AI Strategy for Business

Is It Too Late for a Small Business to Start Using AI?

Is it too late to start using AI? The evidence says no. See what UK and US data show about small business adoption, and use the Ninety-Day Line to start with one job this week.

A gradient paper compass beside a dotted path that arrows toward a map pin, standing for a small business finding its way to a first AI job.

In brief

No, most small businesses are not too late to start using AI. UK government data from 29 October 2026 shows 32% adoption but only 14% calling AI highly important. Pick one job, one person and one decision date within 90 days, then keep, change or stop.

Almost certainly not. If you have not started with AI, you are behind some of your neighbors, but not by much, because most have not got far either. The UK government's 2026 survey found adoption doubled in a year, yet only 14% of businesses rated AI of high or critical importance. Waiting for the right moment is what costs you. Pick one job and give it 90 days.

How many small businesses have actually started with AI?

Roughly one in three UK businesses has, and most of them are barely past the front door. (If you are weighing where to begin, how a small business should start using AI is the companion piece.) The Department for Business, Innovation, Science and Trade (DBT) reported in its AI adoption survey, published 29 October 2026, that 32% of UK businesses were using at least one AI technology, up from 16% in 2025. Adoption among larger firms was 52%.

Look at what that use consists of. The same report says use was concentrated in accessible, low-complexity tools such as text generation, that most businesses were still at an early or exploratory stage, and that half of adopters (50%) reported a productivity increase, generally described as modest.

The Office for National Statistics (ONS) gives the small end of the picture. Its business insights bulletin of 8 October 2026 found 29% of businesses with fewer than 10 employees reported using AI technologies in September 2026, against 51% of those with 100 or more. So a 12-person firm that has done nothing is in the minority, though not an exotic one. If your team is under 10, AI for businesses with fewer than 10 people fits your size.

That is the first thing to see. The crowd you imagine ahead of you is thinner than it looks, and the front of it is mostly drafting emails.

Is "late" even the right question?

No. Calendar position tells you very little, because the gains in the research come from changing how work gets done, and that takes time whenever you start. Late is a measure of the date. What decides the outcome is how many weeks your people have spent doing real work with the thing.

The best evidence for this is from manufacturing, so treat it as a pattern rather than a promise for a bakery or a law firm. A US Census Bureau working paper (Center for Economic Studies paper 25-27, April 2025, by McElheran, Yang, Kroff and Brynjolfsson) used Census data for 2017 and 2021 and found J-curve returns: short-term performance losses came first, longer-term gains after. Firms that had adopted before 2017 grew faster over time, conditional on survival. The paper also found the losses fell hardest on older businesses, and that growth-oriented strategies softened them.

Two readings follow, and they point the same way. Starting earlier has a real advantage, because the awkward early stretch is behind you. And the awkward stretch exists for everyone who starts. Nobody skips it by waiting, and nobody who waits escapes it. They just take it later.

My own view, from training business teams: the owners who say "we're too late" are usually describing a feeling about other people's LinkedIn posts, not a measurement. When I ask what they have tried, the honest answer is generally "nothing structured".

Start now, wait, or buy big: which approach holds up?

Starting now with one small job holds up best on the evidence available today. I compared three approaches against the two sources above, checked on 9 November 2026.

Approach What it looks like What the evidence says
Start now, one job One named task, one owner, 90 days, a decision date Matches the pattern in DBT's findings: gains are real but modest and come from workflow gains in tasks such as drafting and admin, so a narrow job is where they show up. Starts the J-curve clock soonest.
Wait and watch Hold off until tools settle or a competitor moves Among non-adopters, 65% cited the lack of a clear business case (DBT). Watching does not produce one. Only trying does.
Buy big A platform or consultant-led program across the whole firm DBT found 67% of adopters were held back by limited skills. A large purchase assumes the skills already exist. The Census paper traced its short-run losses to costly adjustment inside core processes, which is what a firm-wide rollout touches first.

Two honest caveats. The DBT survey measures businesses of every size and asks what they did, not what would have worked, so it cannot prove that small starts beat big ones. And the Census paper is a working paper that has not had Census Bureau review, about industrial AI, so it supports the shape of the argument, not a number. I give you the table because it is the most defensible reading, not because the data settles it.

What does the Ninety-Day Line say about whether you are late?

The Ninety-Day Line is the test I use to answer "are we too late?": a business is not late if it can name one job, one person and one decision date inside the next 90 days. If you can name all three, you are not behind. You are about to start. This is my suggested way of thinking about it, not established research.

If you have started but see little return, why most small businesses are not getting value from AI covers that. Take a hypothetical 30-person electrical contractor. The office manager spends Friday afternoons turning engineers' site notes into customer completion summaries. That is the job. She is the person. The decision date is 8 February 2027, when she and the owner sit down with 10 summaries done the old way and 10 done with AI, and decide whether it saves real time and whether the quality holds. Nothing about that is a "late" or "early" question. It is a Friday task with a date on it.

The line gives you three ways to fail, which are worth knowing in advance:

  • No job. "Use AI more" is not a job. A job has a start and a finish you can see.
  • No person. If everyone owns it, nobody has done it by week three.
  • No date. Without a date to decide, the trial drifts into an experiment that never ends. That is the failure my Exit Question exists to prevent in larger pilots.

