Customer Experience Is Only Half the Job

January 1930. The Hotel New Yorker opens on Eighth Avenue with forty-three floors and 2,500 rooms, three months after the crash.¹ Its manager is Ralph Hitz, an Austrian who arrived in the country at fifteen and worked his way up from busboy. He will run the hotel, and then a chain of them, on one idea.

Every guest gets a card. The card records his birthday and his wedding anniversary, his credit standing, and what he likes: the room, the floor, the way he wants things done. It records which newspaper he reads at home, so that when a regular from Cincinnati books, the clerk pulls the card, the hotel orders that morning's Cincinnati paper, and it is on the table in his room when he walks in. Coloured tabs on the edge of the card say whether he wants no publicity, whether he is not to be given a room again, whether the address he gave should be checked. On his first stay he gets a letter. On his twenty-fifth, another. On his fiftieth, a suite for nothing. The hotel's own front-office manual, printed in 1931, has a department for this. It calls it guest history.

There is no computer in the building. There are filing cabinets, and there are thousands of cards in them, and a man who has decided that being recognised is the product. That is a single customer view — the thing the customer-experience industry has been promising to build for thirty years — on paper, in a drawer, in 1931. And the card also tells the hotel whether the guest pays his bills. Hold that second fact; it is the point of the essay.

You do not run hotels. But you run this scene. Somewhere in your company there is a record on every customer: an account, an app, a profile, a score. There is a journey map on a wall and a satisfaction number reported each quarter, and a line in the budget called customer experience that has grown every year since someone convinced the board it would. You know what the recognition costs. When the chief financial officer asks what it earned, you have a slide, and you do not quite have an answer. Hitz had one. It fitted on a card.

The essay runs from that filing cabinet to today's algorithms: how customer experience learned to scale, why the single customer view took thirty years to arrive, what the science of delight actually says, why the growth claim does not hold, and what the companies that made money did instead. It will take you about forty minutes.

The instinct Hitz was running on is older than his card, and it never needed one. In 1890, in Bombay, a man called Mahadeo Havaji Bachche began collecting hot lunches from homes in the suburbs and delivering them to offices in the city, for a small monthly fee.² The dabbawalas still do it; the word means the men who carry the dabba, the round steel lunchbox.

Picture the conditions. Mumbai's suburban trains carry six to seven million people a day, in trains rated for two thousand people that carry four and a half thousand at the peak, at what the railway itself classifies as super-dense crush load: fourteen to sixteen people standing in a square metre. In the monsoon the tracks flood and the trains stop. Through this, at its height, about five thousand dabbawalas moved close to two hundred thousand lunchboxes a day.

Each box is collected from a kitchen at about nine in the morning, carried by bicycle to the local station, sorted on the platform, loaded into the luggage van, sorted again at the city end, carried by handcart to an office and delivered before half past twelve; the empty box goes home the same way in the afternoon. A box changes hands several times and travels sixty or seventy kilometres. The sorting runs on a code painted on the lid (a colour for the origin station, a number for the destination, a mark for the building and floor), because for most of its history the workforce could not read. A man carries about forty boxes at a time.

Harvard has taught the case since 2010, and its authors' summary is careful: mistakes are extremely rare. The famous "one error in six million" is a reporter's estimate from 1998 and has never been measured, and the essay does not repeat it. It does not need to. A hot lunch, on time, every working day, through that, for a century, on no software at all, is the most demanding piece of customer experience in this essay.

The dabbawalas needed no database. They needed someone deciding that the customer's experience was worth designing, and then designing it. That is the half of customer experience the industry got right, and it is the half this essay honours before it complicates it.

The first wave of technology did not invent customer experience. It let one hotelier's card index scale past one hotelier's memory, and it did so in three directions.

In 1967 AT&T launched its interstate 800 service: a number a customer could dial from anywhere in the country, with the company at the other end paying for the call.³ Before that, if you wanted to ask a company a question you paid to ask it. Afterwards the question travelled at the company's expense, which is a small thing on a phone bill and a large thing in a relationship. The early subscribers were big corporations that could afford the fixed line; within a few years the mail-order firms were printing the numbers in their catalogues and taking the order on the phone. Every contact centre, chat window and careline since is a descendant.

