AI adoption is rising, but a 2025 OECD report finds a stark gap between large and small businesses. It also shows that many SMEs using generative AI are not yet applying it to core business activities. For small businesses, the next step is practical: choose a useful task, develop staff skills, and measure whether AI improves the work.
Small Businesses Are Falling Behind in AI Adoption
Artificial intelligence is becoming part of everyday business. From writing and research to data analysis and customer service, AI tools are increasingly accessible, affordable, and capable.
But access to powerful technology doesn't necessarily translate into meaningful business results.
TL;DR
Small businesses often have access to AI before they have reliable ways to use it. The best next step is to start with one low-risk, valuable task, train employees through hands-on practice, turn successful experiments into repeatable workflows, and measure whether AI actually improves speed, quality, or consistency.
A 2025 OECD report, AI Adoption by Small and Medium-sized Enterprises, prepared at the request of Canada's G7 Presidency, highlights a substantial divide between large businesses and their smaller counterparts.
The report presents findings drawn from different datasets:
- Across the OECD, approximately 40% of firms with at least 250 employees used AI in 2024 or the latest available year.
- For firms with 10–49 employees, the comparable figure was 11.9%.
- In a separate OECD survey, 29% of SMEs using generative AI reported applying it to their core business activities.
The 29% finding raises a practical question: how can SMEs make AI useful in the activities central to their business?
The OECD reports that generative AI use among surveyed SMEs remains concentrated more in peripheral than core activities. That finding describes where AI is used; it does not, by itself, measure employee fluency or the value of those uses.
An immediate opportunity is to move from isolated tasks to repeatable processes that employees can review, improve and use responsibly.
The OECD identifies four critical factors that enable successful AI adoption among small and medium-sized businesses: connectivity, access to AI-related resources, skills, and financing.
Workforce skills are one part of that picture, and an area managers can address through focused practice and training.
The report identifies skill shortages as a major barrier to adoption. It also finds that reported participation in AI-related training remains limited among SMEs using generative AI in the countries surveyed.
In a 2024 OECD survey covering Japan, Germany, the United Kingdom and Canada, fewer than 30% of SMEs using generative AI in each country reported employees participating in AI-related training.
In Canada, 29.4% of surveyed SMEs using generative AI reported that their employees participated in such training. This measures the share of businesses reporting training, not the share of individual workers trained.
The findings point to a practical training gap: AI tools may be available before employees have established reliable ways to use them.
And AI skills aren't limited to programming or technical expertise.
At HumanSavvy, our 7 Rules of AI Fluency describe transferable habits for working with AI. For most employees, the immediate opportunity is learning how to:
- Communicate more effectively with AI tools.
- Supply the context needed to produce useful results.
- Apply AI to existing tasks and business processes.
- Recognize inaccurate or unreliable outputs.
- Connect AI capabilities with the software and information they already use.
- Develop repeatable workflows rather than starting over with every interaction.
These are practical, learnable skills that can help employees make better use of technology already available to them.
Where Small Businesses Can Start
Smaller businesses do not necessarily have larger training budgets or spare capacity. But some can test a focused change quickly by involving a small team.
Where decisions can be made close to the work, owners and managers may be able to choose a trial task, assign responsibility and adjust the approach without a lengthy rollout.
The objective is to find a task that offers real value, not to introduce AI into every process.
Choose a task with available source information and a clear way to check the result.
For example, an organization could:
- Pick one recurring, low-risk task and describe the current process, including how long it takes and how quality is checked.
- Decide which step AI could assist with and which decisions must stay with a person.
- Practise with approved, non-sensitive examples from the team's actual work.
- Record the instructions, input requirements, review steps and person accountable for the result.
- Compare time spent, corrections needed and output quality with the previous process.
- Repeat the approach elsewhere only when the benefits justify the effort and risks.
A worked example: customer enquiries
Suppose a small service business receives recurring questions about pricing and appointments. A staff member could collect approved, anonymized enquiries and the company's current service information, then ask AI to group the questions and draft responses using only those materials. The employee would check each answer against the source information before anything is sent.
A starting instruction: "Using only the service information below, group these anonymized enquiries by topic, draft a clear answer for each theme, and flag any question the information cannot answer. Do not invent prices, policies or commitments."
To test whether this helps, the team could compare preparation time, corrections required and response quality with its previous process. This is an illustrative example, not a reported case study or measured result.
Not every task is suitable for AI, and not every use case will generate a meaningful return.
But a structured approach makes it easier to distinguish genuinely useful applications from interesting experiments.
Why Hands-On AI Training Matters
Reading about AI and watching demonstrations can help people understand its potential.
But confidence and competence develop through practice.
Employees need opportunities to work with their own documents, business challenges, workflows, and questions.
That's why HumanSavvy focuses on practical, hands-on AI training.
In our workshops, participants practise on business tasks and build reusable approaches to:
- Improve prompts and interactions with AI.
- Develop reusable instructions and workflows.
- Apply AI to research, writing and data analysis.
- Reduce friction between AI and existing business tools.
- Identify opportunities for straightforward automation.
- Establish appropriate verification and quality-control practices.
Participants practise framing a task, reviewing AI output and documenting a process they can reuse. We adapt examples to participants' roles and experience instead of assuming everyone needs the same training.
The objective isn't to turn every employee into an AI specialist.
It's to help people become more capable, confident, and productive in the work they're already doing.
Help Your Team Move Beyond AI Experimentation
HumanSavvy helps organizations build practical AI fluency through hands-on, instructor-led training.
Start with one task your team needs to do better. HumanSavvy's instructor-led training helps teams structure requests, check results and make effective approaches repeatable. Explore the courses to see how training could fit your organization.
Research source
OECD (2025). AI Adoption by Small and Medium-sized Enterprises: OECD Discussion Paper for the G7. Organisation for Economic Co-operation and Development, Paris.
Read the original OECD report (PDF)
Key references: Executive Summary (p. 6); adoption by firm size and business function (pp. 10–11); AI-related training (p. 35, Figure 12). The core-activities and employee-training figures come from survey evidence distinct from the OECD-wide firm-adoption comparison.
The HumanSavvy AI Fluency framework and recommendations presented in this article are independent interpretations and are not OECD methodologies or endorsements.