The best and the rest: what artificial intelligence could mean for small business

Published 24 June 20265 min read

Written by

Fahad IshtiaqFounder, Vela

A small business can buy access to a capable AI system without building a research department. What it cannot buy quite so easily is the time, knowledge and organisational capacity to make that system useful.

That distinction matters well beyond individual firms. In Canada, businesses with fewer than 100 employees account for 98.2% of employer businesses and 46.6% of private-sector employment (Innovation, Science and Economic Development Canada [ISED], 2025). In the United States, whose small-business definition is broader, small firms employ 45.9% of private-sector workers (U.S. Small Business Administration [SBA], 2026).

Improving their performance would reach a substantial part of both economies. It is particularly relevant to Canada’s longstanding productivity problem, which Bank of Canada Senior Deputy Governor Carolyn Rogers described as an emergency in 2024 (Rogers, 2024).

AI gives a small team access to work that previously demanded more specialist support. It also strengthens competitors who know how to use it. The decisive question is what turns access into an advantage.

Capabilities a small team can put to work

In an owner-run firm, several responsibilities may compete for the same person’s attention: answering customers, preparing proposals, maintaining records and delivering the service itself. Assistance with a defined task can release capacity without requiring an additional full-time role.

Workplace experiments establish that useful assistance is possible. Professionals given access to ChatGPT spent 40% less time on assigned writing tasks and produced work assessed as 18% higher in quality (Noy & Zhang, 2023). Among 5,172 customer-support agents, an AI assistant increased issues resolved per hour by about 15% overall, with larger gains among less-skilled and less-experienced staff (Brynjolfsson et al., 2025).

These experiments covered professional writing and a large support operation. For a small firm performing similar tasks, they offer a clear reason to test assistance against its own work.

Owners’ reports provide a complementary perspective. In the Federal Reserve Banks’ 2025 Small Business Credit Survey, 46% of responding small employer firms reported AI use by the business or its employees. Among users, 71% reported increased productivity and 31% higher sales (Federal Reserve Banks, 2026). The figures are self-reported and come from a convenience sample.

A practical starting point is therefore a recurring problem with a visible cost. Consider a hypothetical service firm whose proposals wait because the owner must assemble the same information each time. AI could prepare a draft from approved service details, leaving the owner to check scope and pricing. The relevant test is whether accurate proposals reach customers sooner with less total effort, including review and correction.

The gap that access could help close

The productivity divide predates generative AI. OECD research documents increasing divergence between leading firms and firms further from the productivity frontier, pointing to difficulties in the spread of technology and know-how (Andrews et al., 2016).

AI makes writing support, analysis and routine information handling available through widely used software. A smaller firm can draw on those capabilities without building a specialist department. Its next challenge is to apply them to the work that holds it back.

This is the central economic opportunity: more firms becoming better at the work they already need to do.

Why use and integration are different

Statistics Canada’s spring 2026 survey found that 19.2% of employer businesses had used AI to produce goods or deliver services during the previous year. Among AI users, 68.1% of businesses with at least 100 employees reported training existing staff, compared with 24.0% of businesses with one to four employees (Do et al., 2026).

The training difference is revealing. Access to a tool can be quick; learning where it helps, adapting a process and evaluating results require further work.

Other surveys expose a similar distinction between trying a tool and integrating it. Only 7% of AI users in the Federal Reserve small-business survey reported full integration (Federal Reserve Banks, 2026). Bank of Canada research likewise distinguishes widespread interest and use from significant deployment in core operations (Chawla & Arnburg, 2026). The US Census Bureau found use rising among firms with at least 20 employees while remaining relatively stable among smaller businesses (Grundy et al., 2026).

Survey definitions and populations differ, so the adoption rates are not directly comparable. The OECD identifies persistent gaps between smaller and larger firms, alongside the importance of connectivity, data, skills and finance (OECD, 2025).

For an owner, integration means deciding where information comes from, who checks the output, how the result enters the next stage of work and what happens when the system cannot complete the task. A subscription settles none of those questions.

Advice still requires judgement

A field experiment with Kenyan entrepreneurs provides a useful challenge to simple optimism. Researchers randomised access to a GPT-4 business assistant and found no statistically significant average effect on revenues or profits. Subgroup analyses suggested that initially stronger performers may have benefited by more than 15%, while weaker performers did nearly 10% worse (Otis et al., 2026).

The authors link this divergence to which advice entrepreneurs chose to implement. The same assistance produced sharply different outcomes. Choosing what to act on was part of the work.

That makes training more than a lesson in writing prompts. Owners need to evaluate an answer against their customers, costs and operating circumstances. A plausible recommendation may still be a poor fit for the business receiving it.

Reliability and control also matter. Accuracy was the leading challenge reported by AI users in the Federal Reserve survey, while Statistics Canada listed cybersecurity and privacy concerns among the reported barriers (Do et al., 2026; Federal Reserve Banks, 2026). The UK’s Competition and Markets Authority has also identified risks around choice and switching in foundation-model markets (Competition and Markets Authority, 2023).

Before relying on a system, a small firm should understand what information it will use, how consequential errors will be caught, and whether essential records can be recovered or moved if its supplier changes.

Turning access into an advantage

The scale of possible gains remains uncertain. Acemoglu’s task-based analysis gives an initial upper bound of 0.66% for cumulative total factor productivity gains over ten years, with a lower estimate after accounting for harder tasks (Acemoglu, 2025). Danish research found no detectable average effects on earnings or recorded hours two years after ChatGPT’s launch despite changes in work (Humlum & Vestergaard, 2026).

For an owner, the immediate question is simpler: does this application improve this business? Choose a recurring task, record the current effort and error rate, trial assistance on representative work, and compare the whole process. Include the owner’s time spent checking and fixing results. Decide in advance what improvement would justify continued use.

At an economic level, the OECD points toward differentiated support for firms with different levels of digital maturity (OECD, 2025). Practical training, implementation support and usable data can help more businesses turn access into capability.

AI gives smaller firms a way to challenge the gap between the best and the rest. The next investment is in the skills, information and processes that turn access into useful work. A subscription is easy to buy. A stronger business is what makes it worth buying.

Sources

Every figure in this article was checked against these sources.

All posts