Ten years earlier American Airlines had signed an agreement with IBM to build a reservation system.⁴ Sabre was installed in 1960 in Briarcliff Manor, New York, and fully running by 1964, the largest commercial real-time computer system in the world; it took bookings from every agent at once and knew where every seat was. In 1981 the airline used what the system held. AAdvantage launched on the first of May, the first modern frequent-flyer programme, and its first members were reportedly found by searching Sabre for the phone numbers that kept coming back — about 130,000 people the airline already knew but had never recognised. United had its own version running within about ten days. This is the first single customer view at national scale: one record per passenger, built from the operational system, and used to recognise him by name.

In 1994 FedEx put its package-tracking system on the web.⁵ The company had tracked every parcel centrally since 1979 and scanned them with handheld devices since 1986; what changed in 1994 was who could see the result. The customer typed a number and saw where her parcel was. FedEx says it was the first transport company to do it. Whether or not it was first, it was a new kind of customer experience: the company handing the customer its own visibility, and calling the reassurance a feature.

Read those three as the primitives of everything the CX industry now sells. The 800 number is access: the customer can reach you. The reservation system is recognition: you know who she is when she does. The tracking page is transparency: she can see what you see. Every contact centre, loyalty programme and order-status screen since is one of the three, scaled. And note what none of them yet does. Each makes something visible — who calls, who flies, who is waiting — and none of them, in 1994, is paired with a decision taken on what it sees. The delight half of customer experience had learned to scale. The other half had not started.

By the 1990s companies could imagine the customer experience they wanted. Recognise me everywhere. Remember what I bought. Anticipate what I need. In 1993 two consultants, Don Peppers and Martha Rogers, gave the ambition a name in a book called The One to One Future: a company would treat each customer as a market of one, and the computer would make it affordable.⁶ The industry's term for the record that would make it possible was the single customer view: every interaction a customer has with a company, in every channel, joined into one file. Hitz's card, for everyone, updated by machine. The vision was clear. The machinery could not hold it.

Tesco learned this first, because it tried first. In November 1993 it began testing a loyalty card in three stores, then in about a dozen: a plastic card with a magnetic stripe, shown at the till, that recorded what the shopper bought and, for the first time, who bought it, in return for one per cent back in vouchers by post.⁷

To make sense of the data Tesco hired dunnhumby, a small analysis firm in west London founded a few years earlier by a married pair of statisticians, Clive Humby and Edwina Dunn — about twenty-five staff, three of them on the Tesco account. When the pair presented the trial results to the board in late 1994, the chairman is reported to have said that they knew more about his customers after three months than he did after thirty years.⁷ The line is their own recollection, published nine years later; the board's vote is on the record, and Clubcard launched across Britain in February 1995. On the independent shopping panel for the following two months, Tesco's share of grocery spending was 19.1 per cent against Sainsbury's 18.5.⁸ It was the first time Tesco had ever been ahead.

Then the data came in faster than anyone could use it. The computers of the day could not hold a year of a nation's shopping, let alone query it; dunnhumby could analyse about ten per cent of the card data at a time, one product group a week.⁹ So they made a choice that became the method: rather than know everything about every customer, they would answer, in Humby's phrase, "the biggest questions about most customers." Who was price-sensitive, and on what. Which products brought people into the store. Who had a baby. Who had stopped coming. Ten per cent of the data, well chosen, was enough to run a business on, because somebody had decided which questions mattered before the data arrived.

Most companies did not make that choice. They bought the whole ambition instead. Through the 2000s it was sold as customer relationship management, then as the single customer view, then as Customer 360, then as the customer data platform, and it is sold now as AI-powered customer experience. Each name promised the same thing the last one had promised: every interaction, every channel, one record, one person, seen whole. Companies bought it, found the records would not join, the channels would not talk and the people in the branches would not type, and bought it again under the next name. Anyone who has sat through a data-migration steering committee knows the sequence. The ambition kept its shape across three decades because it was the right ambition; the vendors kept renaming it because the last version had not delivered it. They were not foolish, and neither were the buyers. The ambition was real. The plumbing was not.

For twenty years the customer-experience industry sold a single customer view the infrastructure could not deliver, and a growth promise the view was supposed to make good. The vision was not wrong. It was early — and, as Section VI will show, the growth promise was something else again.

Several things that were impossible five years ago are now so ordinary that nobody notices them. That is what closing a gap looks like, and it is worth being fair about it before the essay turns critical.

