From Rob
This week has been what I’m rapidly learning to call “the solo founder’s rollercoaster”.
It started on a high: I agreed my first paid workshop! Then I looked at my bank balance and wondered where the next one is coming from. I started a new marketing campaign to move beyond just referrals and find fresh leads. I met an old friend who now runs a successful law firm and wants some support. All in one week.
The campaign is where AI earned its place. I used it to research potential customers, then to draft and refine proper paper letters of introduction rather than emails. I’m a slow writer, on my own that would have taken a week, and I’d probably have ended up sending everyone the same letter. Instead it took a day, and each letter was written for the person receiving it.
That’s the efficiency. Whether it works is a different question. There’s nowhere to hide when you need to fill a pipeline. The phone calls start next week, and they’ll tell me how good those letters really were.
Rob
Growth
Expect a modest lift, and find out where your data goes
What they asked. The Federation of Small Businesses surveyed 904 small business owners between 11 and 24 November 2025 about how they use AI and what it has done for them.
What they found. 55% now use AI, up from 20% in 2023. Asked what it had done for them, 59% reported productivity gains, 24% said it had increased revenue and 22% said it had helped them grow. Only 8% said it had helped them reduce staff. The FSB puts the average revenue increase for adopters at 3%.
What’s holding firms back. Worry has risen too: 92% of owners are now worried about AI risks, up from 73% in 2023. The FSB wants standard “model cards” from AI suppliers saying how business data is used and stored, whether it trains their models, and who is responsible when things go wrong. Until those exist, you have to ask.
The honest bit. These are owners’ own reports, not measured results, and a 3% revenue lift is modest. The FSB’s estimate that wider adoption could add more than £42bn a year to the UK economy is an extrapolation, so we’ve left it out of the headline. We could only open news reports of the survey, not the FSB’s own release.
Try this week · about 20 minutes
Pick the AI tool your team uses most. Find three answers in its settings or terms: does it train on what you type, where is your data stored, and how do you delete it? This week’s prompt will read the policy for you. If you can’t find the answers, email the supplier and ask.
Source: Business MK report of the FSB survey, 23 June 2026 (secondary source: The FSB’s own press release wouldn’t load for us, and the full FSB report was described as forthcoming.)
Independent dataNumber of the week
30%
Share of UK businesses using at least one AI technology in September 2026, according to the Office for National Statistics. That rises to 38% for firms with ten or more staff, and is 29% for those with fewer than ten. Among AI users with ten or more staff, 6% said it had reduced their headcount. The FSB’s 55% (above) asked its own members a different question, which is why the figures differ: treat any single survey’s adoption rate with care.
Source: ONS Business insights and impact on the UK economy, 8 October 2026
Your own capacity
Hand AI the data entry, and keep a senior eye on what it produces
The problem. In most small firms the books close weeks after month-end, because someone has to key in and categorise every transaction first. Until then, the owner is making decisions on last month’s guesswork.
What the researchers did. Jung Ho Choi of Stanford and Chloe Xie of MIT surveyed 277 accountants and studied field data from 79 small and mid-sized firms using AI accounting tools, covering hundreds of thousands of transactions. Their working paper came out in May 2025.
What they found. Accountants using AI supported 55% more clients a week, moved about 8.5% of their time from routine data entry to work like client communication and quality checks, and closed the monthly books 7.5 days sooner. Their ledgers were also 12% more detailed. Senior accountants gained more: junior staff were more likely to accept the AI’s output at face value, even when it was flagged as uncertain.
The honest bit. This is US research and a working paper, not yet a peer-reviewed result. The field data came from a company that sells AI accounting software, and firms that chose AI may already have been better run. The pattern to take away is the experience gap: the gain comes from someone who knows the numbers checking the output.
Try this week · about 15 minutes
Ask your bookkeeper or accountant two questions: which month-end steps could AI do for us now, and who checks what it produces? Ask them to show you one item the software flagged last month and what they did about it.
Headwind: The checking hours
Pax8 surveyed 1,150 small business leaders in October 2026, 350 of them in the UK. 55% of decision-makers spend three or more hours a week reviewing, editing or fact-checking AI work. 35% of firms using AI found staff using AI tools nobody had approved in the past six months, and 49% of those paying for AI got a higher bill than they expected. Only 23% of UK firms building their own AI tools have a formal process for reviewing them, the lowest of the four countries surveyed.
Pax8 sells AI and cloud software to small firms through IT providers, so read it as a vendor’s survey. The checking time is real either way: put it in the plan.
Source: Pax8 Q3 2026 SMB AI Pulse Report announcement (published by a company that sells the product or service)
Prompt of the week
Find out what an AI tool does with your data
The problem. Most owners never read the privacy terms of the AI tools their staff use. The answers to the basic questions are usually in there, buried in legal language. Start with the public policy for the assistant you use: ChatGPT (OpenAI’s Europe privacy policy, which covers UK users), Claude (Anthropic’s privacy policy) or Gemini (the Gemini Apps Privacy Hub). These cover personal and individual paid accounts. Team, business and Google Workspace accounts have separate terms, so ask your supplier or IT provider for those.
Below is the privacy policy and terms of service for [tool name]. Answer these questions in plain English for the owner of a small UK business. For each answer, quote the exact sentence it comes from. If the document doesn't answer a question, say "not stated". 1. Is what we type or upload used to train the AI? Can we switch that off, and how? 2. Where is our data stored, and is it sent outside the UK? 3. How long is it kept, and how do we delete it? 4. Who else can see it, including the supplier's staff or other companies? 5. Is there anything here that should worry a business that handles customers' personal data? [Paste the privacy policy and terms here]
Before you use it. Paste only the supplier’s public policy text, never your own data. Check each quoted sentence against the original, because AI can misquote. Policies change, so date your notes, and for anything involving customer personal data get the supplier’s answer in writing.
Disclosure: This newsletter is drafted with Claude, made by Anthropic. Claude’s policy is listed alongside the others for reference, not as a recommendation.
Source: OpenAI Europe privacy policy (ChatGPT, UK users) (published by a company that sells the product or service) · Anthropic privacy policy (Claude) (published by a company that sells the product or service) · Gemini Apps Privacy Hub (Google) (published by a company that sells the product or service)
What changed in the tools
Older OpenAI models switch off in January. Check what’s built on them.
On 14 October OpenAI told developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its developer service (the API) on 14 January 2027, including any custom versions built on them. Its recommended replacement is GPT-6 Luna.
This matters if a website chatbot, an automation or a plug-in your firm uses was built on one of those models. It will stop working on that date unless whoever built it switches it over. The notice is about the developer service; it doesn’t cover the ChatGPT app.
What to ignore. Claims that the newer models are faster or cheaper. They come from OpenAI. What matters is whether your chatbot or automation still gives the right answers after the switch, so ask your supplier to test it on real examples before January.
Source: OpenAI API deprecations page (published by a company that sells the product or service)
Find your 20%.
Want to know where AI would pay back in your business?
Answer four questions and get one small test to try. About three minutes, no email needed. Or tell me which job takes too long.
Subscribe
Get it every week
The Quiet Carriage lands once a week: two short stories about how small firms are using AI, what went wrong, and one thing to try. Free.