THEJOURNEYMAP

A note before you start

Most journey maps are workshop artifacts: sticky notes, an invented persona and an emotion curve nobody measured. A DTC brand doesn’t need to imagine its customers’ journey. Its order, shipping, email, support and returns data already record it, event by event.

The first number says what to map. McKinsey’s surveys found that how customers rate a whole journey tracks their satisfaction far more closely than how they rate each step Reported. The second says how little it takes to learn why. Abbie Griffin and John Hauser found that twenty interviews surfaced over 90% of the needs that thirty did Published. You need a data pull to find where customers leave and a few weeks of conversations to learn why.

So this guide maps the journey you can measure: stages defined by events, counts and time in each, every touch laid against them, and the backstage behind them. Then each stage gets one owner and one number, and you fix one stretch at a time.

A journey map is a management tool. If it hasn’t changed a decision this quarter, it’s a poster.

This is the hub of the series. Every stage has a deeper book, and this one links to it for the fix instead of teaching it again; the one-page map lists which book covers which stage.

How to read it

Start with The Journey Audit, or with the one-page map below. Or follow a path:

• Founder: The Journey on One Page, A Map Is Not a Poster, One Owner, One Number, Fix One Stretch at a Time.
• Retention or CX lead: Every Touch, in Order, The Backstage, What the Data Can’t See.
• Analyst: Stages Are Events, Where the Cohort Leaks, then For Your Analyst.

Three calculators and a scored audit run in the page. Nothing you type leaves your browser.

What’s proven and what isn’t

• Filed: what companies told the SEC, with the filing named.
• Published: peer-reviewed research, and books and articles by the researchers themselves.
• Reported: what a named company, consulting firm or outlet published. Much of the customer-experience literature comes from consulting firms reporting their own surveys; the text says so each time.
• Derived: my arithmetic from stated inputs, with the inputs shown.

Examples that open with Say or Picture use made-up round numbers. Every source is listed in Appendix C.

The main line runs from the first visit to the third order. The detours are where customers step off it: a contact, a return, a lapse. A standard ecommerce dashboard reports the first two stages and a repeat rate, and very little in between.

Two books sit across every row: The Noise Floor on telling a real change from noise, and Cash Before Growth on what each stage does to cash. Your stages may differ (a subscription brand’s second order is its first renewal), but the shape holds.

Do this

• Write the nine numbers for your store, from whatever you have today. Leave blanks blank. Expect to fill in three or four; first use and the contact detour are the usual blanks.

The audit doesn’t ask whether you have a journey map. It asks whether you could answer, today, three questions: where customers leave, how long they wait before they do, and who is responsible for each stretch. If you can answer the first only up to the checkout, that’s where the audit starts.

Score each check 0 to 2: 0 if it failed or nobody can answer it, 1 if partly true, 2 if clean. “It’s in the dashboard somewhere” isn’t an answer.

A journey you can’t count is a journey you can’t manage. Start by counting it.

