Small Business Owners and AI: The Uneven Adoption Curve and What It Means for the Economy

The Expectation Gap

There is a widening gulf between what technology vendors promise small business owners and what those owners are actually equipped to do. In the spring of 2023, a small dry-cleaning business in Milwaukee received a form email from its point-of-sale software provider announcing that its platform was now "AI-powered." The email described new features: automated demand forecasting, dynamic pricing suggestions, and a natural language interface for pulling reports. The owner, who had run the business for 22 years and had no staff dedicated to technology, deleted the email. She did not understand what dynamic pricing meant in practice. She was not sure whether demand forecasting would work for a business that serves a two-mile radius. She had no time to find out.

This story, unremarkable in its individual details, represents a pattern playing out across tens of millions of small and medium-sized enterprises worldwide. The technology sector has moved with extraordinary speed. The commercial availability of capable generative AI tools has compressed what might have been a decade-long adoption cycle into two or three years. Vendors, investors, and policy commentators have greeted this acceleration with enthusiasm. The small business owner trying to keep the lights on has largely been left to navigate it alone.

The gap between expectation and reality is not simply a matter of technical complexity. It reflects deep structural inequalities in access, literacy, time, and capital that have defined the small business technology experience for decades. Artificial intelligence amplifies those inequalities. As McKinsey Global Institute (2023) documented, generative AI has the potential to add trillions of dollars to global economic output, but the distribution of that value is far from even. The businesses best positioned to capture productivity gains are those that already possess the organizational infrastructure, technical staff, and capital to deploy and iterate on AI tools. For the smallest businesses, the asymmetry is stark.

The question this paper examines is not whether AI will transform the small business landscape. It already has. The question is who benefits, who is left behind, and whether the policy and market responses to date are remotely adequate to the challenge.

AI Adoption Rates Among Small Businesses

Measuring AI adoption is complicated by the ambiguity of the term itself. When a small retailer enables an AI-driven product recommendation engine within its Shopify store, is that AI adoption? When a sole trader uses ChatGPT to draft a response to a customer complaint, does that count? Survey data reflects this definitional messiness, and results vary substantially depending on how questions are framed.

Nonetheless, the broad contours are consistent across studies. Large enterprises adopt AI at substantially higher rates than small businesses. The gap in adoption is not merely a matter of scale but of kind: large businesses are integrating AI into core operational systems such as supply chain management, customer relationship management, fraud detection, and recruitment screening. Small businesses, to the extent they adopt AI at all, tend to use it for peripheral tasks: generating marketing copy, answering frequently asked questions, or scheduling social media posts.

The U.S. Small Business Administration (2022) counted approximately 33.2 million small businesses in the United States, employing nearly half the country's private-sector workforce. These businesses operate across an enormous range of industries and contexts. A five-person legal practice, a 12-employee restaurant, a one-person consultancy, and a regional manufacturer with 180 workers are all classified as small businesses. Their technology needs, resources, and readiness are vastly different. AI adoption statistics that lump all these together conceal more than they reveal.

Research from Deloitte Insights (2020) identified a typology of AI adoption stages: experimentation, expansion, and transformation. Most large enterprises surveyed had moved into the expansion phase. Most small businesses, in those few studies that disaggregate by size, remain in early-stage experimentation or have not begun at all. The implications of this lag compound over time. Businesses that are still learning to use tools their competitors integrated years ago face a structural disadvantage that grows wider as the technology continues to advance.

Goldfarb and Tucker (2019) offered a theoretical framework that is useful here. They argued that digital technologies reduce the cost of certain economic activities, but do so unevenly depending on the existing capabilities of the adopter. AI accelerates this dynamic. The marginal cost of deploying an AI tool is lower than ever before. The marginal cost of deploying it effectively, however, depends heavily on the organizational capabilities, data infrastructure, and human expertise available to the deploying firm. Those inputs are distributed in direct proportion to firm size and existing capital.

What the adoption statistics do not capture is the second-order effect: competitive dynamics. A small business that has not adopted AI marketing tools is not simply behind in one dimension. It is competing against businesses that can generate more content at lower cost, personalize customer communications at scale, and analyze their own performance data continuously. The competitive gap created by uneven AI adoption is not linear. It compounds.