Stand at a bus stop in Singapore. The panel above you says the 67 is four minutes away and the 133 is nine. Every bus on the island reports its position every thirty seconds, and the Land Transport Authority combines that with the history of the route to estimate the arrival; the same feed goes to every map app on the island.¹⁰ No loyalty scheme is involved. Nobody has joined anything. And the panel does something for the wait that has nothing to do with the bus arriving sooner. In 1985 David Maister, then at Harvard Business School, wrote down eight propositions about waiting, and the fourth was that uncertain waits feel longer than known, finite waits.¹¹ When researchers in Seattle later tested it on bus passengers, those with real-time arrival information on their phones judged their waits about a minute shorter than those without, five minutes against six — and, oddly, waited less in fact, about two minutes less, because they timed their walk to the stop.¹² The passengers without the information overestimated how long they had stood there; the ones with it were about right. This is customer experience from a public body, with no revenue motive at all: the authority spent its visibility on the passenger, and the passenger got some of her afternoon back.

Waze does the same thing in a private company. Every driver with the app open sends her speed and position; the service uses that to work out where the traffic is and reroutes everyone else around it.¹³ Drivers also report the accident, the hazard, the police car. The company's own description is plain: the more people drive with Waze open, the better the navigation. The data is not collected to price anyone or to sell them anything. It is the product, and every user makes it better for every other user.

Banks have found the same move. DBS, Singapore's largest bank, describes one of the nudges it sends to customers: by analysing past spending, it predicts an upcoming payment and warns the customer of a shortfall before the payment bounces and a fee is charged.¹⁴ That is the bank's own account of the mechanism, and the bank's own account of the scale — over a billion nudges to thirteen million customers in a year, and what it calls three-quarters of a billion Singapore dollars of economic value, a figure that bundles revenue gained with cost avoided and should be read as the bank's arithmetic.¹⁴ But look at what the nudge itself is. It is the bank spending its visibility to save the customer money the bank could have collected.

And there is a quieter form of it, where the customer simply tells the company what she wants and the company remembers. Since 2012 a Sephora shopper has been able to have her skin scanned to a shade number, which then attaches to her profile and follows her online, so that the foundations she is shown are the ones that match.¹⁵ She declares her skin type and her hair type and her concerns, and the recommendations change. Nothing was inferred; she said it. It is Hitz's card, filled in by the guest — personalisation on declared preference, which is the kind that never needs an apology.

What changed, technically, to make these routine? Customer data that used to move in overnight batches between hand-wired systems now moves within seconds of the event; "the same customer" used to mean the same email address and now means a scored probability that two records are one person; the analysis used to live in a warehouse while the action lived somewhere else, and the decision can now be pushed into the app or the call script while the customer is still there; and everything a customer said in a call or an email used to be lost, because nothing could read it, and now a language model can. Fig. 2 sets it out in the trade's terms and in plain ones.

None of this makes the growth claim true. All of it makes the delight real. Modern customer experience can recognise a customer, anticipate her and adapt in the moment — the thing the 1990s could only promise, the 2020s deliver as a matter of course — and the practitioners who built it are right to be proud of the plumbing. Which leaves a question the plumbing cannot answer: now that the single customer view exists, what does it earn? Before that, the science of the delight itself.

Now turn the conversation around and look at customer experience from the customer's side of the counter, because the best of it was never really about technology. It was designed around how people feel time, waiting, effort and fairness — and there is real science under that, most of it thirty years old, most of it ignored by the vendors who sell the platforms.

A story told in the New York Times has an airport in Houston flooded with complaints about the wait at baggage claim.¹⁷ The airport added handlers and cut the wait to eight minutes, and the complaints continued. So it moved the arrival gates further from the terminal and sent the bags to the farthest carousel. Passengers now walked six times as far and waited almost not at all, and the complaints stopped. The story names no official and no date, and should be taken as a story. But Maister's first proposition, from 1985, says why it would work: occupied time feels shorter than unoccupied time. A walk is occupied. Standing at a carousel is not. His fifth belongs on every service designer's wall — unexplained waits feel longer than explained ones — and between them the two account for most of what a good queue does: it gives you something to do and tells you why.¹¹ Neither costs anything, and both were available before the first computer was switched on.