The twelve checks

1. Stages are written as events · 5 minLook at: Your list of journey stages, or the map on the wall.
   Good: Six to nine stages, each starting with an event in your data that has a timestamp: order placed, delivered scan, second order.
   Cost if wrong: Nobody can count a stage called “consideration”, so nobody can tell whether it’s improving.
   Read next: Stages Are Events
2. Stage conversion is read by first-order cohort · 5 minLook at: Whether you can show, for customers who first ordered in a given month, the share who reached each later stage.
   Good: A table by cohort month, refreshed monthly.
   Cost if wrong: Calendar-month numbers mix new and old customers, and a growing brand looks better than it is.
   Read next: Where the Cohort Leaks
3. You can name the stall · 3 minLook at: Ask the team which stage loses the most customers each month, and how many.
   Good: One answer, with a number, that matches the cohort table.
   Cost if wrong: Effort goes to the touchpoint people complain about, not the stage that loses the most customers.
   Read next: Where the Cohort Leaks
4. Time in stage is known · 4 minLook at: Median and 75th-percentile days customers spend in each stage before moving on.
   Good: Both numbers for every stage, and a target window for each.
   Cost if wrong: Flows fire on guesses, too early for some customers and too late for most.
   Read next: Where Customers Wait
5. You have a first-use signal · 4 minLook at: Any event that shows a customer started using the product: a how-to page view, a login, a refill scan, a review, a support question about using it.
   Good: One agreed signal, tracked per customer, even if it’s imperfect.
   Cost if wrong: The stretch between delivery and the second order is a black box, and it’s usually the longest one.
   Read next: Stages Are Events
6. Every message you send is on one calendar · 5 minLook at: Transactional emails, flows, campaigns, SMS, shipping notices from your carrier or tracking app, and review requests.
   Good: One list, by stage and day since first order, with the system that sends each one.
   Cost if wrong: Four tools each send “just one” message on the same day.
   Read next: Every Touch, in Order
7. Customer-started touches are on the map · 4 minLook at: Whether tickets, returns and reviews are tagged with the stage the customer was in.
   Good: Contacts per 100 orders by reason and by stage, reviewed monthly.
   Cost if wrong: The loudest signal of a broken stage goes to a helpdesk queue and stays there.
   Read next: Every Touch, in Order
8. No collision days · 4 minLook at: The full message log for ten customers who first ordered last month.
   Good: No day with more than two marketing messages, and no marketing message on the same day as a delay or problem notice.
   Cost if wrong: A promotion lands an hour after a delay notice, and the customer reads both.
   Read next: Every Touch, in Order
9. The backstage is drawn for your worst stretch · 5 minLook at: Whether anyone has drawn who and what system is behind each touch in the stall stage.
   Good: A one-page service blueprint with fail points marked, less than a year old.
   Cost if wrong: Fixes land on the email copy when the failure is a warehouse cutoff.
   Read next: The Backstage
10. Measured and inferred are marked · 3 minLook at: Your map, and whether it says which parts come from data and which from belief.
    Good: Every stage labeled measured or inferred, with a plan to measure the inferred ones.
    Cost if wrong: A guess made in a workshop becomes a fact in a budget meeting.
    Read next: What the Data Can’t See
11. Each stage has one owner and one number · 3 minLook at: Ask who owns first use, and what number they report.
    Good: A name and a number for every stage, reviewed in a monthly meeting.
    Cost if wrong: The stages between teams belong to nobody, and that’s where customers leave.
    Read next: One Owner, One Number
12. Fixes are ranked and measured with a holdout · 5 minLook at: The last three journey changes you shipped, and how you judged them.
    Good: Each was chosen from a ranked list and read against a random holdout.
    Cost if wrong: You ship five changes at once and credit the one you liked.
    Read next: Fix One Stretch at a Time

Score as you go; your band appears when all twelve are in.

Read your score

If you sell on subscription, score check 5 on the first shipment after sign-up and treat the first renewal as your second order.

The usual journey map is made in a day: sticky notes, a persona, an emotion curve with a dip at “delivery” and a peak at “unboxing”. The photo goes into a deck. Six months later nobody can say what it changed.

The workshop map’s real problem

A workshop isn’t wrong in itself. The Nielsen Norman Group, a user-experience research firm, calls a map built from the team’s existing knowledge an assumption map or hypothesis map, and it recommends one as a starting point. Kate Kaplan’s warning is about what happens next: the process “often stops before validation occurs”, and the hypothesis map ends up driving critical decisions Reported. The emotion curve is the clearest case. It’s drawn with confidence, and in most workshop maps nobody measured it.

A DTC brand is unusually well placed to skip that risk. Its systems record the journey as it happens: the order, the carrier’s scans, every email and text, every ticket, return and reorder, each with a timestamp and a customer ID. The map you need is mostly a query.

Journeys, not touchpoints

The case for mapping the whole journey, rather than polishing each step, comes largely from McKinsey. In a 2013 Harvard Business Review article, Alex Rawson, Ewan Duncan and Conor Jones, then McKinsey partners, reported from their firm’s cross-industry surveys that performance on whole journeys was 30% to 40% more strongly correlated with customer satisfaction than performance on single touchpoints, and 20% to 30% more strongly correlated with outcomes such as revenue, repeat purchase and churn Reported. Their example was a pay-TV provider whose onboarding ran about three months, with six or so phone calls and a technician visit. Each interaction was likely to go well, yet in key customer segments average satisfaction “fell almost 40% over the course of the journey” Reported.

This is a consulting firm reporting its own research, without a journal’s scrutiny, so take the direction rather than the percentages: the customer’s verdict forms across the stretch, and the damage often sits between touchpoints, where no team is looking.

Three kinds of map, and the one this book builds

ReportedFirst three rows: Sarah Gibbons, Nielsen Norman Group, “UX Mapping Methods Compared: A Cheat Sheet”, 2017. The last row is this book’s method.

What the research says a journey is

Katherine Lemon and Peter Verhoef’s 2016 review in the Journal of Marketing splits the journey into prepurchase, purchase and postpurchase stages and calls it “iterative and dynamic” Published. They sort touch points into four kinds: brand-owned, partner-owned (your carrier, your returns app), customer-owned, and social or external, such as reviews and friends. And they’re blunt that firms have “much less control, overall, of the customer experience and the customer journey” Published.

That’s the argument for building from data. The parts you don’t control still leave traces: a carrier delay becomes a ticket, an unopened product becomes a missing reorder. A workshop can only guess at those. A query can count them.

Do this

• Find your current map, if there is one. Circle every element that came from data and cross out every one that came from a meeting. What’s left uncircled is your hypothesis map. Keep it: it’s the list of things to measure.