The Skills and Literacy Deficit

When surveys ask small business owners why they have not adopted AI, cost is rarely the primary obstacle. Across multiple studies, the most consistently cited barrier is a lack of knowledge and skills. Business owners do not know which tools are relevant to their situation. They do not know how to evaluate competing claims from vendors. They do not know how to assess whether an AI tool's output is accurate, appropriate, or safe. And they do not have staff who can answer these questions for them.

This is a literacy problem, and it is more complex than it might appear. AI literacy is not a single skill. It encompasses at least four distinct competencies: an understanding of what AI tools can and cannot do; the ability to evaluate the quality of AI-generated output; the capacity to integrate AI tools into existing workflows; and awareness of the risks, biases, and failure modes that accompany AI systems. Most small business owners possess none of these competencies, not because they are incapable of learning them but because no one has provided accessible, relevant training.

Agrawal, Gans, and Goldfarb (2018) described AI as fundamentally a prediction technology: a set of tools for making better guesses based on data. This framing is helpful because it makes visible what AI requires to function well. Good predictions require good data. Good data requires consistent collection practices, careful storage, and some understanding of what you are trying to predict. Most small businesses do not have systematic data collection practices. Their customer records may be scattered across email, a spreadsheet, and a point-of-sale system that has never been connected to anything else. Before an AI tool can help them, they need to solve problems that are not technological at all.

The literacy gap is also generational in complex ways. Younger owners are sometimes assumed to be more digitally capable, but digital fluency with consumer products does not translate automatically to operational competence with AI business tools. A 28-year-old restaurateur who is comfortable on TikTok is not necessarily equipped to evaluate an AI-powered inventory management system. The skills are different, and conflating them leads to policy and vendor support that misses the actual need.

Brynjolfsson, Li, and Raymond (2023) found in a controlled study of customer service workers that access to a generative AI tool substantially boosted the productivity of less-experienced workers while having a smaller effect on the most experienced. This finding has been interpreted optimistically, as evidence that AI can democratize expertise. But it depends on workers having access to training, time to experiment, and an organizational context that supports the integration of new tools. Solo operators and very small teams rarely have those conditions. The democratizing potential of AI is real but contingent on infrastructure that is not evenly distributed.

The skills deficit also interacts dangerously with the confidence with which AI systems present their outputs. Language models generate text fluently and authoritatively regardless of whether their content is accurate. Small business owners who do not know how to evaluate AI-generated output may publish incorrect information on their websites, use legally problematic contract language drafted by an AI, or rely on financial projections that are statistically nonsensical. The risks of under-competent AI use are not hypothetical. They are accumulating in real businesses right now, largely invisible in the aggregate statistics on adoption and productivity.

Vendor Lock-In and the Platform Dependency Problem

The structure of the AI tools market creates specific risks for small businesses that receive relatively little attention in the mainstream discussion of AI adoption. Most small businesses that do adopt AI do so through platforms they already use for other purposes: their e-commerce provider, their accounting software, their email marketing service. These platforms have integrated AI features at the application layer, meaning the business owner does not need to understand the underlying technology. They simply enable a feature.

This model of adoption is accessible and removes many of the barriers associated with deploying AI independently. But it creates a new vulnerability: platform dependency. When a small business's AI capabilities are entirely mediated by a single platform, that business has no ability to switch providers without losing its AI functionality. It cannot negotiate on price. It cannot evaluate competing AI providers. It is subject to the terms of service, data practices, and pricing decisions of a platform it cannot easily leave.

PwC's Global AI Study (2023) noted that competitive advantage in AI increasingly comes from proprietary data and proprietary model customization. Large enterprises can build and train models on their own data. Small businesses cannot. They are customers of AI infrastructure, not owners of it. This means they benefit from whatever commodity AI features their platform chooses to offer, at whatever price the platform sets, under whatever data-sharing arrangements the platform's terms of service specify.

