AI Customer Service Statistics: What the Data Actually Shows

The most cited number in this field comes from a peer-reviewed study of 5,179 customer support agents: access to a generative AI assistant raised the number of issues resolved per hour by 14% on average. That one result gets quoted in vendor decks, board papers and conference talks, often without the caveats attached. Most AI customer service statistics circulating online are weaker than that study, and many are surveys of what people expect rather than measurements of what happened.

This post collects the figures worth using, shows where they come from, and explains why they disagree. It draws on government statistics, a regulator, an academic study and two research and vendor surveys. It is written for readers in the US, UK, EU and beyond, and where a statistic belongs to one market we say so. Every figure here is third-party data. None of it is Klevere's own.

Quick answer

The best-evidenced finding is a 14% productivity gain for support agents using AI assistance, with a 34% gain for novices. Business adoption is rising fast but unevenly: about 20% of EU firms and about 35% of UK firms use AI. Customers remain split, with 64% in one Gartner survey preferring that companies did not use AI for service.

The statistics at a glance

The list below sets out the headline figures from each source, with the year and scope so you can judge how far each one travels. Read the scope before you quote the number, because a survey of service leaders and a measurement of resolved tickets are not the same kind of evidence.

How many businesses use AI in customer service

Government statistics give the cleanest view of overall adoption, though none of them isolate customer service on its own. Eurostat reports that 20.0% of EU enterprises with 10 or more employees used AI technologies in 2025, against 13.5% in 2024 and 8.1% in 2023. The spread between countries is wide: Denmark is at 42.0%, while Romania is at 5.2%. If you sell across the EU, the same service tool will meet very different customer expectations in different markets.

The UK figure is higher. The ONS reports that self-reported AI use among UK businesses with 10 or more employees rose from about 12% to about 35% between late 2023 and June 2026. Size matters: the same release puts use at 28% for the smallest businesses and 49% for those with 250 or more employees. Among the technologies named, large language models were used by 18% of businesses.

These two numbers should not be compared directly. The UK series covers a different set of sectors, uses a different survey and ends more than six months later than the EU series. What they agree on is the direction. Adoption roughly doubled or tripled in two to three years, and it is now normal rather than unusual for a mid-sized firm to use some form of AI.

The US has no single equivalent, but the regulator evidence is telling. The CFPB found that over 98 million people, about 37% of the US population, engaged with a bank chatbot in 2022, and that each of the ten largest US commercial banks had deployed one. That is a sector-specific figure, and banking is an early adopter, so treat it as an upper bound for customer-facing chatbot exposure rather than a national average.

For a deeper look at small-business uptake specifically, see our post on AI adoption statistics for small business.

What AI does to resolution and productivity

The study behind the opening figure is the strongest evidence in the whole field because it measured outcomes rather than opinions. The authors studied support agents at one company and found that AI assistance raised productivity, measured as issues resolved per hour, by 14% on average. The gain was concentrated among less experienced staff, at 34% for novice and low-skilled workers, with minimal impact on experienced and highly skilled workers. The abstract also reports improved customer sentiment and higher employee retention.

Two cautions apply. First, this was an assistant that suggested responses to human agents, not an autonomous agent answering customers alone. Second, it studied one firm using one tool. The direction is credible and the size is plausible, but the exact numbers will not transfer to your team.

Industry surveys point the same way on agent time. Salesforce reports that reps using AI spend 20% less time on routine cases, freeing an estimated four hours per week for complex work. The company itself notes that these findings may reflect selection bias among early adopters, which is a fair warning to apply to any vendor survey, including the ones in this post.

The Salesforce survey also contains the projection that gets repeated most: service teams estimate that AI handles 30% of cases today and expect it to reach 50% by 2027. Note the wording. It is an estimate by service professionals, not a count of resolved cases, and it is a forecast of what they expect rather than a measurement of what happened.

If your interest is automated resolution rather than agent assistance, our guide to chatbot resolution rate benchmarks explains why published containment and resolution rates vary so widely, and what each definition counts.

How customers feel about AI service

Customer acceptance is where the data is most divided. Gartner surveyed 5,728 customers in December 2023 and reported that 64% would prefer that companies did not use AI in customer service, and 53% would consider switching to a competitor if they found out a company was going to use it. The release says customers' top concern is that AI will make it more difficult to reach a person, followed by job displacement and wrong answers.

Zendesk's figures look quite different. Its statistics article reports that 81% of consumers believe AI has become part of modern customer service and that 51% say they prefer bots over humans when they want immediate service. Nearly 8 in 10 say bots are helpful for simple issues. The article does not state a sample size or survey year for each of these figures, so they are harder to weigh than the Gartner data.