The first decision is the one most maps get wrong: what the stages are. Get it right and the rest of this book is arithmetic.

Four rules for a stage

1. It starts with an event.Something that happens to one customer at one moment and leaves a record: an order, a scan, a login, a ticket. Not a feeling and not a phase.
2. Most repeat buyers pass through it.A stage only a few reach is a detour.
3. Someone can change it.If nothing you control moves it, it’s context.
4. It nests.Count a customer at a stage only if they also passed every stage before it. That keeps the counts falling in order, and it tests your signals: if many customers reorder without ever showing your first-use signal, the signal is too weak.

Six to nine stages is enough.

A standard set, and where each event lives

Appendix A has the query that turns these into one row per customer with a timestamp for each stage.

Finding a first-use signal

First use is the stage that order data can’t see, and it tends to sit in a long silent stretch. You won’t find a perfect signal. You need one that’s good enough, per customer, and cheap to collect. Candidates, by what you sell:

• Anything with instructions: a view of the how-to page or video, or a click on the how-to link in the delivery email.
• Consumables and supplements: a reply to a day-seven check-in (the template is in Appendix B), a scanned QR code on the pack, a refill reminder opted into.
• Beauty and apparel: a review, a photo, a size or shade question after delivery, or the return window passing with no return.
• Anything with an app or account: the first login or setup after delivery.

Then test it: customers with the signal should reorder at a clearly higher rate than those without. If not, it measures something else, such as who opens email. Used, Not Bought covers how to design the first-use stage itself.

Name stages for the customer’s progress

Clayton Christensen and colleagues define a job to be done as “the progress that a person is trying to make in a particular circumstance” Published. That’s a useful test for stage names. “Delivered” is your milestone; the customer’s is “I have it and I know how to start”. Where you count only your milestone, add a signal for theirs.

Do this

• Write your stages as a list of events, each with the table and field it comes from. Pick one first-use signal and test it against reorder rate this week. If you can’t find one, the check-in email in Appendix B creates one.

The core of the map is one table: for customers who first ordered in a given month, how many reached each stage.

Why cohorts, not calendar months

A calendar-month report mixes customers at every point in their journey: acquire twice as many in March and April’s second orders rise even if nothing improved. A first-order cohort follows one group from the same starting line. The Second Order covers cohorts in depth; here you need one table.

Give each stage a fixed window, measured from the first order, so cohorts compare fairly: delivered by day 14, first use by day 30, second order by day 120, third order by month 12. A cohort can only report a stage once its window has closed.

A worked example

Say a brand’s March cohort has 5,000 first-time customers. 4,900 are delivered (the rest canceled or were refunded before delivery). 3,000 show the first-use signal by day 30. 1,050 place a second order by day 120, and 600 a third by month 12.

Rank by customers lost. The step to the third order loses 43% of those who reach it, alarming in a rate report, but only 450 people. The steps into first use and the second order lose 1,900 and 1,950. With two stages this close, the next section breaks the tie.

Customers, contribution and the value of a point

Not every lost customer is worth the same. One lost before first use still had two stages to pass before a third order; one lost at the second order had one. A point gained early reaches more customers, but fewer of them go on. In this example, a 5-point gain at the second order is worth about $123,000 a year and the same gain at first use about $71,000 Derived, both computed below.

There’s a simple rule inside that arithmetic: a fixed gain in points is worth most at the stage with the lowest conversion rate, because it’s the largest relative lift there. Chapter 11 adds cost and confidence.

With the defaults, the stall is the second order: 1,950 customers a month stop there, with up to about $134,000 of contribution at stake per cohort. A 5-point gain at the second order is worth about $123,000 a year; first use loses nearly as many customers but the same gain there is worth about $71,000 Derived.

Read the leak carefully

• Look at three cohorts, not one. If the same stage leads in three in a row, it’s the stall. The Noise Floor shows how big a month-to-month change has to be before it means anything.
• Split by how the first order was acquired. A cohort bought on a 40%-off offer may leak at the second order for reasons no journey fix will cure; The First Offer covers that.

Do this

• Build the cohort table for the last three cohorts whose third-order window has closed, using the query in Appendix A. Put the counts into the tool. Write down the stall, and the runner-up if it’s within 10%.

Two stages can have the same conversion rate and very different journeys. In one, customers move on in a tight window. In the other, half move quickly and the rest drift for weeks, and that’s where flows fire at the wrong moment.

Median and 75th percentile, not the average

For each stage, take the customers who moved on to the next one and measure how many days it took. Report two numbers: the median (half were faster) and the 75th percentile (three in four were faster). Skip the average, which a few very slow customers drag far from typical. The gap between the two is the stage’s tail.

The target window

Then set a target for each stage: the day by which a customer should have moved on if the journey is working. It can come from a promise (delivery in five days), the product (a 30-day supply) or judgment (first use within a week). Write it down before you look at the data.

