Customers Want AI Service, Just Not the One You Built

Customer service leaders put a larger share of their 2025 budget into AI than any other business function. Use of company-provided chatbots has not moved since 2022.

Both of those facts come from Gartner, drawn from two separate surveys, and together they describe the most expensive misunderstanding in customer experience right now. The instinct is to read flat chatbot usage as customer resistance to AI. The data says close to the opposite, particularly in this region. Customers have adopted AI for service enthusiastically. They have adopted somebody else’s.

What the numbers actually show

Gartner surveyed 3,566 B2B and B2C customers in February and March 2026. Customers were roughly three times more likely to use a third-party generative AI tool, something like ChatGPT, Gemini or Copilot, than a company-provided chatbot when resolving a service issue. Third-party use during service interactions nearly doubled in a year. Company chatbot use was statistically unchanged from 2022.

A separate Gartner survey of 1,303 senior leaders, run between January and April 2026, sits usefully next to that. Service and support leaders committed a median of 12% of their 2025 budget to AI, the highest of the ten business functions assessed. Only 24% of them could demonstrate positive financial returns across their AI use cases.

Gartner’s Eric Keller attributes the disappointing return to something other than the technology, saying it has “less to do with technology limitations” than with misalignment with customer expectations. That distinction matters, because the two diagnoses lead to opposite decisions. If the model is not good enough, you wait for a better model. If the design is wrong, waiting changes nothing.

There is a longer-running version of the same problem. Gartner reported in August 2024 that only 14% of customer service issues were fully resolved in self-service. Self-service has been failing to finish the job for years. Generative AI was widely expected to fix that, and in company-owned channels the usage data suggests it has not yet.

The gap is between answering and doing

The clearest signal in the Gartner data is about capability rather than sentiment. Among customers who use generative AI, 58% have used it to complete a task on their behalf. In B2B settings that rises to 74%.

Most company chatbots cannot complete a task. They were scoped as deflection: a natural-language layer over a help centre, built to answer a question and reduce ticket volume. That was a reasonable design in 2021. It is now a mismatch with what customers have learned to expect from AI everywhere else in their lives, which is that it does something.

A customer who wants to change a booking, submit a document, update a policy or chase a delivery is not looking for an article explaining how to do those things. When your bot offers the article and a third-party assistant offers to draft the complaint email, the bot loses on merit.

Why this bites harder in the Gulf

The regional picture makes the cost of getting this wrong higher here than in most markets, for two reasons that pull in the same direction.

The first is appetite. BCG’s biennial Digital Government Citizen Survey, which covered 44 countries and 48,500 respondents in 2026, reports GCC findings drawn from 1,500+ respondents across the UAE, Saudi Arabia and Qatar. Only 4% of GCC respondents preferred human-only service with no AI involvement, against almost one in ten globally. Preference for a hybrid model, with AI handling simple services and humans handling complex ones, rose nine percentage points to 50%, the highest of any region surveyed. Expectation that services should be AI-centred ran at 79% in the GCC against 66% globally.

That is a population telling you it wants this. Resistance is not the constraint.

The second reason is the benchmark. In 2026 the GCC was the only region surveyed where citizens rated public digital services above private sector ones, on BCG’s net measure of 69%. On the separate measure of those who simply believed public sector digital services were superior, the UAE reached 76%, and the GCC average of 72% compares with a global average of 44%.

That gap is closing, and BCG is explicit about why. The same net figure fell from 75% in 2024 to 69% in 2026, which BCG attributes to a marked improvement in citizens’ opinion of private sector digital services, and its recommendations note that the GCC private sector is now innovating faster than the public one. So the benchmark is not a fixed disadvantage. It is a moving target that some companies here are already meeting.

nn

Read that as a customer expectation problem rather than a public sector achievement. A UAE resident who renews a licence through a government app without friction, then opens your chatbot and receives a link to a help article, is not comparing you with your competitors. They are comparing you with the best digital service they used that week, and in this region that was probably not built by a company.