Daniel Kahneman and his colleagues found something stranger. In a 1993 experiment, people held a hand in painfully cold water for sixty seconds, and on another occasion for sixty seconds followed by thirty more in water very slightly warmer.¹⁸ Asked which to repeat, most chose the longer one. The extra half-minute of lesser pain had improved the memory of the whole. In 2003 the same team ran a randomised trial on 682 patients having a colonoscopy, and in half of them left the instrument resting, causing little discomfort, for up to three minutes at the end.¹⁸ Those patients remembered the procedure as less painful, and were somewhat more likely to come back for the next one. People do not remember the average of an experience. They remember the peak and the end, and they barely register how long it took. Kahneman called the last part duration neglect, and it has an uncomfortable consequence for anyone who budgets customer experience by the minute: an hour of adequate service is remembered as whatever its worst moment and its last moment were. A farewell matters more than a lobby. The hotel that gets the checkout right has bought more memory than the one that gets the check-in right, and for less.

Then the finding every tracking screen is built on. In 2011 Ryan Buell and Michael Norton showed people a travel website that made them wait while it searched.¹⁹ When the screen showed the work — searching this airline, now that one — people valued the result as highly as if it had been instant, and preferred it to a faster site that showed nothing. The authors called it the labour illusion: seeing the effort makes the wait feel honest. Domino's had launched its pizza tracker three years before the paper; Uber's small car crawling across the map is the same idea, and so is a chef you can see. In a later field experiment in a cafeteria, when cooks and customers could see each other, diners rated the food twenty-two per cent better, and the cooks got faster.¹⁹ The boundary is worth knowing: the effect reverses when the outcome is bad. Showing the work only helps when the work is good.

And sometimes the experience was designed into the object itself, before any department existed to own it. In 1989 Sam Farber, a retired housewares executive, watched his wife Betsey, who had arthritis, struggle with a vegetable peeler in a rented kitchen in France.²⁰ He went to a design firm with a brief that the new tools had to work for everyone, so that nobody using them would feel singled out. OXO Good Grips launched in 1990 with fat, soft, finned handles, and every kitchen drawer in the world now has something shaped by that afternoon. Heinz did the same for ketchup in 2002, turning the bottle upside down so the sauce sat at the cap.²⁰ Nobody in either company called it customer experience. They called it making the thing work.

So delight is not soft. It is a design discipline with a literature: waiting has propositions, memory has a rule, effort has an illusion, and all of them can be engineered into a queue, a screen or a handle for almost nothing. That is the case for taking customer experience seriously, and it is stronger than anything in a vendor's deck. It is also what makes the next section necessary. If delight is this real, this designable and this cheap, and companies have spent thirty years and hundreds of billions on it, the growth it was supposed to produce should be easy to find in the accounts. Go and look.

Which brings the essay to the spreadsheet, because there is one, and the customer-experience industry has been standing on it for more than three decades.

You have seen the chart. Two lines, or two bars: the companies that lead on customer experience, and the ones that lag, and the leaders racing away. It has been redrawn every year since the 1990s, and its greatest hits can be said in one breath. CX leaders grew revenue five times as fast as laggards, or fourteen percentage points faster, depending on which vendor drew it.²¹ A modest improvement in customer experience is worth about 775 million dollars over three years to a company with a billion in revenue.²¹ A portfolio of CX leaders beat the S&P 500 by 415 points.²¹ Each of these is presented as a discovery. Each was built as a correlation, and the people who built them say so, in the methodology sections and footnotes that nobody who bought the slide has read.

Three things are wrong with the chart, and none of them is subtle.

The first is that nobody controls for the obvious. Companies that are winning can afford lovely experiences. A retailer with a fat margin can staff its stores, answer its phones and take back the returns without asking; a retailer in a price war cannot, and its customers notice. The chart cannot tell you whether the experience caused the growth or the growth paid for the experience, and the firm behind the best-known version wrote the words "correlation is not causality" under its own finding. Any chief financial officer would want to see the chart drawn with market share and gross margin controlled for. Nobody has drawn it.

The second is how the pairs were chosen. That firm's headline study compared five pairs of companies, picked by hand: Amazon against Walmart, Southwest against United, and so on. Two years later its own follow-up, using its full index rather than five pairs, found that the correlation between CX scores and share price was weak — and that in seven of fifteen industries the laggards had out-returned the leaders.²² The confession is on the record. It did not make the slide.

