73% of AI-Using Small Businesses Report an Impact on Employee Roles — What a U.S. Survey of 750 Owners and Operators Shows

In a survey commissioned by the U.S. Chamber of Commerce Foundation and conducted by Ipsos, 54% of owners and operators of small businesses that use AI said AI had a “mostly positive” impact on how quickly their employees complete work.

For employees’ ability to take on more challenging work, 50% reported a mostly positive impact; for the quality of their output, 47% did.

“Mostly negative” responses came in at 3–4% for each of these measures.

That said, these figures do not come from direct measurement of processing time or output quality. They are employers’ self-reported perceptions.

The remaining responses were “mixed impact” and “no impact at all,” so it would be a misreading to treat everything outside that 3–4% as favorable.

The same survey found that, among businesses using AI, 73% reported some impact on employees’ roles and responsibilities, 73% on customer expectations and behavior, 70% on expectations for employees’ job performance, and 65% on hiring and personnel decisions.

These numbers do not directly indicate that formal reorganizations or changes to performance review systems took place.

Even so, the fact that employers perceive AI’s effects as extending beyond task speed and output quality into roles and day-to-day operations is worth noting when planning an implementation.

This article separates what the survey does show from what it does not, and then works through the organizational changes to monitor from the outset.

One note on terminology: “small business” in this survey means a U.S. business with 500 or fewer employees, excluding sole proprietorships. That does not align with the statutory definition of a small or medium-sized enterprise in Japan, so these percentages should not be applied directly to Japanese companies.

What the survey measured, and what it did not

The survey was commissioned by the U.S. Chamber of Commerce Foundation and conducted by Ipsos among 750 owners and operators of small businesses with 500 or fewer employees, excluding sole proprietorships, between May 19 and June 4, 2026 (published July 14, 2026).

The credibility interval for all respondents is reported as plus or minus 4.4 percentage points.

The key point to establish upfront is that this survey measures perceptions — whether owners and operators perceive an impact — rather than actual productivity or formal changes to organizational structures or policies.

It is not a survey of employees or customers themselves, nor is it an objective measurement of productivity or quality.

It also does not establish that AI caused any of the changes reported.

Employers report more positive than negative effects on speed and quality

Among businesses using AI, 54% reported a “mostly positive” impact on how quickly work is completed, 50% on employees’ ability to take on more challenging work, and 47% on the quality of their output.

“Mostly negative” responses were limited to 3–4% across those measures.

Because the remaining responses were “mixed impact” and “no impact at all,” it would be wrong to read everything outside that 3–4% as favorable.

Still, in employers’ own assessment, positive reports clearly outnumbered negative ones.

How employers say AI-enabled gains are being used

The same survey also asked how employers believe employees are using the time or capacity freed up by AI — but it asked this only of respondents who reported a “mostly positive” impact on task-completion time and/or work-output quality.

That narrower base, rather than all AI-using respondents, is important to keep in mind.

Among those respondents, 65% said they saw employees using the gain to deliver more or higher-quality work than before. Fifty-six percent cited learning, planning, or reviewing existing work; 38% cited stretch assignments or new responsibilities; 36% cited avoiding overtime or extra hours; and 30% cited breaks or personal tasks (multiple responses were allowed, so the total exceeds 100%).

Employers therefore reported both business-directed uses, such as increased output and more challenging assignments, and employee-directed uses, such as reduced overtime and time for breaks.

A separate study offers a useful reminder that AI’s effects can vary substantially across workers.

In a study of customer support work, access to a generative AI assistance tool was associated with a 15% average increase in issues resolved per hour. Less experienced and lower-skilled agents saw larger gains, including an increase of roughly 30%.

The most experienced and highest-skilled groups saw much smaller gains, and the highest-skilled agents showed small declines in some quality measures (Brynjolfsson, Li, Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140(2), 2025).

The main analysis covers 5,172 workers and uses a quasi-experimental difference-in-differences design based on a staggered rollout. It was not a study in which all participants were randomly assigned.

The paper also reports an earlier pilot involving approximately 50 workers, about half randomized to treatment, although the authors lacked information on the pilot’s control-group workers.

The study examines a single company’s customer-support operation, including its subcontractors. It does not directly explain the U.S. Chamber of Commerce Foundation’s survey results or establish which KPIs should be used for different experience levels.

It does, however, provide a reason to measure outcomes by experience level and job type rather than assuming that AI will affect every employee in the same way.

Beyond task speed: roles, expectations, and personnel decisions

The other set of findings concerns roles, expectations, customer interactions, and personnel decisions.

Among businesses using AI, 73% reported that AI had some impact on employees’ roles and responsibilities.

Another 73% reported an impact on customer expectations and behavior, 70% on expectations for employees’ job performance, and 65% on hiring and personnel decisions.

In each case, these figures combine “major” and “minor” impact as reported by employers themselves. They do not indicate the direction of the impact, nor whether any formal internal policy was actually changed.

The finding on customer expectations also warrants care: it reflects how business owners perceive their customers, not a survey of customers directly.

Even with those caveats, evaluating AI’s impact solely through task speed and output quality risks missing changes underway in how work is divided, what is expected of employees, how customers are handled, and how personnel decisions are made.

Model and system performance remain important to the outcome of any implementation.

What this survey suggests, however, is that technical evaluation alone will not capture all of the organizational effects that may accompany adoption.