The data-sharing dimension is particularly underexamined. When a small business enables an AI feature in a third-party platform, its customer data is typically used to generate AI outputs. Depending on the platform's terms of service, that data may also be used to train or improve the underlying model. Small business owners who have not read these terms carefully may not realize that their customer data, including purchase histories, contact information, and behavioral patterns, is contributing to a model that their platform sells to other businesses, including competitors.

Accenture (2022) identified platform dependency as one of the defining technology risks for SMEs in the early part of this decade. The risk is not that platforms are necessarily malicious, but that the power asymmetry between a large platform and a small business customer is so extreme that the small business has no practical means of redress when the relationship goes wrong. AI intensifies this asymmetry because the value created by AI tools depends on data that accumulates over time. Switching costs become prohibitive not because of contractual lock-in but because the business's AI capabilities are built on data and model customization that cannot easily be moved.

Customer Expectations and Competitive Pressure

The pressures on small businesses to adopt AI do not come solely from vendors. They come from customers, and increasingly from the competitive environment created by larger businesses whose AI deployments have reset expectations about service quality, responsiveness, and personalization.

Customer expectations for response times have compressed dramatically over the past decade. A customer who sends an inquiry to a small business expects a response within hours, sometimes minutes. Large businesses with AI-powered customer service systems respond instantly, around the clock, in multiple languages. The small business owner who answers customer emails personally, during business hours, is not competing on a level playing field. This is not simply a technology gap. It is an expectation gap created by the technology decisions of large enterprises and felt acutely by small businesses that cannot match them.

Personalization is another dimension where AI has reset competitive expectations. E-commerce giants use AI to present each customer with a personalized selection of products based on their behavior, purchase history, and demographic profile. Small online retailers, operating on platforms that offer simpler recommendation engines, cannot replicate this. Customers who have been trained by their interactions with large platforms to expect personalized experiences may find the more generic experience offered by smaller businesses actively disappointing, even if the products themselves are superior.

Bessen (2019) argued that AI-related productivity gains tend to increase demand in ways that expand employment even as they automate specific tasks. This is the optimistic view of AI's labor market effects. But this demand expansion is not uniformly distributed. The additional demand generated by AI-powered efficiency may accrue primarily to businesses that have already adopted AI, creating a positive feedback loop that further concentrates market share. The small business that has not adopted AI does not simply fail to gain from the productivity benefits. It may lose customers to competitors whose AI-enhanced experience attracts and retains buyers more effectively.

The competitive pressure from AI adoption is also felt in pricing. AI-driven dynamic pricing, automated procurement optimization, and demand forecasting give larger businesses tools to optimize margins in ways that smaller businesses cannot easily replicate. A small retailer competing with a chain that uses AI to optimize its purchasing and pricing in real time faces structural disadvantage not just in efficiency but in the prices it can sustainably offer.

Case Studies: Businesses That Adapted and Those That Did Not

The empirical evidence on how small businesses are actually navigating AI adoption is sparse, because most rigorous research on AI in business focuses on large enterprises. What exists is a collection of case studies, survey data, and anecdotal evidence from practitioners that begins to reveal a pattern.

Among businesses that have successfully integrated AI, certain common characteristics emerge. They tend to have an owner or manager who takes personal responsibility for understanding technology. They have a relatively clear sense of which specific operational problems they are trying to solve with AI. They start with a narrow use case and expand only after gaining confidence. And they are embedded in networks, whether industry associations, peer groups, or supplier relationships, that provide access to practical knowledge about what works.

A bakery chain with six locations in the Pacific Northwest provides an illustrative example. The owner, who had a background in operations management, identified labor scheduling as her primary pain point. She evaluated three AI scheduling tools, selected one with a low monthly cost and a trial period, and integrated it with her existing payroll system over a period of four months. The tool reduced her scheduling time by roughly six hours per week and improved her ability to predict staffing needs during peak periods. She has not used AI for anything else. She was explicit in interviews that she did not want to expand her use of AI tools until she fully understood the one she had.

This careful, narrow approach is arguably the right one for a small business. But it requires a level of organizational discipline, personal technical interest, and available time that is not widely distributed. The business owner whose primary constraint is simply getting through each day, who has no staff and no time for extended evaluation of new tools, who operates in an industry without a strong peer network, and who has never had reason to develop interest in technology, is not in a position to replicate this pattern.