The third is the number under all the others. In 1990 two consultants at Bain, Frederick Reichheld and Earl Sasser, published an article in the Harvard Business Review containing the sentence that paid for three decades of conferences: companies can boost profits by almost a hundred per cent by retaining just five per cent more of their customers.²³ It is a spreadsheet. Take a customer's value over the years she stays, assume fewer customers leave each year, and she stays longer and is worth more. It was arithmetic about what would happen if defection fell. It was never a study of companies that reduced defection and then earned more, and the article does not claim to be one. Its readers did the claiming — and the same author's Net Promoter Score, launched in 2003 as "the one number you need to grow", was correlated in its founding paper with growth that had already happened.²³

Adelaide, 2010. Byron Sharp, a marketing professor at the University of South Australia, takes the sentence apart in How Brands Grow, and what he offers is not a quibble but a different theory of where growth comes from.²⁴ Start with the arithmetic. "Five per cent more" means five percentage points: a company losing ten per cent of its customers a year must lose five, which is to halve its defection rate, and the spreadsheet assumes it can do so for nothing. As Sharp puts it, "Reichheld and Sasser did no research that revealed that companies that reduced defection rates increased profits."

Then the empirical law that makes the halving unavailable to most. Across categories and decades, a brand's defection rate is set mostly by its size. Small brands lose a larger share of their buyers than big ones, and no amount of wanting changes it; the pattern is called double jeopardy, because the small brand is punished twice, with fewer buyers and less loyal ones, and it has held in every market anyone has measured. The practical meaning is blunt. A brand's retention rate is largely a consequence of its market share, not a lever for changing it. When Sharp and his colleagues later decomposed where growth actually came from, acquisition mattered roughly twice as much as reduced defection: the brands that grew were the ones that recruited more buyers, not the ones that held the buyers they had.²⁴

The loyalty schemes designed to hold customers fared no better under the same lens. Australia's Fly Buys produced what Sharp called "weak excess loyalty" in two of six brands, and the same pattern appeared among shoppers who were not members.²⁵ French panel data showed that the heaviest buyers joined a grocer's scheme first, because they were already loyal, and that whatever change the scheme made had eroded within nine months. Two other researchers summarised a decade of evidence in one sentence: "Most schemes do not fundamentally alter market structure."²⁵ None of this says that recognising a customer is worthless. It says that recognising her does not, by itself, make her stay, and that a CX strategy built on her staying is built on sand.

Put the two halves together. The chart that sold the customer-experience industry is the one document in it that would not survive due diligence. The companies that actually made money from customer experience never relied on it, because they were doing something the chart does not measure.

What they were doing is best seen where it began. Tesco's card gave its members one per cent back on their shopping, as vouchers, by post. That was the delight, and it was real; people liked the vouchers, and more than ten million people used the card regularly.²⁶ But the card also told Tesco who bought what, and Tesco used it to decide who got which coupon. A coupon mailed to everyone was redeemed by about three people in a hundred.²⁶ A coupon mailed to the people the data said would want it was redeemed by more than twenty. By February 1999 the quarterly statement went out in eighty thousand variations of letter, offer and magazine, and the vouchers issued since launch had passed half a billion pounds. Underneath, the audited accounts: Tesco's sales went from 10.1 billion pounds in the year Clubcard launched to 18.8 billion five years later, and it has not been behind Sainsbury's since.²⁶ Ireland, central Europe and an acquisition in Scotland are inside those numbers too; Clubcard is not the whole story. It is the part of the story that made Tesco the company that bought the rest.

Look at what is inside the envelope. The voucher is the thank-you; the choice of which voucher is the decision; and the customer sees only the first. That is the whole of the machine, in 1999, on paper, four years after a magnetic stripe: the delight the customer saw, and the decision she did not, in the same envelope. Seeing is not deciding. The winners did both.

Section VI made a claim that a CX director will want tested against her own industry: that the money in customer experience arrives not from the delight but from the decisions the delight makes possible. So here is the pattern six times over, each time in three parts — the experience the customer sees, the decision the company takes on what the experience reveals, and the growth number that follows, graded. Retail and airlines have had their turn; what follows is coffee, formula milk, banking, telecoms, jet engines and software, and the shape does not change.