When the scope of work handled by AI changes, organizations may also need to revisit human roles, expectations for employee output, customer interactions, and personnel decisions.

Roles, expectations of employees, and hiring decisions are not independent of one another.

When AI changes what someone is responsible for, it is worth checking whether existing evaluation criteria and hiring requirements have drifted out of alignment.

The percentages should not be transferred directly to Japan. The underlying questions, however — how AI affects roles, expectations, customer interactions, and personnel decisions — remain useful for Japanese companies reviewing their own adoption plans.

How adoption spread is separate from who is accountable

The survey also asked how day-to-day AI use had grown within the organization.

29% said it was mostly driven by ownership or leadership providing guidance, tools, or setting expectations. 26% said both contributed equally, and 19% said it was mostly driven by employees exploring AI and finding use cases on their own.

What this question captures, though, is how adoption spread — not where final decision-making authority or accountability sits.

Even when AI use begins on the front line, decisions about which data may be entered, which use cases are approved for official use, and who verifies the effect on customers and employees still have to be made separately.

Conversely, when adoption is driven from the top, there needs to be a channel for confirming whether workloads actually decreased in practice and whether any requirements were overlooked.

In companies that cannot easily staff dedicated AI, HR, or operations-design roles, this kind of review has to run alongside existing day-to-day work.

Company size alone, however, does not determine operational capacity. What matters here is not headcount but whether someone owns the ongoing review of roles and evaluation criteria, and whether time is set aside for it.

Five areas to track before and after AI adoption

The following is a starting point for translating these findings into your own situation.

It is intended as a starting point rather than a universal framework.

The relative weight and priority of each item will vary by industry and by the structures already in place.

• Roles and responsibilities: Which tasks AI supports and which judgments stay with people. After the system is running, check whether anyone’s scope has shifted in ways you did not intend.
• How AI-enabled gains are allocated: Whether time savings go to output volume, quality, learning, new responsibilities, reduced overtime, or rest. Check that you are not imposing a single use on everyone.
• Expectations of employees and how they are evaluated: Whether expected volume or quality has shifted since adoption. Review whether existing evaluation metrics ignore the gap between work where AI helps and work where it does not.
• Customer interactions and personnel decisions: Whether expectations around response speed, hiring requirements, or staffing assignments have changed — and whether that shows up in actual customer and employee data.
• How adoption spread, and who is accountable: Whether AI use began with leadership or the front line. Separately from that, confirm who approves official use, who owns the data, and who is responsible for periodic review.

Each of these is easier to discuss at the point when AI adoption is first being considered.

None of it needs to be settled completely up front.

Setting initial hypotheses before adoption and building a mechanism to revisit them against operational data and frontline feedback is the practical next step.

In closing

In the U.S. Chamber of Commerce Foundation survey, owners and operators of small businesses using AI reported largely positive assessments of task speed and output quality.

At the same time, a majority reported some impact on roles and responsibilities, on expectations for employees’ job performance, on customer expectations and behavior, and on hiring and personnel decisions.

This survey is based on employers’ self-reports, however. It does not directly measure actual productivity or formal policy changes, and it does not establish that AI caused the changes reported.

Even so, evaluating AI adoption purely on operational efficiency risks overlooking the changes taking place on the organizational side.

At the point of adoption, alongside the question of which model or product to use, it is worth confirming how work is divided, how AI-enabled gains are used, what is expected of employees and how they are evaluated, how customers are handled, how personnel decisions are made, and who holds formal accountability for ongoing use.

Setting initial hypotheses before adoption and revisiting them against operational data and frontline feedback can help organizations identify whether reported efficiency gains are translating into sustainable improvements.

It can also help surface confusion in day-to-day work or misalignment in employee evaluation earlier than would otherwise be possible.

About This Information

This article was prepared using information reviewed as of August 2026. Its primary sources are the U.S. Chamber of Commerce Foundation article “Small Business Owners Say AI Is Already Changing Work,” published on July 14, 2026, and the accompanying “Appendix: Survey Questionnaire” prepared with Ipsos.

The survey was conducted online and in English between May 19 and June 4, 2026, among 750 owners and operators of U.S. small businesses. In this survey, “small business” means a company with 500 or fewer employees, excluding sole proprietorships.

The credibility interval for all respondents is reported as plus or minus 4.4 percentage points. Many of the figures cited in this article — including 54%, 50%, 47%, 73%, 70%, and 65% — come from questions asked only of respondents whose businesses use AI. The published materials do not provide separate intervals for those subgroups.

The question about how employees use AI-enabled gains was asked only of respondents who reported a “mostly positive” impact on task-completion time and/or work-output quality.

The survey results represent perceptions reported by owners and operators. They are not a direct survey of employees or customers, an objective measurement of productivity, or an estimate of the causal effect of AI adoption.

“Generative AI at Work” by Brynjolfsson, Li, and Raymond examines a single company’s customer-support operation, including its subcontractors. Its main analysis covers 5,172 workers and uses a quasi-experimental design based on a staggered rollout. The paper also reports an earlier pilot involving approximately 50 workers, although the authors lacked information on the pilot’s control-group workers. Its findings cannot be generalized directly to other occupations or companies.

The five-part framework used in this article — roles, allocation of AI-enabled gains, employee expectations and evaluation, customer and personnel effects, and operational accountability — represents MIF’s analysis.