Can AI Improve Small-Business Customer Service?

The problem, in numbers

In a small business, customer service is usually whoever is free: the owner between jobs, a casual on a Saturday shift, a phone that rings out after hours. The quality of an answer depends on who happens to pick it up. AI promises to fix both halves of that, by answering when no one can and by helping whoever does answer to answer well.

Customers are wary of the first half. In a 2025 survey of 1,031 Australians, commissioned by a customer-messaging vendor and run by YouGov, 49% said they would rather start with a human agent even if it takes longer. The vendor's Australian release adds that 84% of consumers said being able to move easily from an AI agent to a human was important.^{Source 1} 78% of Australians had used AI tools such as chatbots, yet only 39% reported satisfaction with them, and the release says one in five consumers would stop using a service or switch to a competitor when automated service takes longer than expected.^{Source 1}

What improves, and why

There are two very different products sold under the same label. One is the assistant: AI that suggests a reply, finds the policy, or drafts the email, while a person decides what is sent. The other is the agent: a chatbot that answers customers on its own. Most small businesses meet the second first, because it is what website and messaging platforms switch on by default.

The quality gain is clearest for the first. An assistant spreads the habits of the best person on the team to everyone else. For a small business with high staff turnover or casual rosters, that is the real prize: a new starter who answers like someone with two years behind them. The bot answering alone can cover the hours no one is there, but it raises the floor only if it is held to a narrow, accurate script.

What the evidence actually says

The best study is a field experiment, not a vendor deck. Brynjolfsson, Li and Raymond followed the staggered rollout of an AI conversational assistant to 5,172 customer support agents. In the published version, access raised issues resolved per hour by 15% on average, including a 30% improvement for less skilled and less experienced workers, while the most experienced saw only small gains in speed.^{Source 3} The authors also found that AI assistance improved customer sentiment and reduced attrition among newer workers.^{Source 2Source 3} An earlier working paper reported slightly different figures (14% and 34%); we use the journal's.^{Source 2}

The peer-reviewed version adds the detail that matters for quality. Customers dealing with assisted agents were less likely to ask for a manager, and the most experienced and highest-skilled agents saw little productivity change and small declines in quality.^{Source 3} In other words, the tool lifted the bottom of the team towards the top, and did little for the top.

Two cautions keep this honest. The study covers one large support operation where every AI suggestion passed through a trained person; it is not evidence that an unattended chatbot on a florist's website performs the same way. And the survey numbers above come from a vendor-commissioned poll, which is useful for direction but not for precision.

Where it falls short

• The evidence is for assistance, not autonomy. No study of comparable rigour shows unsupervised small-business chatbots improving service quality.

• Experts gain little. An owner who already knows every answer may see almost no quality lift, and possibly a small dip if they lean on suggestions.^{Source 3}

• Confident errors. A bot that invents a refund rule or misstates a guarantee does damage at scale, out of hours, with no one watching.

• Customers want a way out. Not being understood is the top frustration, and a bot with no handover turns a small problem into a lost customer.^{Source 1}

Where governance decides the outcome

The cautionary case is Air Canada. Its website chatbot told a grieving customer he could apply for a bereavement fare retroactively, after booking; the airline's policy did not allow it. Air Canada argued it could not be held liable for what the chatbot said. The tribunal called that a "remarkable submission" and ordered the airline to pay C$812.02 in damages, fees and interest.^{Source 4}

Australian law points the same way. Treasury's 2025 review notes that misleading or deceptive conduct can be contravened without fault, and that an AI system's tendency to hallucinate does not warrant re-examining that rule. Its own example is a small business paying $1,000 a month for an off-the-shelf support chatbot: if the bot misleads customers, the small business is liable, though it may have a remedy against the supplier.^{Source 5} Treasury's advice is that businesses should make the limitations of the technology clear to consumers.^{Source 5}

Privacy is the second line. The OAIC recommends that public-facing chatbots are clearly identified as AI, that privacy policies explain how AI is used, and, as best practice, that personal and especially sensitive information is not entered into publicly available generative AI tools.^{Source 6} Safer, here, is earned by deciding in advance what the bot may never say, telling customers it is a bot, and putting a person within one step.

How to think about adoption

Start with the assistant, not the agent. Let AI draft replies, summarise the history and find the policy, and keep a person pressing send. That is where the evidence is strongest, and it gives the most to newer or casual staff. Measure it the way the field study did: how many enquiries are resolved, how often customers ask for the owner, and whether new staff reach a confident standard sooner. If those numbers do not move within a few months, the tool is not earning its place.

If a customer-facing bot is worth having, give it a narrow job: opening hours, booking, order status. Keep refunds, guarantees, pricing exceptions and complaints with a person, and make the handover obvious. Label it as AI. Review a sample of its conversations every week, and treat each wrong answer as a statement the business made, because in law it is one.