The stall in time is the stage where the 75th percentile runs furthest past the target. That’s where the slow quarter of your movers, and everyone behind them who never moves, wait longest after they should have acted.

DerivedThe number on the right is the 75th percentile divided by the target. These are the defaults in the tool below.

In days, the reorders take longest. Against target, first use is worse: the slow quarter take 2.6 times as long as they should, in a stage that often gets no useful message at all.

Silence means different things for different customers

The repeat-purchase models marketers use to predict which customers are still active, such as Peter Fader, Bruce Hardie and Ka Lok Lee’s BG/NBD model, need only two facts per customer: recency, when they last bought, and frequency, how often Published. In those models, thirty quiet days means more for a customer who ordered every two weeks than for one who orders twice a year. Judge time in stage the same way: against what’s normal for that stage and that kind of customer. The Second Order covers timing the reorder reminder; this chapter is about finding which stage to time first.

With the defaults, first use is the stall in time: its 75th percentile is 2.6 times its target, and the slow quarter waits 11 days past the 7-day target, with a tail of 3.0 (the 75th percentile is three times the median). The median is on time, so this is a fast group and a slow group, not a bad target.

Do this

• Write a target for each stage before you pull the numbers. Then pull median and 75th-percentile days for the last two closed cohorts (Appendix A), and run the tool. If the stall in time differs from the stall in customers, you have two candidates; chapter 11 ranks them.

Ask five people how many messages a new customer gets in their first two weeks and you’ll get five answers. Messages come from the store platform, the email and SMS tool, the tracking app, the review app and the helpdesk, each set up by someone thinking about their own. Only the customer sees the whole sequence.

Two lists

List what you send: order and shipping notices, flows, campaigns, SMS, tracking-app notices, review requests, subscription reminders. Then what customers start: tickets, chats, replies, return requests, reviews, cancellations. In Lemon and Verhoef’s terms, the first list is mostly brand-owned and partner-owned; the second is where customer-owned touch points leave a trace Published.

For each touch, record its stage, the day it fires (from the first order), the trigger and the system that sends it.

Lay it on one timeline

Then pull the actual message log for a handful of real first-time customers and lay it out by day. Picture a customer who orders on a Friday, during a sale week, and whose parcel is delayed:

Made-up example. Day 0 is the first order. Day 3: a delay notice, a flow email, a sale campaign and a sale SMS. Delivery on day 6; review request on day 7, before most customers have used the product. Days 8 to 14: nothing.

Two patterns stand out.

Collision days

A collision is a day when messages from different systems pile up, or when a marketing message lands next to a problem. The delay notice on day 3 is the worst kind: the customer learns their order is late and, an hour later, gets a text urging them to buy more. Each system behaved correctly. Nobody owned the day.

Three rules prevent most collisions:

• A problem suppresses marketing. A delay, a failed payment, an open ticket or a return in progress pauses campaigns and sales SMS for that customer for 48 hours, or until it’s resolved.
• A daily cap across systems. No more than two marketing messages a day per customer, counting email and SMS together.
• One calendar owner. One person approves every new automated message, from any tool, against the timeline.

The Whole Machine covers flows, frequency and deliverability. The point here is narrower: a collision is a journey problem, and you only see it when every system is on one timeline.

Silent stretches

The opposite problem sits at days 8 to 14: after the review request, nothing, until a replenishment email weeks later. That’s the stretch between delivery and first use, the stage chapter 5 found running furthest past its target. Silence isn’t always wrong. But it’s usually where the next question goes unanswered.

The Corporate Executive Board’s contact-center research, published in HBR in 2010, is useful here. Matthew Dixon, Karen Freeman and Nicholas Toman reported that 22% of repeat calls involved “downstream issues related to the problem that prompted the original call”, and that “fifty-seven percent of inbound calls came from customers who went to the website first” Reported. That’s the firm’s own research, from service centers, but the lesson carries: the next question is predictable. In a DTC journey it’s how to start, what to expect in week one, and when to reorder. Answer it before it’s asked.

What customers start is a stage signal

Tag every ticket with the stage the customer was in when they wrote, which most helpdesks can do from order status, and report contacts per 100 orders by reason and stage. A rise in how-do-I questions from customers without a first-use signal tells you more about that stage than any survey. Where-is-my-order contacts belong to The Kept Promise; returns to The Return Trip.

Do this

• Export the full message log for ten customers who first ordered last month, from every tool, and lay it out by day as above. Circle every day with more than two marketing messages or any marketing next to a problem notice. Mark every stretch of seven days or more with nothing useful in it.

Failures usually start where the customer can’t see: a warehouse cutoff, a tracking app that doesn’t talk to the email tool, a helpdesk that can’t see order status. The service blueprint is the tool for that half of the map.