Coffee. Starbucks Rewards is, to its members, an app that lets them order ahead, skip the queue and earn a free drink; in the latest quarter about one in three transactions in a company-run store in the United States was ordered on the phone before the customer arrived.²⁷ That is the delight, and it is good enough that members now account for nearly sixty per cent of the company's US revenue. The decisions sit behind it. Since 2019 the company has run an in-house analytics programme it calls Deep Brew, which by the chief executive's description would "power our personalization engine, optimize store labor allocations and drive inventory management": which offer goes to which member, how many baristas a store needs on Tuesday, what to stock. And there is a decision the customer never sees at all. Money loaded into the app is held by the company until it is spent — 1.75 billion dollars of it at the last year-end — and the slice that is never spent is booked as revenue: 222 million dollars last year, at almost no cost.²⁷ The experience is a queue skipped. The growth is sixty per cent of revenue through one channel, and a balance sheet that borrows from its customers at no interest.

Formula milk. In Britain a company may not advertise infant formula to the public, hand out samples or send its marketing staff to approach mothers; the rules descend from a World Health Organization code of 1981 that most of Asia has also written into law.²⁸ So each of the four biggest brands in the British market — Aptamil, Cow & Gate, SMA and HiPP — runs a careline: a helpline staffed by midwives, nutritionists and feeding specialists, reachable around the clock by phone and, now, by WhatsApp, where a parent with a feeding question can ask it. Do parents use them? When the British competition regulator surveyed parents in 2025, fifty-five per cent said the availability of expert support through a brand's careline was important to them, and Nestlé has said that a change to a formula's packaging alone produces a measurable volume of calls.²⁹ No company publishes the call count, and the essay does not invent one. The decision is in the privacy policy. One brand's policy states that it records the baby's date of birth and feeding type and the call itself, and uses them to build profiles, send stage-based communications and find look-alike audiences.²⁹ That is the machine, stated by the company: the experience is a midwife at three in the morning; the decision is which mother gets which message at which week of her child's life. The WHO's own monitors describe these carelines and the baby clubs around them as the industry's principal route to mothers. No growth number is published, and the regulation tells you why one is not needed: a company does not build a 24-hour midwife service in a market where it may not advertise unless the service is doing the advertising's work.

Banking. In 1988 two consultants, Richard Fairbank and Nigel Morris, persuaded a regional bank in Richmond, Virginia, to run its credit-card business as a laboratory.³⁰ The experience was an offer in the post that seemed made for you — a lower rate, a balance transfer, a fee waived — and it was, because every offer was a test against a control group, and the winners were mailed to millions. The decision was the price of credit to each individual, set by what the bank had seen of people like her. By 1999 the business, spun out as Capital One, had nearly seventeen million customers and ran tens of thousands of experiments a year; a decade earlier it had been one regional bank's card division. There is no loyalty programme anywhere in this story. The growth came from deciding, one customer at a time, whom to offer what. The counterweight is also on the record: a former employee later wrote that the same machine had been pointed, with the same rigour, at people who could least afford the credit it was extending.³⁰ The machine is indifferent to what it is pointed at, which Section VIII takes up.

Telecoms. In 2018 Eva Ascarza published two field experiments, one of them at a mobile operator in the Middle East with more than four million subscribers.³¹ The company took 12,137 customers it wanted to keep, gave two-thirds of them a text offering extra credit if they topped up within three days, and left the rest as a control. The experience is the offer. The decision is who gets it, and here the experiment is uncomfortable for every churn model in the world. Had the operator sent the offer to the forty per cent of customers its model said were most likely to leave, churn would have fallen by 1.9 percentage points. Had it sent the same offer to the forty per cent whose behaviour the offer was most likely to change, churn would have fallen by 6.0 points — three times the effect from the same budget. The two lists barely overlap: of the customers in the riskiest tenth, only sixteen per cent were also in the most persuadable tenth. Predicting who will churn is the dashboard. Predicting whom an offer will move is the decision, and they are different customers.

Jet engines. In 1962 Bristol Siddeley, later part of Rolls-Royce, offered airlines a deal it trademarked as Power by the Hour: the airline paid for engines by the hours they flew, not for repairs.³² By the late 1990s the descendant of that contract was called TotalCare; by 2009 about half of Rolls-Royce's civil engines flew under an agreement of the kind, and today more than ninety per cent of its Trent fleet does. The airline's experience is engines that do not surprise it. Underneath, about twenty-five sensors on each engine report temperatures, pressures and vibration several times a flight, over the aircraft's radio link, to a room in Derby staffed around the clock, and that visibility decides when each engine comes off the wing, what the contract should cost, and how the spares are planned. The growth number is on the income statement: of Rolls-Royce's 4.6 billion pounds of civil aftermarket revenue in 2023, 3.3 billion came from these long-term agreements.³² The customer bought certainty. The company bought the data that lets it sell certainty at a margin.

