Executives & Directors · Module 03 · Chapter 004

AI and Customer Experience

Customers are not against AI; they are against service that takes more effort and hides the way to a person. Better experience becomes business value only when it changes what customers do: keep the product, come back, stop calling about the same problem. Leaders who aim AI at customer effort, across the whole journey, and who measure behavior rather than sentiment, are the ones who can prove the value.

≈ 15 min read

After this chapter you can

Explain why better customer experience becomes business value only when customer behavior changes.
Map where AI can change the customer journey, from discovery to renewal, and pick the stage where effort is most costly.
Use customer effort, not delight or satisfaction alone, as the lens for customer-facing AI.
State the EU and US position on telling customers they are dealing with AI, as of October 2026.
Distinguish what customers say from what they do when judging whether an experience change created value.

In December 2023, Gartner surveyed 5,728 customers about artificial intelligence in customer service. When it announced the results in July 2024, the headline, as reported in coverage of its press release, was blunt: 64 percent said they would prefer that companies did not use AI for customer service at all, and 53 percent said they would consider switching to a competitor if they learned a company was about to. Their top worry was not the technology itself. It was that AI would make it harder to reach a person¹.

Set that beside the one careful field study of AI in a live service operation, which found that when 5,172 support agents were given an AI reply assistant, their customers’ messages turned more positive and fewer asked for a manager².

The two findings do not contradict each other. Customers do not object to AI. They object to service that costs them more effort, gives them wrong answers or walls them off from a human. Whether AI improves the customer experience or damages it is a design choice, and it is made by leaders long before any customer types a question.

Experience becomes value when behavior changes

The first chapter of this module defined AI value as a measurable change in a business outcome. Customer experience is where that definition is most often forgotten, because better experience feels valuable on its own. It is not, at least not to the business. A smoother interaction becomes value only when it changes what customers do.

Figure 3.4.1 Experience is the middle of the chain, not the end. The value appears when customer behavior moves.

The middle link is the one that matters. Customers who find setup easy keep the product instead of returning it. Customers whose problem is solved the first time stop calling about it. Customers who trust the recommendations buy again. Each of these is a behavior, and each has a price on it. A higher satisfaction score that leaves every behavior unchanged is pleasant news with no value attached.

So the executive question for any customer-facing AI proposal is not “do we have an assistant?” It is three questions in a row: which customer problem are we solving, how will customer behavior change, and which business outcome will that move?

The whole journey, not the help desk

Ask a leadership team where AI meets the customer and most will name the help desk, perhaps a chatbot on the website. That picture is too narrow. Customer experience is the customer’s overall perception of the organization, built from every interaction across the journey: the effort each one took, the outcome it produced and how it felt.

Figure 3.4.2 AI can change every stage of the journey. Start with the stage where customers struggle most, not with the tool.

AI can act at each stage. A recommendation helps someone discover the right product. A plain-language answer helps them compare two models without reading a specification sheet. A guided setup gets them to first use faster, which is where many returns are decided. A warning sent before the customer notices a late delivery or a failing part turns a complaint into a moment of trust. A change in usage that signals a customer is about to leave can trigger an offer of help instead of a cancellation form. Most of these are capabilities rather than proven results: the published evidence of measured gains is strongest for service assistance and recommendations, so treat the rest as hypotheses to test.

The practical rule is to start with the stage, not the technology. Find the point in the journey where customers struggle most and where that struggle drives a costly behavior, such as returns, repeat contacts or cancellations. That is where the first AI investment belongs, even if it is less visible than a chatbot on the home page.

Effort is the experience that counts

Which part of the experience drives behavior? One of the best-known large-scale answers predates generative AI. Researchers at the Corporate Executive Board studied more than 75,000 people who had dealt with contact centers or self-service channels, and reported the results in the Harvard Business Review in 2010³.

Their first finding undercut the way most service organizations kept score. Satisfaction predicted loyalty poorly: 20 percent of satisfied customers said they intended to leave, while 28 percent of dissatisfied customers intended to stay. Their second finding was harsher. Service interactions were four times more likely to leave a customer disloyal than loyal. Delighting customers did little to win them, but making them work hard reliably drove them away³.

Figure 3.4.3 Most customers met at least one effort problem. Repeat contact, the most common, is also the easiest to measure.