Shostack’s blueprint

G. Lynn Shostack, then a senior vice president at Bankers Trust, introduced the service blueprint in Harvard Business Review in 1984. Her method has four steps: identify the processes, isolate the fail points, set a time frame for each step, and analyze profitability. The blueprint separates what the customer sees from what happens behind it with a “line of visibility”, and marks each place the service is likely to go wrong with an F Published. Her example was a shoeshine: a standard execution time of two minutes, and a customer who would tolerate up to five before thinking less of the service Published. That’s a target window, the same idea as chapter 5.

Mary Jo Bitner, Amy Ostrom and Felicia Morgan turned it into a practical technique in California Management Review in 2008. Their blueprint has five layers: physical evidence, customer actions, onstage contact employee actions, backstage contact employee actions and support processes, split by a line of interaction, a line of visibility and a line of internal interaction Published.

Their cases include ARAMARK Parks and Resorts, which blueprinted its Lake Powell houseboat vacations and redesigned around the gaps. The authors report that ARAMARK saw “50 percent fewer complaints” and repeat business up 12 percent afterward Reported. That’s one company’s before-and-after, not a controlled test, so read it as what’s possible, not what to expect.

When a touch fails, look below the line of visibility first. The email is rarely the problem. The system that was supposed to trigger it usually is.

A DTC blueprint for one stretch

For a DTC brand the “employees” are mostly systems. Here’s the stretch from order to delivery, with fail points marked:

Every fail point in that table sits at a handoff between two systems or two teams. That’s the typical pattern, and it’s why a stage needs an owner who can see both sides of every handoff (chapter 10). The Kept Promise covers fixing the delivery stretch itself.

Blueprint the stall, not everything

Don’t blueprint the whole journey. Blueprint the stall from chapter 4, and the stall in time from chapter 5 if it differs, one page each. For first use the backstage is often thin: a how-to email written at launch and no signal feeding back. That’s a finding too.

Do this

• Draw the blueprint for your stall with whoever runs operations, retention and support, in one hour. Five columns as above. Mark every fail point with an F, and for each one name the system and the person who can fix it.

The cohort table can tell you that 1,900 customers a month never show a first-use signal. It can’t tell you whether the product was confusing, was a gift, or was used without triggering your signal. Those are different fixes, and the only way to tell them apart is to ask.

Four sources, and what each misses

Use the first three to form a view, and interviews to test it. For reading tickets and return reasons as a defect log, see The Return Trip. For session recordings and on-site research, see The Honest Test.

How many conversations

Fewer than most teams think. Abbie Griffin and John Hauser, studying how firms identify customer needs, found that “interviewing 20 customers identifies over 90% of the needs provided by 30 customers”, and hypothesized that 20 to 30 interviews find 90% to 95% of needs. Two one-on-one interviews were about as productive as one focus group Published. For website problems, Jakob Nielsen’s rule of thumb is that five users surface about 85% of usability issues, with more needed for distinct groups Reported.

For a journey stage, that suggests eight to ten interviews with customers who stalled and five or so with customers who passed, per round. The comparison is what makes it useful: the difference between the two groups is the fix.

Who to call, and what to ask

• Recruit from the stage, within two weeks, while it’s fresh: say, customers delivered 10 to 20 days ago with no first-use signal.
• Ask about the circumstance, not the brand. What were you trying to get done when you ordered? What happened when it arrived? You want the progress they were after and what got in the way.

The script is in Appendix B. Offer a small thank-you, and be clear it doesn’t depend on what they say.

Label every part of the map

Mark each stage, and each claim about why customers stall there, as measured (from data), heard (from customers in their words) or inferred (believed, not yet checked). The Nielsen Norman Group’s advice on hypothesis maps is to validate them with research before using them for decisions Reported. The label keeps that visible. An inferred claim can still guide a test; it shouldn’t drive a budget.

Do this

• Book ten calls with customers from your stall and five with customers who passed it, this month. Before the first call, write down what you think the problem is. After the last one, compare.

Two findings sit behind most emotion curves: people remember experiences by the most intense moment and the end, and customers reward low effort more than delight. Both are real. Both weaken when stretched across a four-month journey.

What the peak-end research found

In a 1993 experiment, Daniel Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier had people hold a hand in 14°C water for 60 seconds and, in another trial, 30 seconds longer as the water warmed to 15°C. “A significant majority chose to repeat the long trial”, preferring more discomfort that ended better Published. Redelmeier and Kahneman then found that 154 colonoscopy and 133 lithotripsy patients judged total pain by the worst moment and the last three minutes, and that long procedures were not remembered as especially bad Published.

The strongest test was a randomized trial. Redelmeier, Katz and Kahneman added a short, less painful interval to the end of colonoscopies for half of 682 patients. They remembered the procedure as less unpleasant, and were slightly more likely to come back for a repeat colonoscopy over the following years: an odds ratio of 1.41, against an average return rate of 50.4% Published. That’s the result closest to what a brand cares about, a changed ending that changed a later decision, and the effect was modest.

