Everyone Uses AI, Few Use It Deeply: Moving a Small Business Past Shallow Adoption
Ask ten small business owners whether their company uses AI and about seven will now say yes. Ask where, and the answer is mostly emails, copy edits and looking things up. When Ireland's Small Firms Association (SFA) published its survey in August 2026, it used the word "shallow" to describe the situation. The question for small business AI adoption is no longer whether, but how deep. This article covers what three recent surveys show, the common signs of shallow use, and three steps for turning AI from a personal helper into part of how the company works.
Small business AI adoption today: widely used, rarely deep
Three surveys point the same way.
- SFA (Ireland, 404 small businesses): 92% use generic off-the-shelf tools such as ChatGPT and Microsoft Copilot. Only 30% use AI models customised to their own needs, up from 22% a year earlier. Among firms taking a customised, strategic approach, 65% report efficiency gains.
- Thryv (United States, 561 owners and decision makers): adoption rose from 55% a year ago to 66%, yet 70% of owners say they need more training to use AI effectively.
- Intuit QuickBooks (US, Canada, UK, Australia): roughly seven in ten businesses use AI regularly, but in the US only 12% paid for dedicated AI tools between 2021 and 2025.
Taken together: adoption is high, most of it sits in free or generic chat tools, and few companies have connected AI to their operations.
The SFA also asked what holds firms back: 31% cite a lack of technical expertise, 28% a lack of time to investigate, and 17% a lack of finance. Intuit's report adds that owners worry most about privacy and security, the risk of errors, and uncertainty about what AI can do. Cost is not the leading concern.
What shallow use looks like: four signs
If several of these describe your company, you are probably still in the shallow stage:
- Everyone on their own: each person has their own account and their own way of asking, and good techniques never reach colleagues.
- Copy and paste: AI output is moved by hand into emails, spreadsheets or systems. Nothing connects AI to the company's tools.
- AI does not know the company: products, prices and tone have to be explained again every time, because the AI holds none of your information.
- No measurable result: everyone feels it "helps", but nobody can say which process got faster, or by how much.
Shallow use is not a failure. It is the necessary first stage. The problem is staying there too long: individuals save a few scattered minutes, and the way the company runs does not change.
Three steps from shallow to deep
1. Pick just one process
Do not begin with "rolling out AI across the company". Choose one process that repeats weekly, follows fixed steps and is cheap to get wrong, such as answering common enquiries, sorting orders or drafting quotes. Write down every step, then decide which ones to hand to AI.
2. Give AI your company's information
Vague answers usually mean the AI does not know your products, prices and rules. Turning your common questions, product details and brand voice into documents AI can read is the best-value step. We covered how in the model is smart but does not know your company and can AI read your brand guidelines?.
3. Make the know-how the company's, not one person's
Seven in ten owners say they need more training, but training does not have to mean a course. It works better to keep good prompts and examples in a shared place, spend fifteen minutes a week having a colleague demonstrate one use, and name someone to maintain it. Knowledge should stay with the company, not in one person's chat history.
If generic tools still cannot connect to your systems at this point, that is the time to consider something custom. The SFA survey shows the share of small firms building standalone in-house software has risen from 10% to 20%, so this route is becoming practical.
AI does the work, people choose it and check it
The deeper the adoption, the more people matter. Their position simply changes.
- AI handles: repetitive work that follows a set process, first drafts, organising data, and responding around the clock.
- People handle: deciding which process is worth changing, setting what information AI may see, checking anything sent outside the company, dealing with exceptions and complaints, and judging whether the investment pays off.
Intuit's report notes that AI use is highest in marketing, administration and customer service, and lowest in roles that need significant human judgement. That pattern is a sensible division of labour in itself: hand over the work with clear rules first, and keep judgement with people.
Conclusion: depth is organised, not bought
The gap between shallow and deep rarely comes down to how expensive the tool is. It comes down to whether the company has put its own processes, information and methods in order. No AI can do that organising for you, and it decides how much AI can help. This is what TitApeX means by "AI handles perfection, people handle what moves us": let AI make the repetitive work reliable, and keep people's time for customers and judgement.
If you would like to find the one process in your company best suited to start with, take a look at our services or get in touch.
Comments