One caveat worth stating plainly: the BCG data measures government services, not commercial customer service. It tells you what GCC customers are willing to accept from AI and what standard they have been trained on. It does not directly measure their behaviour with a retailer or a bank. The inference that those expectations carry across is mine rather than BCG’s, though it is difficult to argue that people maintain two separate standards for what good service feels like.

Where the trust line actually falls

Commercial data from the region suggests customer comfort with AI is not uniform. It has a specific boundary, and it sits exactly where money moves.

Visa’s Stay Secure study, conducted by Wakefield Research in January and February 2026 across 5,800 adults in 17 CEMEA markets including the UAE, found that 85% of UAE consumers have used AI tools to assist with shopping: checking reviews and ratings (60%), comparing prices (59%) and finding gift ideas (55%). Ninety-three percent said new technologies including AI are making online shopping faster and easier than before.

Then the line. Only 32% would trust an AI agent to complete checkout.

That pairing is the most useful finding in any of this research. The same consumer who happily delegates research will not delegate the transaction. Willingness to use AI is high and willingness to hand over authority is low. They are different questions, and they get collapsed into one in most internal AI discussions.

What follows for anyone running this

Audit what customers are already asking third-party AI about you. They are doing it now, roughly three times more often than they use your own bot, and the answers they receive are assembled from sources you do not control. That is a brand and accuracy exposure before it is a service one, and it is measurable today.

Judge the bot on resolution, not containment. Deflection metrics reward a bot for ending a conversation. Customers reward it for finishing a job. Any organisation whose service AI dashboard leads with containment rate is optimising the wrong variable, and will keep reporting success while usage stays flat.

Scope for action, then for conversation. If the AI cannot change a booking, process a return or update a record, it is a search box with better manners. Integration into the systems of record is the expensive, unglamorous part, and it is the part that determines whether any of it gets used.

Build the human route deliberately, not as a failure state. The same Gartner survey found that 87% of customers say access to a human agent is essential when companies use generative AI, and that only 27% would be willing to try a chatbot again after a negative experience. Those two findings together mean the escalation path is not a safety valve. It is what makes the first attempt worth risking, and you get roughly one attempt per customer. That route is also staffed by people whose roles are being rewritten by the same AI programme, which is where most of this quietly comes undone.

Put the boundary where the customer puts it. Informing, comparing, drafting and summarising are low-resistance activities. Paying, committing, cancelling and authorising are not. A journey designed around that split will outperform one that assumes trust scales uniformly with convenience.

What this data does not tell you

Three limits are worth stating, because these figures get quoted loosely.

The Gartner customer survey is global and does not break out the GCC, so its behavioural findings are not regional. The BCG figures concern government services, as noted above. The Visa study covers 17 markets with 5,800 respondents in total and does not publish a UAE-specific sample size, so the UAE percentages come from a subset of unstated size.

None of that undermines the direction of the evidence, which is consistent across independent studies from Gartner, BCG and Visa. It does mean that anyone building a business case should treat these as indicative of a pattern rather than as precise regional measurements.

The uncomfortable summary

Most organisations are not facing customer reluctance. They are facing a product that does not do what customers now assume AI does, in a market where customers are more open to AI than almost anywhere else, and where the comparison set has been established by someone other than their competitors.

The spending has already happened. What has not happened, in most places, is the harder work underneath it: connecting the AI to the systems where things actually get done, measuring whether a customer’s problem went away, and deciding consciously which actions a customer will permit a machine to take on their behalf.

None of that requires a better model. It requires treating the problem as transformation and adoption work, which is what it is.

Martin Alva

Marketing, Brand & Digital Transformation Leader, United Arab Emirates

Senior Manager, Employee Experience (Brand Positioning, Digital Platforms & AI) at Space42 (A G42 & Mubadala Company). 15+ years in marketing, brand and digital transformation across MENA and Europe. 5 MENA Effie Awards.

More about Martin Alva → Connect on LinkedIn →