A 2022 meta-analysis by Balca Alaybek and colleagues, pooling 174 effect sizes, found a large peak-end effect on how people rate past experiences (a correlation of 0.58) and almost no effect of duration. It also found the peak-end effect was about as strong as the simple average of the whole experience Published. So the rule holds up, but the average matters as much: you can’t fix a bad journey with a good ending.

How far it travels

Most of that research is on single, short episodes rated moment by moment. Stretched episodes look different. Simon Kemp, Christopher Burt and Laura Furneaux had 49 students text their happiness daily through a week-long vacation and found “the peak-end rule was not an outstandingly good predictor” of how they remembered it Published. In a call-center study, Peter Verhoef, Gerrit Antonides and Arnoud de Hoog found that average performance and positive peaks drove satisfaction, while the end of the call had an unexpected negative effect they called “difficult to explain” Published.

A DTC journey is many episodes over months. My reading: treat each detour (a delivery problem, a support contact, a return) as its own episode and make sure it ends well. Don’t draw one emotion curve over the whole journey and call it measured.

Effort

The Corporate Executive Board’s research, from more than 75,000 people who’d used phone or self-service support, argued that exceeding expectations adds little and that reducing effort is what keeps customers. Among customers who had a hard time solving a problem, 81% said they intended to speak negatively about the company Reported. The authors proposed a Customer Effort Score as a better predictor of loyalty than satisfaction or Net Promoter Score.

An independent test disagrees. Evert de Haan, Peter Verhoef and Thorsten Wiesel compared the metrics across customers of 93 firms in 18 industries: top-2-box satisfaction predicted retention best overall, the best metric differed by industry, and combining metrics beat any one Published. Measure effort where it’s obviously at risk: the detours.

What to put on the map instead of a curve

Put measured points where the curve was: a one-question survey after a support contact, review stars at the first-use stage, return reasons. Add effort markers from data, such as a second contact about the same order within seven days. And make every detour end in writing: the refund confirmed, the replacement tracked, the answer sent. The Proof File covers when to ask for reviews.

Do this

• Replace the emotion curve on your map with three measured points and the repeat-contact rate. If you keep a curve, label it inferred.

Ask who is responsible for the stall and each fail point you’ve found. The first order has an owner, because acquisition is measured on it, and the second order usually does too. Delivered and first use sit between operations, support and retention, and often belong to none of them.

Journeys as products

David Edelman and Marc Singer, writing in Harvard Business Review in 2015, argued that companies should manage journeys the way software firms manage products, with a journey product manager who leads a cross-functional team and is accountable for the journey’s return on investment Reported. A $10 million brand doesn’t need a new title. It needs the same idea at its own scale: one named person per stage, measured on that stage’s number, with the standing to ask other teams for changes.

Stages between teams belong to nobody, and that’s where customers leave.

The stage table

The guardrail stops the number being gamed: second orders are easy to buy with a deep discount, and contacts easy to cut by hiding the contact form. Report both, every month.

One person can own several stages. What matters is one name per stage, and that each owner also owns the handoff into the next stage, where the fail points from chapter 7 sit.

The monthly map review

A map nobody reviews is a poster again within a quarter. Hold a 45-minute review each month with every stage owner in the room, on a fixed agenda: the newest closed cohort against the three before it, time in stage against target, every automated message added or changed since last month, each live fix and its holdout, and the next fix from the ranked list. The agenda is in Appendix B. The rule that makes it work: every number has a name next to it, and every fix has a holdout.

Do this

• Fill in the stage table with names, today. Where two names compete for a stage, pick one. Where no name fits, that stage is your likeliest stall; check it against chapter 4. Book the first review for the first week of next month.

By now you’ll have more fixes than time, and every one has an advocate. The ranking turns the argument into arithmetic, and the holdout tells you afterward whether the arithmetic was right.

The ranking

For each candidate fix, estimate five things:

• Reach: customers entering the stage each month.
• Lift: the change in the stage’s conversion you expect, in points.
• Value: what one extra customer passing the stage is worth. From the chapter 4 defaults, an extra customer reaching first use is worth about $24 (20% of them go on to a third order worth $120) and one reaching the second order about $69 Derived.
• Confidence: how likely the fix is to work at all.
• Cost: everything for the first year: build, tools, discounts, time.

Expected value for a year is reach × 12 × lift × value × confidence. Divide by cost and rank. Because it’s in money, not points, it can be checked against what happened.

Setting confidence on a fixed scale

Confidence is where rankings get rigged, because the advocate sets it. Use a fixed scale: 80% for a fix that already worked in a holdout elsewhere in your own business; 50% for one with good outside evidence; 20% to 30% for a reasonable hunch. Never go above 80%. If one fix only ranks first at 90%, it doesn’t rank first.

