Get running costs in writing at expected volume and at double it, including usage-based charges.
6. Add the internal lines: staff time, training, data clean-up, policy and measurement.
7. Agree who owns it after launch, and budget support for at least a year.
8. Review quarterly. Cut seats nobody uses and scale what's working.

A worksheet for your first-year budget

Copy these lines into a spreadsheet. Put a named owner and a quoted or measured assumption next to each. Keep VAT treatment and currencies consistent, and separate recoverable tax from the economic cost.

Calculate expected and high-volume scenarios. Avoid counting the same staff review time both as lost benefit and as an additional cost. Carry the total into the ROI worked example; a cost total on its own says nothing about whether the project is worth doing.

Frequently asked questions

How much does AI cost for a small business? Many small businesses start with business subscriptions at about $20 to $30 per user per month. UK government research found the median annual AI spend among businesses using or planning AI that could estimate it was about £2,000, though a small number spend far more.

What is the cost of implementing AI in a business process? If an off-the-shelf tool fits, it's mostly subscriptions and setup time. A custom pilot costs roughly $8,000 to $60,000 by vendors' own published estimates, and a departmental rollout can run to six figures.

Are AI agents more expensive to run than chatbots? Yes, usually. Gartner says routing a task to an agentic reasoning model costs at least five times as much in inference as a basic chatbot interaction, because agents make more model calls and take more steps.

What are the hidden costs of AI implementation? The common ones are your own staff's time, training, cleaning up data and access permissions, governance, and ongoing usage charges. Support and maintenance after launch are often left out of the first budget too.

How much should I budget for an AI pilot? Budget for a fixed, small scope with a clear success measure. Vendor estimates for custom pilots start around $8,000 to $15,000 and run to about $60,000, depending on integrations and data.

Sources

• AI Adoption Research, Department for Science, Innovation and Technology, updated 13 February 2026 (fieldwork February to May 2025)
• Gartner Predicts AI Inference Costs Per Agentic Workflow Will Increase More Than Fivefold Through 2028, Gartner, 17 August 2026
• ChatGPT Business and Enterprise pricing, OpenAI, checked 10 October 2026
• Claude plans and pricing, Anthropic, checked 10 October 2026
• Microsoft Copilot for business pricing, Microsoft, checked 10 October 2026
• AI implementation cost, Folio3 (vendor estimates)
• How much does artificial intelligence development cost?, Cubix (vendor estimates)

If you're putting a number on an AI project and want an independent check on what it should really cost, let's talk.