73% of US Small Business Owners Say AI Has Already Changed What Their Customers Expect

Ipsos surveyed 750 small business owners and operators for the US Chamber of Commerce Foundation between 19 May and 4 June 2026. The results were published in July, with a credibility interval of plus or minus 4.4 percentage points.

Two findings sit at exactly 73 percent. Seventy-three percent say AI has changed employee roles and responsibilities. The same proportion say it has changed customer expectations and behaviour.

The first of those gets all the attention. The second is the one that costs money, because a business can decline to adopt AI internally and still be judged against the response times, availability and clarity that customers have learned to expect elsewhere.

What does it mean that AI changed customer expectations?

It means the comparison set moved. Customers now experience instant answers, round-the-clock availability and immediate order visibility from the largest companies they deal with, and they carry that expectation into every other interaction, including with a twelve-person business that answers email twice a day.

This is not a new phenomenon, but the gap widened sharply. In the same survey, 54 percent of employers using AI reported a mostly positive impact on how quickly tasks get completed and 47 percent on the quality of work. Only three to four percent reported a mostly negative impact on those measures. Firms that adopted are getting faster while firms that did not are standing still, and customers experience the difference as a service gap rather than a technology gap.

What is an AI automation agency and what does one actually build?

An AI automation agency connects the systems a business already runs and inserts automated handling where the decisions are simple and repeatable. The output is usually a set of workflows rather than a product: enquiries captured and routed, quotes assembled, follow-ups sent on schedule, records updated without anyone retyping them.

Very little of it involves training models. The work is closer to plumbing than to research, and the skill is in knowing which joins are worth making. A good engagement starts by counting how often something happens and ends with a system that handles the common case reliably and escalates the rest.

Which processes give the fastest return?

The ones with high frequency, clear rules and a measurable delay today. Intake and follow-up almost always top the list, because leads arrive at unpredictable times and the cost of a slow response is a lost sale rather than an inconvenience.

The survey supports this reading of where the time goes. Among employers seeing positive effects, 65 percent said employees deliver more or higher-quality work, 56 percent said the time went into learning, planning or reviewing, and 36 percent said it removed overtime. The gain lands as capacity and quality, not as headcount reduction, which is why intake automation tends to pay back first: it protects revenue rather than trimming cost.

Scheduling, quoting and status updates follow for similar reasons. Each is frequent, rule-bound, and currently performed by someone whose actual job is something else.

What should a business be careful of?

Be careful of automating a broken process, of building on accounts you do not own, and of per-task pricing that behaves very differently at scale than in a demonstration. Each of these is cheap to avoid at the start and expensive to correct later.

Automating a broken process is the most common and least discussed. If the intake form asks the wrong questions, automation simply routes bad information faster. Fixing the process first is unglamorous and usually delivers a fair share of the benefit on its own, which is a good reason to be suspicious of any proposal that does not begin with how the work currently happens.

Ownership deserves the same scrutiny. If workflows run inside an agency's accounts, the business has rented a process rather than bought one, and it stops when the relationship does. Our position on this is set out on our what you own page.

How does this connect to marketing rather than operations?

Because the first process most businesses automate is the one where marketing hands over to sales, and that handover is where most spend is wasted. Advertising that generates enquiries nobody answers within the hour is not an advertising problem.

This is the part that gets missed when automation is treated as an IT project. A business can double its ad budget and see no change in closed revenue because the constraint sits after the click. Response speed, routing and follow-up consistency determine what proportion of demand converts, and those are automation questions rather than media questions.

The reverse is also true. Automating follow-up on a channel that produces almost no enquiries improves nothing. The two have to be looked at together, which is the approach we set out in our guide to digital marketing for accounting firms and apply across every sector we work in.

Why do small businesses stall on this?

Usually because the decision has no obvious owner. Automation sits between operations, marketing and whoever handles technology, and in a small business those are frequently the same overloaded person. The project is genuinely worthwhile and permanently third on the list.

The second reason is that the benefit is hard to picture in advance. A business owner can imagine a new website or a new advert. Imagining the absence of an interruption that currently happens forty times a day is harder, because the cost of that interruption has never been measured. It shows up as general busyness rather than as a line anyone can point to.

This is why counting is the whole first step. Once a business knows that enquiry handling consumes eleven hours a week and that the average first response takes six hours, the conversation stops being abstract. Both numbers can be collected in a fortnight with no technology at all.

How should results be measured?

Pick two numbers before anything is built, and be willing to be embarrassed by them. Time from enquiry to first response, and the proportion of enquiries that receive a second follow-up, will tell you more than any dashboard a vendor supplies.

Those two measures are deliberately unflattering. Most businesses believe they respond quickly and follow up consistently, and most are wrong, because the exceptions happen on busy days that nobody remembers afterwards. A fortnight of honest measurement generally produces the business case without anyone needing to argue for it.

What does a sensible first project look like?

One process, measured before and after, live within weeks rather than quarters. Pick the workflow you can count today, instrument it so you know the current response time and completion rate, then automate it and compare.

Resist the programme-shaped proposal at this stage. A first project exists to establish whether the business can absorb automation at all, which is a question about internal process discipline more than technology. Firms that clear that hurdle can expand quickly. Firms that cannot will not be rescued by a larger contract.

Agree a review at two weeks and six weeks, with a list of what the system handled correctly and what it did not. The exceptions found in that window are the actual specification for everything built afterwards.

What about the businesses that have not adopted AI at all?

They are not necessarily behind, but they are being measured against a standard they did not agree to. The Chamber Foundation figures describe expectations that have already shifted across the market, and a customer comparing two suppliers does not adjust for whether either of them uses AI internally. They notice which one replied.

That reframing is useful because it removes the pressure to have an AI strategy. A business does not need one. It needs to know how long it currently takes to answer a customer, and whether that number is defensible against the alternatives that customer has. If it is, no action is required. If it is not, the fix is operational and specific rather than strategic and vague.

Plenty of well-run businesses will look at their numbers and find nothing urgent. That is a legitimate outcome of measuring, and a better use of a fortnight than commissioning a programme nobody needed.

Does size change the answer?

It changes the sequencing rather than the principle. A business with three people should automate one handover and stop. A business with thirty has enough volume to justify connecting several processes, and enough internal variation that documenting the current state becomes the hard part rather than the build.

What does not change is the order. Count first, fix the process second, automate third. Businesses that invert this tend to buy a tool, discover it does not match how they actually work, and conclude that automation was oversold. The tool was not the problem, and neither was the ambition. The sequence was.

Where this leaves a US business in 2026

The Chamber Foundation data describes a market where customer expectations have already reset and roughly three-quarters of owners can feel it. The businesses closing that gap are not running elaborate AI programmes. They are removing the delays between a customer's action and the business's response.

That work is measurable, which is the main reason we like it. You can state the response time before and after, and the argument ends. Engagements built on that kind of evidence are the ones that last, and it is the basis of our 82% client retention.

If the gap you can feel is between enquiries arriving and anyone answering them, start there. Our AI and automation service, our analytics and reporting work and our work with businesses across the USA all begin from the same question: what happens, today, in the hour after somebody contacts you.

FAQs