If you cannot name the job, you are not late. You are unsure where to start, and the next section is for you.

What I tell business owners who feel behind

Choose the job that is dull, frequent and checkable. Across the business teams I train, the first wins are almost never glamorous: summarizing meeting notes, turning rough bullet points into a customer email, tidying a quote before it goes out. Those jobs share three traits. They happen weekly, someone already knows what a good result looks like, and a mistake is caught before it reaches a customer.

Then read the research against your own staff. The DBT survey found limited skills affected 67% of businesses that had adopted AI. In my experience that is the real gap. The tool is easy to reach and slow to use well, so a head start is mostly practice, not purchases, and practice has to be given to people as working time. The people who end up using AI well are, in the teams I see, the ones who were given time to use it badly first.

One more thing. Your staff may already be ahead of you. If a few of them are using AI on their own phones, you are not at the start line at all. You are behind on governance, not adoption, and shadow AI is the better place to begin.

What should you do on Monday?

Pick the job, the person and the date this week. It should take a morning, not a project.

  1. Monday: list five tasks that happen every week and take someone more than 30 minutes. Cross out anything involving sensitive client data you have no policy for yet.
  2. Tuesday: choose one. Name the person. Write down what "good" looks like in two sentences, so you can tell the difference later.
  3. Wednesday: put the decision date in two diaries, about 90 days out, with the question "Keep, change or stop?"
  4. By Friday: the person does the job once with AI and once the usual way, and writes down the time and what needed fixing.
  5. At the date: decide. Keep it, change it, or stop and try a different job. Any of the three is a result.

If you want to know where the whole business stands first, the free Business AI Readiness Check will show you whether the gap is skills, permission or something else before you spend a month guessing.

Where to go next

If your team is using AI but nobody has shown them how to use it well, that is exactly what my AI training for teams is built for. If you would rather know where you stand first, take the free Business AI Readiness Check. And if you are planning an event and want your members or leadership team to hear this argument in person, I give keynotes to business audiences.

Sources and further reading

Dan Fitzpatrick is The AI Educator, a Forbes contributor, keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries.

Key takeaways

  • The UK government's AI adoption survey (published 29 October 2026) found 32% of businesses using at least one AI technology, up from 16% in 2025.
  • Only 14% of UK businesses rated AI of high or critical importance, and half of adopters reported productivity gains described as modest.
  • The ONS found 29% of UK businesses with fewer than 10 employees used AI in September 2026, so a firm that has not started is in the minority but not an outlier.
  • A US Census Bureau working paper found J-curve returns in manufacturing: short-term losses first, longer-term gains after. Waiting postpones the awkward stretch, it does not remove it.
  • The Ninety-Day Line says a business is not late if it can name one job, one person and one decision date inside the next 90 days.
  • Start with a dull, frequent, checkable task, run it with AI for 90 days, then decide to keep, change or stop.

Frequently Asked Questions

Is it too late for my small business to start using AI?

No. The UK government's 2026 survey found 32% of businesses used at least one AI technology, yet only 14% rated AI of high or critical importance. Most firms are still early. What matters is putting one job through AI for 90 days, with a named person and a decision date.

What percentage of small businesses use AI?

The ONS business insights bulletin of 8 October 2026 found 29% of UK businesses with fewer than 10 employees used AI technologies in September 2026, against 51% of those with 100 or more. The Department for Business, Innovation, Science and Trade put all UK businesses at 32%. The surveys differ in method, so compare each with itself.

Do you lose out by waiting to adopt AI?

Possibly, but waiting does not avoid the awkward early stretch. A US Census Bureau working paper on manufacturing found short-term losses followed by longer-term gains, and firms that adopted earlier grew faster, conditional on survival. That evidence is from manufacturing, so treat it as a pattern, not a promise for your sector.

What is the Ninety-Day Line?

The Ninety-Day Line is Dan Fitzpatrick's test for whether a business is too late to start with AI. If you can name one job, one person and one decision date inside the next 90 days, you are not late. You are about to start. It is a suggested way of thinking, not established research.

Where should a small business start with AI?

Start with a dull, frequent, checkable task, such as summarizing meeting notes or tidying a customer email. Name one person, write two sentences describing a good result, and set a decision date about 90 days out. Leave out tasks involving sensitive client data until you have a written policy.

Should a small business wait for AI to settle down?

Waiting does not supply what most non-adopters say they lack. In the 2026 UK government survey, 65% of non-adopters cited the lack of a clear business case, and you build one by trying a job, not by watching others. A 90-day trial on one task is easy to stop if it does not earn its place.

Dan Fitzpatrick helps businesses use AI well through keynotes, practical AI training for teams, and AI strategy and governance for leadership teams.

Dan Fitzpatrick

Written by

Dan Fitzpatrick

Dan Fitzpatrick is The AI Educator: a Forbes contributor, international keynote speaker and author who has trained more than 150,000 people across 30+ countries. He works with businesses and leadership teams on AI training, AI strategy and governance, and keynotes.

He also works with schools as an AI keynote speaker for schools, through school AI strategy consulting and with AI training for schools.