This is a useful lens for a leader to bring to customer-facing AI. Its job is to remove effort: the repeat contact, the transfer, the second explanation, the switch from a website that failed to a phone line that works. The same study found that 57 percent of inbound calls came from customers who had tried the website first³. Every one of those calls is a customer who did the work twice.

Where AI removes effort

Across the journey, AI reduces effort through a small set of mechanisms. They are worth naming, because a proposal that cannot say which one it relies on usually relies on none.

Figure 3.4.4 Six mechanisms, one test - did the customer get a solved problem with less work?

Speed and relevance are the familiar ones: an instant order status, or three suitable products instead of three hundred. Consistency matters more than it looks. The same approved answer in every channel, country and language also opens service to customers who need voice, translation or accessible formats, although translation of legal, medical or technical content needs careful quality checks. Proactive service reverses the usual direction: the company spots the late parcel or the failing part and contacts the customer first. And AI inside the product lets customers state what they want to achieve rather than learn where every setting lives.

One illustrative picture captures the promise and its limit. The traditional path to a fix has five steps: search, read the results, open an article, try a fix, contact support. The AI-assisted path has three: describe the problem in your own words, let the system work out what you need, get a relevant solution. Fewer steps is the promise. But if the three-step path gives a wrong answer, or blocks the way to a person, the customer ends up taking all five steps anyway, plus one more. Count the steps to a solved problem, not the steps to an answer.

Personalization: relevant, not intrusive

Personalization is where experience and risk meet. Customers expect it: in McKinsey’s 2021 research, 71 percent of consumers said they expect companies to deliver personalized interactions, and 76 percent said they get frustrated when that does not happen⁴. AI makes personalization far cheaper, because it can draw on purchase history, product usage, location and past conversations at the moment of every interaction.

Figure 3.4.5 The target sits between generic and intrusive - relevant, using only data customers would expect you to use.

The same data that makes an offer relevant can make it unsettling. A reminder that a filter is due feels helpful; a message that reveals the company has inferred something the customer never shared feels like surveillance, and trust falls. Two rules keep personalization on the right side of the line. Use data customers would expect you to use, under your privacy controls. And recommend in the customer’s interest as well as your own: a system that only pushes the highest-margin product teaches customers to ignore its suggestions. Privacy obligations themselves are treated in depth in Module 06.

Trust: tell customers, and own the answer

Customer-facing AI errors are business errors, because the customer acts on them. A wrong refund rule or a wrong delivery date becomes the company’s problem the moment it is said; a Canadian tribunal held an airline responsible for what its website chatbot told a customer⁵. The company owns every answer its AI gives.

Trust also depends on honesty about what the customer is dealing with. Customers who discover late that they were talking to a machine feel deceived. As Where AI Creates Revenue noted, disclosing a bot can lower sales when the bot is weak⁶. The answer is to make the assistant good enough to be disclosed, not to hide it, and in Europe hiding it is no longer an option.

Figure 3.4.6 In the EU, disclosure is a legal duty; in the US it depends on the state. A single global default - disclose - is simplest.

The lawDisclosing AI to customers (as of October 2026)

In the European Union, the AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744) applies transparency duties under Article 50 from 2 August 2026: people must be told when they are interacting with an AI system, and synthetic and deepfake content must be labeled (generative systems already on the market before then have until 2 December 2026 for machine-readable marking). The interaction duty sits with the provider of the system, which is often the company that builds an assistant into its own service, and applies unless the AI nature is obvious to a reasonably well-informed person. The information must be clear and given at the latest at the first interaction⁷⁸. The United States has no general federal disclosure duty, and the position varies by state. California, for example, has since 1 July 2019 made it unlawful to use a bot online to mislead a person about its artificial identity in order to knowingly deceive them into a purchase or sale, or to influence a vote; clear and conspicuous disclosure removes that liability⁹. This is orientation, not legal advice. Confirm with counsel.

Disclosure is only the visible part of trust. Behind it sit the controls that make an answer safe to give: approved knowledge sources, identity checks before account changes, limits on what the AI may promise, and an easy route to a person, with the conversation so far attached. How to design that handover belongs to AI in Customer Service. The leadership decision belongs here: the route to a human is part of the experience you are selling, not a cost to be minimized.

