🤖

This article is AI-generated, it is researched, drafted and then edited by AI. We're open about this, it's part of how we demonstrate what's possible with AI.

Matters now October 2026 · 4 min read

Only 23% of UK small and medium-sized businesses that build their own AI tools have a formal review for them

A Pax8 survey published on 12 October 2026 found formal, documented review of home-built AI at 23% in the UK, the lowest of the four countries surveyed. It also found shadow AI, tools running without anyone's approval, and AI bills higher than businesses expected.

A
Written by Aether · Edited by Greg Shaw
Aether is Greg's AI writing assistant, named after the Greek primordial force of the bright upper heavens, infinite, radiant, and unconstrained. Every article written by Aether is reviewed before publication.

While every effort is made to ensure accuracy, AI-generated content may contain errors or omissions and is provided for general information only, not professional, legal, or financial advice. Opinions expressed are those of our AI personas using skills created for the very purpose of researching, drafting and writing articles on AI for small businesses. Just AI Talking accepts no liability for any loss arising from reliance on content published on this site.

The takeaway

Building AI tools is now ordinary for small businesses, and keeping track of them is not: in the UK, 23% of those that build have a formal, documented review. Ask the team what they have connected, list every AI tool, automation and agent with an owner, what data it sees and how it is billed, and go through the list every three months. That one page finds shadow AI before it finds you, and makes the next surprise bill easier to trace.

Small businesses have stopped only buying AI tools and started building their own: an automation that sorts the enquiries inbox, say, or an assistant that drafts quotes, or an agent that chases unpaid invoices. A survey from Pax8, published on 12 October, asked 1,150 business leaders in the US, the UK, Germany and Sweden about the AI they use, build and pay for. In the UK, 23% of those that have built something have a formal, documented review1 of it. That is the lowest of the four countries.

Building it makes you the supplier

Buy a tool and the software company looks after it. Build one, even a small one, and that job comes back to the business. Pax8's report puts it plainly: a business that deploys an agent or brings AI into its finance system “has taken on work that used to belong to software vendors: maintaining what was built, securing it, deciding who can change it, and paying for the compute it consumes”.2

In a business of six people, that work usually lives in one person's head. The report names what happens next: tools that become difficult to maintain once the person who built them leaves. More than half of the UK businesses using AI already run something they made themselves. That is no reason to stop, but it is a reason to know what is running.

One-off session

AI Fit Session

Not sure where to start? Over 1 week and three 60-minute 1-to-1 sessions, we map your business, find the quick wins, and build a practical starting point specific to you.

Eight-week programme

AI Foundations

Eight weeks, one coaching session a week, seven recorded modules. By the end you'll have AI working across real areas of your business and a 90-day plan to keep building.

Monthly coaching

AI Growth

Already using AI but want to keep developing it? Each month we work on real tasks from your business together, refining what's in place and building what's next.

£200/mo See AI Growth

The tool nobody approved, and the bill nobody expected

The survey's name for the first problem is shadow AI: tools, workflows or agents running in a business without anyone having approved them. In a small business it usually looks ordinary. A free transcription app someone connected to the shared calendar. An automation a freelancer set up last year that still sends every enquiry through a chatbot. In the UK, 24% of businesses using AI found something like this in the past six months, and another 13% are not sure whether they have any.

What matters is what the tool can see. Data security and privacy exposure was the risk businesses named most often for AI built by staff, in every country surveyed. If an unapproved tool reads customer emails, that customer information has gone somewhere the business never chose.

The second problem arrives on the card statement. Some AI services charge by use, so a workflow that runs on every incoming email, or an agent that keeps working at a task, costs more the more it runs. Pax8 says usage-based charges are the hardest to forecast. In the UK, 39% of businesses paying for AI had a bill higher than they expected, and only 12% keep AI on a budget line of its own. Otherwise a surprise charge hides among the other subscriptions.

No formal review does not mean nobody is checking. In many countries the most common approach is a review that is consistent but undocumented. What is missing is the written version: which tools exist, who owns them, and what happens when that person is on holiday.

How far to trust the numbers

Pax8 runs the marketplace that IT providers use to buy and manage software for the small businesses they look after, so knowing what those businesses are doing with AI is part of its job. The report ends with advice for them on where to focus. The survey counts firms of up to 499 staff, and Pax8 calls results for smaller groups within one country directional. Read the UK figures as a direction, not a measurement.

A review small enough to do this week

  • Ask everyone, without blame, what AI tools they have connected to work email, files, calendars or accounts. Asking is the simplest way to find shadow AI.
  • List every AI tool, automation and agent the business runs, including the ones you set up yourself. Put one person's name against each.
  • For each one, write down what data it can see and what it can do without a person checking first, such as sending an email or changing a record.
  • Check how each one is billed. Where the charge depends on usage, look for a spending limit or an alert and switch it on if the provider offers one.
  • Put a date in the diary to go through the list again in three months.
Frequently asked questions
What is shadow AI in a small business?

Shadow AI means AI tools, workflows or agents running inside a business without anyone having approved them, such as a free app a member of staff has connected to work email. In Pax8's Q3 2026 SMB AI Pulse survey, published on 12 October 2026, 24% of UK businesses using AI had found some in the past six months, and a further 13% were not sure whether they had any. The risk businesses named most often was data security and privacy exposure.

How do I review an AI tool my business has built?

Write down four things for each tool: who owns it, what data it can see, what it can do without a person checking first, and how it is billed. Keep the list in one place and go through it again every three months. For a small business, a single page that someone other than the builder can follow is a reasonable place to start.

Why are AI bills higher than expected?

Some AI services charge by use, so costs rise the more an automation or agent runs, and Pax8 says usage-based charges are the hardest to forecast. In its Q3 2026 survey, 39% of UK businesses paying for AI had received at least one bill or usage charge higher than they expected. Only 12% kept AI spending on a budget line of its own, which makes a surprise charge harder to spot.

How reliable is the Pax8 SMB AI Pulse survey?

It is a quarterly online survey by Pax8, which runs the marketplace IT providers use to buy and manage software for the small businesses they support, and it uses the findings to advise those providers. The Q3 2026 edition covered 1,150 decision-makers at companies with 5 to 499 staff: 400 in the United States, 350 in the United Kingdom, and 200 each in Germany and Sweden. The margin of error for the UK sample is about 5.2 percentage points, and Pax8 says results for smaller subgroups should be read as directional.

Book a call ← Back to October 2026