Software. Snowflake, a data company in Bozeman, Montana, reported a net revenue retention rate of 126 per cent for its 2025 financial year: the customers it had a year earlier were paying 26 per cent more.³³ Product revenue grew thirty per cent, so most of the growth came from existing accounts before a single new customer signed. The experience is software that works and scales without a call. The decision is which account gets the call: the company can see, hour by hour, how much each customer uses, and usage tells it which accounts are about to expand and which are about to leave, and its sales force is deployed on that list.

Six industries, one pattern, and the same three-part shape every time. In every case the experience is genuine: the queue is skipped, the midwife answers, the engine does not fail, the software works. And in every case the growth number belongs to the decision, not to the delight. Take the decision out — send every coupon to everyone, price every card the same, call every account — and the delight stays just as delightful while the growth line goes flat. That is what a CX budget is for, and it is what most CX budgets never fund.

Section VII has a problem inside it, and a reader running a CX programme will have felt it. If the money is in the decision, and the decision is taken on what the customer let you see, then the customer is being used — and Capital One's machine, pointed at the people who could least afford it, is the same machine as Tesco's. What separates the two? Not the technology. The essay's answer is a test, and it comes from a coffee counter.

You pay a dollar, the barista says thank you, and you say thank you back. In 2007 the American journalist John Stossel called this the double thank-you moment and pointed out how much economics lives in it: both parties say thanks because both are better off — you wanted the coffee more than the dollar, the shop wanted the dollar more than the coffee.³⁴ Nobody was coerced and nobody was fooled, and the two thank-yous, one from each side, are the proof. Amit Varma has carried the phrase to Indian readers for a decade, in his columns and on The Seen and the Unseen, as the plainest test of whether a trade is fair: if only one side has reason to say thank you, look harder at the trade.

Now apply it to customer experience, because every act of it is a transaction, and both sides are supposed to leave it better off. The customer's thank-you is for the delight: the queue skipped, the bus time right, the midwife on the line. The company's thank-you is for what the customer gave it in return — her business, and her visibility: what she bought, where she stood, what she asked at three in the morning. And a company only has reason to say thank you for visibility if it does something with it. A CX programme that delights and then decides nothing is a one-way thank-you: the customer's. The company has spent the money and has nothing to be grateful for, which is the state Section VI described and most CX budgets are in. A programme that uses the visibility against the customer is a one-way thank-you the other way: the company's alone, and the customer's withdrawn the day she finds out. The double thank-you is the narrow condition in which both are real — the customer is delighted, the company has decided and earned, and the customer, shown the decision, would still mean hers.

Tesco's envelope is a double thank-you. The shopper thanks Tesco for the voucher; Tesco thanks the shopper for the basket that told it which voucher to send. Both are better off, and the shopper, shown the targeting, would shrug and keep the voucher.

Waze is a double thank-you. The driver gets the route; the company gets the map, which every other driver improves. If Waze explained the exchange on the screen, nobody would switch it off.

DBS's nudge is one, on the bank's own description of it. The bank sees your balance and your bills coming, and spends that sight to warn you before a fee it could have collected; it keeps a customer who trusts the app, and the customer keeps the fee. Thank you, and thank you.

FairPrice is the structural case, because it shows a company declining the one-way version on purpose. Singapore's largest grocer holds store-level data on every neighbourhood it trades in. It prices every item the same in every store, the premium format included, while its most profitable rival prices by location and by whoever is next door.³⁵ FairPrice Group's profit from operations runs at about one per cent of revenue; Sheng Siong's net margin is nine to ten per cent, and the eight points between them are, in part, the price of the refusal, paid on purpose, year after year.³⁵ Members are paid a rebate on what they spend. The company could use what it sees to charge one estate more than another for the same bag of rice, and does not; what it takes instead is the share and the trust of a third of the country's grocery spend. That is a loan of trust honoured in the structure of the business, not in a values statement.