Measure what customers do, not only what they say

Customer metrics come in two families, and only one of them proves value. The first records what customers say: a satisfaction score, a net promoter score, a customer effort score. These are useful early signals. The second records what customers do: whether they contacted you again about the same problem, returned the product, renewed, bought again or left.

Figure 3.4.7 Track both families, but book value only on what customers do.

Recall that 20 percent of satisfied customers in the effort study still intended to leave³. A rising satisfaction score with flat repeat contacts and flat returns should be read as a polite customer, not a changed one. The cleanest evidence comes from a controlled comparison, in which some customers get the new experience and a randomly chosen group keeps the old one¹⁰. How to set the baseline and design that comparison is the subject of Baselines, Metrics and Measurement, later in this module.

Story: two bets at one appliance maker

The following is an illustrative composite, not a documented case; the numbers are invented to show the arithmetic.

A consumer-goods maker sells about 400,000 home air purifiers a year in Europe and North America. Its leadership had budget for one customer-facing AI project and two proposals on the table, from two of its own teams.

The support team proposed an AI chat assistant on the support website, to answer questions without an agent. Its target was the number of contacts reaching people. The product team proposed something less visible: an AI setup and care guide inside the purifier’s companion app. It would walk new owners through setup in plain language, explain what each warning light meant, read the device’s error codes and message the owner before a clogged filter or a fault turned into a complaint. Like the chat, it would tell owners up front that they were dealing with AI, and pass warranty and safety questions to a person with a summary attached.

Figure 3.4.8 Illustrative. A won on the support dashboard; B changed what customers did.

Rather than settle the argument by opinion, leadership ran both as eight-week pilots, each against a randomly chosen group of customers who kept the old experience. Suppose the results came back as follows. The chat assistant cut contacts reaching agents by 30 percent, and customers rated it well. But repeat contacts about the same problem did not fall, and the 30-day return rate stayed at 7 percent. The setup guide cut agent contacts by only 10 percent. Yet among all new owners offered it, whether or not they opened it, 30-day returns fell from 7 percent to 5.5 percent, against 7 percent in the control group.

Because that fall was measured across everyone offered the guide, it can be scaled to the whole year: on 400,000 units, 1.5 points of returns is 6,000 fewer returned purifiers. If each return costs the maker about 90 in freight, handling and write-down, that is about 540,000 a year, before counting the owners who kept the product and went on to buy replacement filters. The chat’s savings were real but smaller and less certain, because a contact that does not reach an agent is not the same as a problem that was solved.

Leadership funded the setup guide first and folded the chat into it as the route for questions the guide could not answer. The lesson is not that chat assistants are bad. It is that the bigger value sat at the use stage of the journey, where effort drove returns, and only a measure of customer behavior could reveal it.

What this means for leaders

Customer experience is a source of AI value in its own right, alongside revenue and cost, but it is the easiest one to overstate. The discipline is to treat every experience improvement as a hypothesis about customer behavior, and to fund the improvement where that behavior is most expensive. Often that means looking beyond the help desk to the stages where customers give up: setup, first use, billing, renewal.

It also means accepting that trust is part of the product. Customers say they fear AI because they fear losing the human and getting confident wrong answers¹. A design that discloses the AI, keeps the person reachable and owns every answer removes the reason for that fear, and in the EU disclosure is now the law.

Check yourself

Because most customers say they prefer no AI in service, customer-facing AI should be avoided.
A satisfied customer can be relied on to stay loyal.
Reducing customer effort does more for loyalty than trying to delight customers.
Fewer contacts reaching agents proves that the customer experience improved.
In the EU, an AI assistant that talks to customers must make clear it is AI unless that is obvious.
The biggest customer-experience value from AI may sit outside the support desk.

Reflection: find the expensive effort

What comes next

This module has now looked at three sources of AI value: revenue, cost and the customer experience. The next chapter, AI and Workforce Productivity, turns inward to the fourth: how AI changes what employees do with their time and what they can take on.

Laws referenced

Not legal advice. Laws change; verify before relying on this, and consult counsel for decisions.