Among businesses that have struggled, the most common failure mode is not catastrophic but gradual. They continue operating as they always have while their competitive environment shifts around them. Their customer acquisition costs rise because their marketing is less targeted and efficient than competitors who have adopted AI-driven advertising tools. Their response times fall below the implicit expectations of a customer base trained by interactions with larger businesses. Their pricing becomes less competitive because they lack the optimization tools that competitors use. None of these effects is immediately fatal. Together, over time, they accumulate into a structural disadvantage that becomes increasingly difficult to reverse.

A second failure mode involves adoption without adequate understanding. Small businesses that enable AI features without understanding how they work, or without the ability to evaluate the quality of their outputs, can make decisions based on AI recommendations that are wrong. An AI-powered inventory management system recommended overstocking a product line based on seasonal data that turned out to be anomalous. The business owner, trusting the system's confident output, followed the recommendation and ended up with excess inventory that took months to clear. The loss was not catastrophic, but it eroded trust in the system and left the owner unwilling to rely on AI tools for anything consequential.

Policy Responses and Support Gaps

Government responses to the challenge of AI adoption among small businesses have been, by most assessments, inadequate to the scale of the problem. Support programs have been fragmented, often focused on awareness rather than practical capability building, and poorly targeted to the businesses most at risk of being left behind.

The OECD's SME and Entrepreneurship Outlook (2021) identified the absence of integrated digital and AI support programs as a significant policy gap. While most OECD countries have some form of digital adoption support for small businesses, these programs were designed for an earlier generation of technology. The transition from digitization, moving business records and transactions online, to AI integration is qualitatively more complex, and existing support structures are not calibrated for it.

In the United States, the Small Business Administration (2022) has acknowledged AI as an emerging priority but has not yet developed comprehensive support programming. Existing SBA resources on AI are primarily informational. They describe what AI is and suggest that small businesses consider adopting it. They do not provide the hands-on, industry-specific, workflow-integrated training that small business owners actually need to make meaningful use of these tools.

There is also a fundamental misalignment between how policy is designed and how small businesses actually operate. Policy responses to technology adoption challenges typically assume that the primary barrier is financial, or that awareness programs will trigger uptake if they can reach enough businesses. The evidence suggests that the primary barriers are time, skills, and context, all of which are structural problems that resist simple policy solutions. A webinar about AI will not help a business owner who has no time to attend it. A subsidy for AI software will not help an owner who does not know how to use it.

Deloitte Insights (2020) argued that effective AI adoption support requires what they call "trusted intermediaries," organizations with credibility in specific industries that can provide hands-on support, curate tool recommendations, and help businesses integrate AI into their specific workflows. Trade associations, chambers of commerce, and industry-specific small business development centers could serve this function. In most cases, they do not have the resources or expertise to do so.

There is a growing body of work suggesting that the policy response needs to address not just adoption barriers but also the risks of adoption without adequate support. Small businesses that adopt AI without understanding it are vulnerable to data privacy violations, vendor manipulation, and operational failures that can be harder to recover from than simply not adopting at all. A framework that treats AI adoption as uniformly desirable, regardless of context or capacity, misses this dimension entirely.

PwC (2023) and McKinsey (2023) both pointed to the likelihood of significant productivity concentration in the near term: the businesses best positioned to capture AI's economic benefits are already those with the largest resources. Without deliberate policy intervention to close the adoption gap, the economic effects of generative AI may accelerate existing trends toward market concentration and worsen the competitive position of small businesses relative to large ones. The macroeconomic implications of this dynamic, including its effects on employment, regional economic vitality, and income distribution, deserve far more attention than they have received.

There is no single policy solution to the problem described in this paper. The barriers are multiple, intersecting, and vary substantially by industry, firm size, geography, and owner characteristics. What is clear is that awareness campaigns, technology subsidies, and voluntary vendor commitments to accessibility are insufficient. The gap between what AI offers and what small business owners can practically use is growing. Without more serious attention to the structural conditions that determine adoption capacity, the promise of AI as an equalizing technology will remain largely fictional.