With the defaults, the delivery-day how-to email ranks first at 6.4 times its cost, about $25,400 of expected value in its first year. The reorder reminder is close behind at 6.2 times, and it has the larger net value ($31,260 against $21,402), so the ranking can’t separate them. The offer returns 3.6 times. The faster carrier returns less than a fifth of its cost in the first year, because a one-point gain at a stage that already passes 98% of customers is worth little further down the line Derived.

One stretch, with a holdout

Ship one fix per stage at a time, and keep a random share of customers on the old journey. Two fixes in one stage can’t be told apart, and a fix read against last month mostly measures the calendar. The Honest Test covers holdouts and reading them; The Noise Floor covers how big a change has to be before it’s real.

Size it first. To see first-use share move from 61% to 64%, you need roughly 4,200 customers in each group Derived, from the rule of thumb 16 × p(1 − p) ÷ d² at 80% power and 5% significance. With 4,900 delivered customers a month split evenly, both groups fill in under two months; a 10% holdout would take more than eight. For small effects, split evenly.

Do this

• List every fix anyone has proposed for your stall and put the top four into the tool, with confidence set on the fixed scale. Ship the first one with a holdout sized as above, and write down the result that would make you stop.

Companies rarely publish their journey maps. They do publish the decisions a map leads to. Both cases here are far larger than the brands this guide is written for; what transfers is the method, not the scale.

Warby Parker retires Home Try-On

For most of its life Warby Parker’s best-known stage sat between the first visit and the first order. Its 10-K for fiscal 2024 describes it: customers pick five frames on the site and try them at home “for five full days”, free Filed. Home Try-On answered the hardest question in buying glasses online, whether they’d suit you, by adding a stage to the journey: frames arrive, you try them, you send them back and order.

On August 7, 2025, in its second-quarter earnings release, the company disclosed its decision to sunset the program at the end of the year Filed. Retail Dive reported the reason executives gave on the earnings call: most recent home try-on users lived within 30 minutes of one of its roughly 300 stores Reported. Co-founder and co-CEO Neil Blumenthal called Home Try-On “a novel way to help customers shop for glasses online” Reported. The same 10-K describes the alternatives, a Virtual Try-On tool and an online quiz, and says customers who shop across product lines and channels “tend to convert to highly loyal returning customers” Filed.

Read as a journey decision: a stage designed for customers with no store nearby was now used mostly by customers who had one. Data about who was in the stage, not a workshop, redrew the map. Ask the same of your oldest stages: the sample program, the quiz, the welcome series. Who uses each one now?

Chewy designs the reorder

Chewy treats the second order as a stage to design, not a hope. Its 10-K describes Autoship as “customizable and convenient automatic reordering” Filed. In its fiscal 2025 results, Chewy reported that sales to Autoship customers reached $10.5 billion, 83.3% of net sales, up from 79.2% the year before, with 21.3 million active customers spending $591 each on average Reported.

The backstage is part of the design. Chewy’s 10-K for the year ended February 1, 2026 says it operates “customer service centers 24/7”, and describes surprising customers “with a hand-painted pet portrait” and offering “personalized expressions of sympathy” when a pet dies Filed. In this book’s terms, those are good endings to detours: a contact that ends with an answer at any hour, and a loss that ends with care rather than a cancellation form.

Copy the two decisions, not the scale: make the reorder a designed stage with its own owner (The Standing Order and The Free Choice cover how), and staff the backstage behind the detours as part of the journey.

What the two have in common

Neither decision is a better email. Both change what customers do between two events, both rest on a count of who is actually in the stage, and both needed an owner with the authority to change the backstage. That’s why chapter 10 puts one name on each stage.

Do this

• List every stage-level program you run (samples, quizzes, trials, subscriptions, loyalty) with the year it launched. For each, pull who used it in the last 90 days. Flag any program whose users no longer look like the customers it was built for.

Most dashboards report revenue, orders and a repeat rate, and none tells you which stage is failing. The scorecard follows the journey, so a problem shows up where it starts, next to its owner’s name.

Quarterly, add the share of the map that’s measured rather than inferred (chapter 8) and a check that every stage still has an owner.

Read the scorecard as a journey

The numbers affect one another. A cheaper carrier moves number 3 at the delivered stage and number 6 a week later. A new review request moves number 8 and, if it lands before first use, number 4. When one number moves, look one stage upstream before you look at the stage itself. The Noise Floor covers control limits for deciding when a monthly change is worth a meeting.

When a stage number moves, look one stage upstream first. The cause is usually where the customer was a week earlier.

Do this

• Build the ten numbers for last month and put them on one page with the owner’s name beside each. Fill what you can; for each blank, write who owns the data and when it will exist. Review it at the monthly map review in chapter 10.