Amazon broke the exchange once, in public. In September 2000 it ran a test that showed different shoppers different prices for the same DVDs.³⁶ Customers compared notes online and found out. The company refunded 6,896 of them and its founder wrote a rule that still stands: "we've never tested and we never will test prices based on customer demographics." The company had found something to thank the data for. The moment the customers saw it, they took theirs back.

Delta broke it in eleven days. In September 2023 the airline announced that elite status would henceforth be earned by spending alone — 35,000 dollars a year for the top tier — with caps on the lounges.³⁷ Frequent flyers who had been told for years that they were valued discovered the dollar figure at which they were, and it was higher than most of them spent. Eleven days later the chief executive conceded, "No question we probably went too far," and the thresholds came down.

The science of why the customer's thank-you is so easily withdrawn is the same science as the snow shovel. In 1986 Daniel Kahneman, Jack Knetsch and Richard Thaler asked people whether a hardware store that raised the price of snow shovels from fifteen to twenty dollars the morning after a blizzard was behaving fairly.³⁸ Eighty-two per cent said no. The store had broken no law and told no lie; it had simply used a position it found itself in. The researchers called the rule people were applying dual entitlement: a firm is entitled to its usual margin, and a customer to the usual price, and an advantage the firm happens to hold is not a licence to use it against the customer. Visibility is such an advantage. It is granted, like the blizzard, by circumstance — the customer had to give it to get the service — and the customer's sense of what the firm may do with it is set by the same rule. She granted the margin. She did not grant the shovel price.

So the condition on everything in Section VII is plain. A CX programme has earned its budget when both sides of the counter have reason to say thank you: the customer for the delight, the company for what the delight let it see and decide — and when the customer, shown the company's half of the exchange, would still say hers. One thank-you is a cost centre. The other alone is a scandal waiting for a screenshot. Both together are the only version of customer experience that has ever made money and kept it.

Most CX programmes are designed backwards, and it is nobody's fault in particular. The conversation starts with what the technology enables — the platform demonstration, the shortlist of customer data platforms, the roadmap with the word "AI" in the title — and the delight and the decisions are retrofitted to what the tools can do. The vendor's chart from Section VI is on the second slide. It is a natural order, because the technology is the thing that arrives with a salesperson. It is the wrong order, and the essay has spent its length showing why.

The order that the winners followed has three steps, and the sequence is the point.

First: what delight do we owe this customer? Not what can we collect — what would she thank us for? Hitz's newspaper, the bus-stop panel, the shade that follows her online, the engine that does not surprise, the midwife who answers. Sections I and V are the evidence that this is worth designing seriously and that there is science to design it with.

Second: what decisions, and what revenue, could the visibility enable? A coupon that goes to twenty in a hundred instead of three. A top-up offer sent to the persuadable rather than the at-risk. An engine brought in a week early. An account called before it churns, because usage said so. Sections VI and VII are the evidence that this is where the money has always been.

Only then: what technology serves both? The plumbing comes last, sized to the two answers above. Humby ran the most valuable instrument in British retail on ten per cent of the data because he had decided the questions first. Hitz ran his on index cards. Rolls-Royce's twenty-five sensors are not there because sensors were available; they are there because someone had already decided that the maintenance schedule, the price of the contract and the location of the spare engines would be set by what the sensors saw. A customer data platform chosen before the decisions is a platform that will collect everything and decide nothing, and it will be replaced, under a new name, in five years.

None of this is an argument against the technology. It is an argument about sequence. The same platform, bought third, is an instrument; bought first, it is a hope.

Run in that order, a CX programme produces both thank-yous by design: the delight is chosen for the customer, the decisions are chosen for the business, and the technology is chosen because it serves the two. Run in the usual order, it produces a platform, a satisfaction score and a chart — and a chief financial officer who is still waiting for an answer.

Back to the card. Hitz's guest history recorded the newspaper a man read and whether he paid his bills, on the same piece of paper, in the same drawer. The hotel used the first to delight him and the second to decide about him, and it never pretended the drawer held only one. A century of technology later the drawer is a customer data platform and the card is a profile, and the question has not moved. Would the guest, shown his card, still write the thank-you note? And would the hotel, looking at what the card had earned it, have one to write back?

Thank you for the forty minutes. That is one thank-you, and it is sincerely meant. Whether the essay has earned the other is yours to judge — and if it has not, the line below is the place to say so.