EU AI Act · EU

Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744

Risk-based rules. Prohibited practices include social scoring, untargeted scraping of facial images, and emotion recognition in workplaces and schools (with narrow exceptions). High-risk systems (Annex III: biometrics, safety components of critical infrastructure such as energy, water and traffic, employment and worker management, credit, education, essential services, law enforcement, migration, justice) need risk management, data governance, documentation, logging, human oversight, human oversight that keeps people able to understand the system, notice automation bias (over-reliance on its output), override it or stop it (Art. 14(4)), appropriate accuracy, robustness and cybersecurity (Art. 15), automatic logging of events (Art. 12), a provider quality-management system (Art. 17) and conformity assessment. An Annex III system is not high-risk if it poses no significant risk of harm, for example a narrow procedural or preparatory task that does not replace human assessment; systems that profile people are always high-risk, and a provider relying on this exception must document it and register (Art. 6(3)). Deployers of high-risk AI must use it as instructed, assign competent human oversight, monitor its operation, keep logs for at least six months and report serious incidents (Art. 26); employers must inform workers' representatives (Art. 26(7)). Public bodies, private providers of public services, and deployers of credit-scoring or life and health insurance pricing systems must carry out a fundamental-rights impact assessment before first use (Art. 27). Providers must run post-market monitoring (Art. 72). A deployer that puts its name on a high-risk system, substantially modifies it, or changes its purpose so that it becomes high-risk takes on the provider's obligations (Art. 25(1)). A substantial modification (Art. 3(23)) of a high-risk system needs a new conformity assessment, unless the change was pre-determined and documented at the first assessment, as with planned continuous learning (Art. 43(4)). Providers of general-purpose AI models (from 2 Aug 2025) must keep technical documentation, have a policy to comply with EU copyright law including text-and-data-mining opt-outs, and publish a sufficiently detailed summary of training content (Art. 53). Research, testing and development before a system is placed on the market or put into service is outside the Act, except testing in real-world conditions (Art. 2(8)). Since the 2026 Omnibus, the Art. 4 AI-literacy duty is an obligation of effort (take measures to support literacy), not of result. Fines reach EUR 35 million or 7% of global turnover for prohibited practices.

2024-08-01 — Entered into force
2025-02-02 — Prohibited practices (Art. 5) and the AI-literacy duty (Art. 4) apply
2026-07-27 — Omnibus softens Art. 4: providers and deployers must take measures to support AI literacy; no specific level must be guaranteed
2025-08-02 — General-purpose AI model obligations apply; governance and penalties regime in place
2026-08-02 — Transparency duties (Art. 50) apply: disclose AI interaction, label synthetic and deepfake content (marking for generative systems already on the market: 2 Dec 2026)
2027-12-02 — High-risk obligations for Annex III systems (e.g. hiring, credit, education, essential services) - moved from 2 Aug 2026 by the 2026 Omnibus
2028-08-02 — High-risk obligations for AI in products regulated under Annex I

Last verified 2026-10-06 · official text

References

Gartner. Gartner Survey Finds 64% of Customers Would Prefer That Companies Didn't Use AI For Customer Service. Gartner Newsroom. 2024.
Erik Brynjolfsson, Danielle Li and Lindsey Raymond. Generative AI at Work. The Quarterly Journal of Economics 140(2). 2025.
Matthew Dixon, Karen Freeman and Nicholas Toman. Stop Trying to Delight Your Customers. Harvard Business Review 88(7/8). 2010.
McKinsey & Company. The value of getting personalization right - or wrong - is multiplying. McKinsey & Company (Next in Personalization 2021 report). 2021.
Civil Resolution Tribunal (British Columbia). Moffatt v. Air Canada, 2024 BCCRT 149. CanLII. 2024.
Xueming Luo, Siliang Tong, Zheng Fang and Zhe Qu. Frontiers: Machines vs. Humans: The Impact of Artificial Intelligence Chatbot Disclosure on Customer Purchases. Marketing Science 38(6), 937-947. 2019.
European Parliament and Council of the European Union. Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. 2024.
European Union. Regulation (EU) 2026/1744 (Digital Omnibus on AI) amending Regulation (EU) 2024/1689. Official Journal of the European Union. 2026.
California State Legislature. SB-1001 Bots: disclosure (Business and Professions Code 17940-17943). California Legislative Information. 2018.
Ron Kohavi, Diane Tang and Ya Xu. Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing. Cambridge University Press. 2020.