How can AI improve customer experience

TL;DR

How Can AI Improve Customer Experience? (And Where It Makes It Worse)

  • The best-evidenced deployment is AI behind the agent: +13.8% resolution, better customer sentiment, fewer escalations, turnover down 8.6 points.
  • Customers' stated preference runs the other way from the industry's: 64% would rather companies didn't use AI in service at all.
  • The failure pattern is consistent - AI as a wall between customer and human, deployed on cost logic. Klarna ran that experiment and reversed it.
  • Placement decides everything: the same technology that damages experience in front of the person improves it behind them.

Ask the internet how AI can improve customer experience and you join one of the most crowded question queues in business. South African searches alone produce dozens of variants: how can AI improve customer experience, how AI is transforming customer experience, can I use ChatGPT for customer service.

The truthful answer has two halves, and companies keep learning the second half the expensive way. AI measurably improves customer experience in some deployments and measurably damages it in others - and the dividing line is not the sophistication of the model. It is where you put it.

How can AI improve customer experience?

The strongest evidence is for AI behind the agent rather than instead of them. In the landmark 5,172-agent field study published in the Quarterly Journal of Economics, a generative AI assistant lifted issues resolved per hour by 13.8%, improved customer sentiment and reduced escalations to supervisors - while the human kept the conversation. AI also improves experience through instant resolution of genuinely routine contacts, 24/7 availability for simple queries, and consistency. Gartner predicts agentic AI will resolve 80% of common customer service issues by 2029.

Unpack the two deployment patterns in that answer, because everything else follows from the distinction.

Pattern one: AI resolves the routine directly. The password reset, the delivery status, the balance query. Here full automation genuinely serves the customer - instant, correct, done at 2am. Nobody's experience is improved by holding for nine minutes to ask a question a machine answers in nine seconds. Gartner's prediction that agentic AI will resolve 80% of common issues by 2029 describes this layer, and the operative word is common.

Pattern two: AI assists the human on everything else. The complaint, the cancellation, the confused customer with three interacting problems. Here the evidence says keep the person and arm them - which is what the QJE study measured, and what the numbers below describe.

What the field evidence shows

How does AI help customer service agents?

By putting the accumulated craft of the best agents behind everyone. In the Quarterly Journal of Economics study of 5,172 agents, an AI assistant suggesting responses lifted average productivity 13.8%, with novice agents improving around 35% because the tool compressed months of tacit learning into weeks. Turnover fell 8.6 percentage points among agents with AI access, customers expressed more positive sentiment, and requests for a manager fell. The agent stayed in charge of the conversation throughout - the AI never spoke to the customer.

The customer experience consequences of that study are easy to miss because it is usually reported as a productivity result. Look at what customers got: faster resolutions, more positive interactions, fewer occasions where they felt the need to demand a manager - and, over time, a more experienced workforce, because the agents stopped leaving as fast. Attrition is a customer experience metric wearing an HR badge: every departed agent is replaced by a novice, and every novice conversation is, on average, a worse conversation. A technology that cuts agent turnover by 8.6 percentage points is quietly one of the biggest CX interventions on record.

+13.8% Issues resolved per hour with an AI assistant behind the agent - alongside improved customer sentiment and fewer escalations Brynjolfsson, Li & Raymond, QJE 2025
64% Customers who would prefer companies did not use AI for customer service at all - with 53% prepared to consider switching over it Gartner, July 2024
80% Share of common customer service issues Gartner predicts agentic AI will resolve autonomously by 2029 - leaving humans the uncommon ones Gartner, March 2025

The customer's verdict

Do customers like AI customer service?

Mostly not, when asked directly. Gartner's survey of 5,728 customers found 64% would prefer companies did not use AI in customer service, and 53% would consider switching to a competitor over it - with their top concerns being difficulty reaching a human, job displacement and wrong answers. A later Gartner survey found 54% of US customers trust a human agent more than AI for recommendations, against 32% the other way. Customers are not rejecting speed or automation; they are rejecting being walled off from a person when it matters.

Read the concern list carefully and the apparent contradiction with the field evidence dissolves. Customers who say they dislike "AI in customer service" are describing the wall: the chatbot that cannot understand the question, will not fetch a human, and answers wrongly with total confidence. They are not describing the assistant behind the agent - which, by design, they never see. The QJE customers whose sentiment improved did not know an AI was involved at all. They just experienced a person who was quicker and surer.

This is why the survey data and the deployment data can both be right. The 64% is a verdict on pattern-one-done-badly: full automation extended past the routine into territory it cannot handle. It is not a verdict on the technology's best use.

Attrition is a customer experience metric wearing an HR badge: every departed agent is replaced by a novice conversation.

Where it goes wrong

When does AI make customer experience worse?

When it is deployed as a cost wall rather than a service layer. The failure modes customers name in surveys: bot loops with no route to a human, confidently wrong answers, and forced deflection of problems too complicated for the script. Klarna provided the cautionary tale - after claiming its AI did the work of 700 agents in 2024, it reversed course in 2025, its CEO conceding that cost-driven deployment had produced lower quality service and that investing in human support quality was the way forward.

The Klarna story is told in full in the companion article, but its diagnostic value belongs here: the deployment was engineered around a cost number - the work of 700 agents - rather than an experience number, and the experience number eventually presented the bill. CEO Sebastian Siemiatkowski's own post-mortem named the mechanism: when cost is "a too predominant evaluation factor", what you end up having is lower quality.

A practical test falls out of all this evidence. Before any AI deployment in a customer journey, ask one question: does this put capability behind a person, or a barrier in front of one? Assistants, suggested responses, instant routine resolution with a visible human escape hatch - capability. Mandatory bot gauntlets, deflection targets, hidden phone numbers - barrier. The first pattern has a top-tier economics journal behind it. The second has 64% of your customers against it.

The human half of the augmented centre

One consequence of getting this right is under-planned almost everywhere. If the machines absorb the routine and the humans keep the difficult, the human job intensifies: all-hard queues, sustained emotional labour, judgement on every call. The experience your customers get in 2027 depends less on which model you license than on whether the people handling the hardest conversations are people that work energises rather than erodes.

That fit is readable. Sariio MAPS maps each agent's working preferences - contact load, pace, structure, autonomy - from a ten-minute survey retaken at least twice a year, and shows leaders where the redesigned work and the actual workforce match and where they grind. AI will keep improving the machine half of the augmented contact centre on its own schedule. The human half is yours to see to - and it is measurable now.

The companion pieces: the replacement evidence is in can AI replace a call centre agent; the South African jobs angle is in will AI take call centre jobs in South Africa; the engagement mechanics are in can AI improve employee engagement.

Sources