Why this lesson comes first
Before you spend money or time on AI, you need a filter for the claims you will hear. Some come from careful surveys. Some are forecasts. Some are sales pitches, and some are scare stories. This lesson gives you a simple way to sort them.
We will use three sample firms throughout the course:
- Bean & Bloom: a café with 6 staff
- Okafor Electrical: a sole-trader electrician with one apprentice
- Studio Nine: a hair salon with 4 stylists and a part-time receptionist
Task 1: Check any AI statistic in four steps
Think of a statistic like a second-hand van. Before you buy it, you check the logbook. Do these four checks, in this order:
- Source. Who produced it? Is it an official statistics body, an independent researcher, or a company selling AI?
- Date. When was the data collected? AI changes quickly, so a figure from 2023 may be out of date by 2026.
- Sample. Who was asked, and how many? Was it all businesses, or only members of one group?
- Method. How was it gathered and adjusted? Was it a weighted survey, an online poll, or a model built on assumptions?
Worked example: two figures that look like they disagree
- Office for National Statistics (ONS), published 8 October 2026: 30% of UK businesses said they used at least one type of AI in September 2026. For firms with fewer than 10 employees the figure was 29%, and for firms with 250 or more it was 51%. In September 2023, when the question was first asked, the overall figure was 9%. The figures come from about 9,933 responses, weighted to represent all UK businesses. ONS labels them "official statistics in development".
- British Chambers of Commerce (BCC) with Atos, published 19 March 2026: 54% of firms surveyed were actively using AI. The online survey covered 668 businesses between 12 January and 6 February 2026, and 94% of them had fewer than 250 staff.
Which figure is right? Possibly both. They asked different groups, at different times, with different methods. A weighted national survey of thousands of firms and an online survey of 668 chamber-network businesses will not give the same answer. Do: never quote one of these figures without saying which survey it came from and when.
For Bean & Bloom, the useful fact is the ONS one: in September 2026 about 29% of firms its size used AI. Most did not. Using AI, or not using it, does not make a firm unusual.
Task 2: Sort claims into evidence, forecast, hype or fear
Put every claim you hear into one of these four groups:
- Evidence: something that was measured, with a named source, a date and a method. Example: the ONS figures above.
- Forecast: an estimate of what might happen, based on assumptions. Useful, but not fact.
- Hype: a promise of a result with no measurement behind it. Example: "This AI tool will double your bookings."
- Fear: an alarming claim that skips the detail or bends the source. Example: "AI will wipe out small businesses like yours."
Spot where forecasts disagree. In April 2023, Goldman Sachs economists forecast that generative AI could raise global GDP by about 7% over 10 years. In 2024, MIT economist Daron Acemoglu estimated a much smaller boost of roughly 1% over a similar period. He assumed only about 5% of tasks could be done profitably by AI in that time. Both are forecasts. The gap comes from different assumptions, not from one side measuring and the other guessing. When forecasts differ this much, say so. Do not pick the one you like.
Watch how a source gets bent. The same 2023 Goldman Sachs report said AI could expose the equivalent of 300 million full-time jobs to automation. "Exposed" means that some tasks in those jobs could be affected. It does not mean the jobs disappear. A headline such as "AI will take 300 million jobs" turns a forecast into fear. Compare the measured evidence: in the BCC survey (March 2026), 95% of AI-using SMEs said their workforce size had not changed over the past year.
Do: the next time Studio Nine's owner sees an advert saying "Salons using AI earn 40% more", she asks four questions: whose figure is it, when was it measured, which salons were counted, and how was it measured? If there are no answers, she files it as hype.
Task 3: List three tasks where AI might save time
Now look at your own business. Do not start from a tool. Start from the work.
- Write down the jobs that are repetitive, text-heavy, or that pile up at the end of the day.
- Choose three of them.
- For each one, note roughly how long it takes now each week. This is your baseline.
- Write "might help". Do not write "will save X hours". You have no evidence yet.
Worked examples:
- Bean & Bloom: writing weekly specials posts (about 1 hour), replying to catering enquiries (2 hours), drafting the staff rota notes (30 minutes).
- Okafor Electrical: writing up quotes after site visits (3 hours), chasing unpaid invoices (1 hour), answering "are you available?" messages (2 hours).
- Studio Nine: appointment reminder messages (1 hour), aftercare advice sheets (30 minutes), social media captions (1.5 hours).
The BCC research noted that, for many smaller firms, the hard part was not cost. It was picking a specific task worth automating. Your list is that first step. Later levels look at what AI can do reliably and which tools fit.
Checklist
- I can name the four checks: source, date, sample, method
- I have sorted one AI claim I have seen into evidence, forecast, hype or fear
- I have written down three tasks, each with a baseline time and no promised result
Key takeaways
- Check every statistic for its source, date, sample and method before you rely on it.
- Two honest surveys can give different numbers. In 2026 the ONS found 30% and the BCC found 54%, because they asked different firms in different ways.
- Forecasts are not evidence. Major forecasts disagree widely, for example about 7% (Goldman Sachs, 2023) against about 1% (Acemoglu, 2024).
- Promised results with no measurement behind them are hype. Bent or stripped-down sources are fear.
- Start from your own tasks and baseline times, not from a tool or a promise.