These findings are less contradictory than they look. They ask different questions. Gartner asked about a company using AI for service in general, which invites a view on the principle. Zendesk's questions focus on specific situations, such as an immediate answer to a simple question. People tend to accept AI for quick, low-stakes requests and reject it as a wall between them and a human.

The CFPB report shows what happens when that wall goes up. It cites a survey in which 80% of consumers who used a chatbot left more frustrated and 78% needed a human afterwards, and it describes poorly designed bots creating doom loops that block access to human agents. That survey was cited in a 2023 regulator report, so it predates the current generation of models, but the failure pattern it describes is still the one customers complain about.

Why the statistics disagree

When two credible sources report very different numbers, the cause is almost always one of five things. Checking for these before you quote a figure will save you embarrassment later.

The realistic range

Putting the evidence together, a cautious reading looks like this. For agent productivity, a gain in the low double digits is well supported, with larger gains for newer staff. For business adoption, somewhere between one in five and one in three firms with 10 or more employees reports using AI, depending on the country and the definition. For automated handling, the figures that circulate are mostly expectations: about half of cases by 2027 is what service professionals say they expect, not what has been measured.

For customer acceptance, the honest answer is that it depends on the task and on whether a human remains reachable. A majority of customers in one large survey said they would prefer no AI in service, while large shares in other surveys say they like it for quick answers. The practical reading is to design for both groups: let AI handle the simple and urgent, and make escalation to a person easy and obvious.

How we compiled this

We searched for primary sources first: national statistics offices, a financial regulator, a peer-reviewed economics study, and then research and vendor surveys. We opened each page and checked the figure on that exact page before including it. Where a source did not state its sample or method, we say so. We excluded aggregator sites and anything we could not trace to an original publication.

Several figures are older than the date of this post. We have kept them because they are still the primary source, and we give the year each time. Please check the original before you quote a number in a board paper or a client proposal, because releases get revised.

How Klevere approaches AI customer service

Klevere is an AI agency that has deployed 500+ AI agents across 50+ projects and 12 industries, and we hold a 98% client retention rate. Those numbers shape how we read the statistics above: the studies that matter most are the ones that measure outcomes, and the clients that do best are the ones that decide upfront what they will measure.

Our approach starts with a support audit rather than a tool. We look at the volume and mix of enquiries, which ones are repetitive and low risk, and which ones need judgement or empathy. Then we design the agent around the first group, with clear rules for handing over to a person on the second. The customer acceptance data above is the reason we treat easy escalation as a requirement rather than a feature.

We also ask clients to agree measures before launch: resolution without escalation, handover quality, time to first response and customer satisfaction, tracked against a baseline. That is how you avoid arguing later about whether a dashboard number means anything. For the broader picture of how this works in practice, read our guide to AI support agents and our comparison of AI versus human customer support.

If you want to model the return on a support deployment, our post on how to measure AI ROI covers the inputs, and our support agent page describes what the finished system does.

Frequently asked questions

What percentage of companies use AI in customer service?

No official statistic isolates customer service. Government data measures AI use overall: 20.0% of EU enterprises with 10 or more employees in 2025 and about 35% of UK businesses of that size in mid-2026. Customer service is one of several common uses, so the share using it specifically is lower than these overall figures.

Does AI improve customer service productivity?

Yes, in the best-evidenced study. Support agents with an AI assistant resolved 14% more issues per hour on average, and novice agents gained 34%. The tool suggested responses to human agents, so the result applies to AI-assisted agents, not to fully automated service. Results at your company will vary with your tooling and processes.

Do customers prefer human agents to AI?

It depends on the task. Gartner found 64% of surveyed customers would prefer companies did not use AI in service, while Zendesk reports 51% prefer bots for immediate service. Customers appear to accept AI for simple, quick requests and resist it when it blocks access to a human. Offer a clear route to a person.

How much of customer service will AI handle by 2027?

Salesforce reports that service professionals expect AI to handle 50% of cases by 2027, up from an estimated 30% today. That is a forecast from a survey of 6,500 respondents, not a measurement. Treat it as a view of sentiment among service teams, and test it against your own case mix.

Are AI customer service statistics from vendors reliable?

They are useful but should be weighed carefully. Vendor surveys often have large samples but state limited method detail, and they may reflect early adopters. Prefer sources that publish sample size, dates and wording, and cross-check against government statistics or peer-reviewed research before using a figure in a decision.

Which AI customer service statistic should I trust most?

The peer-reviewed NBER study is the strongest because it measured outcomes for 5,179 agents rather than asking opinions. Government adoption statistics from Eurostat and the ONS come next. Surveys of customers and service leaders are worth reading for sentiment, but they are weaker evidence of what actually happens in a contact centre.

If you are weighing a support deployment and want figures tied to your own enquiry mix rather than industry averages, book a free AI audit and we will show you where the realistic gains are.