Most of this guide takes a month and no new software. The order matters: count first, then look at every touch, then listen, then fix one thing. Skipping to the fix is how brands end up with a new flow for a stage they never measured.

1. Week 1 · CountRun the audit in chapter 1. Write your stages as events and pick a first-use signal (chapter 3). Pull the cohort table for the last three closed cohorts and find the stall (chapter 4). Set a target for each stage and pull time in stage (chapter 5).
2. Week 2 · Every touchExport the message log for ten new customers from every system and lay it out by day. Put the three collision rules in place (chapter 6). Start tagging tickets by stage. Blueprint the stall with operations, support and retention in one room (chapter 7).
3. Week 3 · Listen and assignHold ten interviews with customers from the stall and five with customers who passed it (chapter 8). Label every part of the map measured, heard or inferred. Replace the emotion curve with measured points (chapter 9). Fill in the stage table with one name per stage and book the monthly review (chapter 10).
4. Week 4 · Rank and ship onePut every proposed fix for the stall into the ranker, with confidence on the fixed scale (chapter 11). Size the holdout and ship the top fix. Build the scorecard (chapter 13) and hold the first monthly review on day thirty.

Count the journey, see every touch, hear the customers who stalled, then fix one stretch and prove it.

At day thirty you should have a cohort table, the stall named with a number, a timeline with no collision days, a blueprint, fifteen interviews, an owner for every stage and one fix running against a holdout. The map won’t be finished. It’s meant to be reviewed next month, with new numbers.

Do this

• Put the four weeks on the calendar today, with one owner for each week’s list, and book the first monthly map review for day thirty.

Whoever takes on the journey needs six things on day one.

1. Order and fulfillment dataEighteen months of orders with customer ID, discount and delivery timestamps, so cohorts can close their twelve-month windows.
2. Every message logSend logs, with customer ID and timestamp, from every tool that messages customers, plus admin access to set the collision rules.
3. The helpdesk and returns exportsTickets with reason, timestamp and order; returns with reason and outcome.
4. Contribution per orderFrom finance, so a stage can be valued and a fix ranked.
5. The current map, if anyWith the date it was made and who made it, to mark measured and inferred.
6. A list of stage ownersEven if half the names are blank. The blanks are the first finding.

Do this

• Ask for all six in one message today. Whatever comes back slowest is where the business knows least about its customers’ journey.

• Understanding Customer Experience Throughout the Customer JourneyKatherine Lemon and Peter Verhoef, Journal of Marketing, 2016The research view of the journey: its stages, the four kinds of touch point, and how little of it a firm controls. Start here.
• Service Blueprinting: A Practical Technique for Service InnovationMary Jo Bitner, Amy Ostrom and Felicia Morgan, California Management Review, 2008How to draw the backstage, with worked company cases. The most useful single paper for chapter 7.
• Designing Services That DeliverG. Lynn Shostack, Harvard Business Review, 1984The original blueprint. Read it for fail points and for the idea that every step has a time standard.
• The Truth About Customer ExperienceAlex Rawson, Ewan Duncan and Conor Jones, Harvard Business Review, 2013The consulting case for managing journeys rather than touchpoints. Take the direction, not the percentages.
• The Voice of the CustomerAbbie Griffin and John Hauser, Marketing Science, 1993How many interviews it takes to hear most of what customers need, and why one-on-ones work as well as groups.
• Competing Against LuckClayton Christensen, Taddy Hall, Karen Dillon and David Duncan, 2016Jobs to be done at book length. Use it to name stages for the customer’s progress and to run better interviews.
• The Effortless ExperienceMatthew Dixon, Nick Toman and Rick DeLisi, 2013The effort argument in full, from the firm that ran the research. Read it alongside de Haan, Verhoef and Wiesel (2015) for the counterpoint.

Full references, including the research on memory in chapter 9, are in Appendix C.

Andrew Lauchner is a growth and retention operator for consumer brands and the author of sixteen free books on the method. He fixes the second-purchase problem through his practice, Growth Legend, as one engagement: The Second-Purchase Sprint. He is the author of The Second Order, on turning first-time buyers into second-time buyers, and The Whole Machine, on the fundamentals of DTC growth, along with a series of field guides for DTC operators at andrewlauchner.com.

As Senior Director of Growth and Retention Marketing at Gallery Furniture, he rebuilt the customer journey and the sales playbooks together. He has worked on growth and retention at Binance and 3Commas, and has been Head of Growth and Retention at Greatness Wins and at Nexus Agriscience.

Andrew answers every note from operators working on this. Write to andrew@growthlegend.com or message him on LinkedIn.

The formulas

One row per customer, one timestamp per stage

-- orders: order_id, customer_id, created_at, cancelled_at,
--         source ('web','exchange','replacement', ...)
-- fulfillments: order_id, shipped_at, delivered_at
-- first_use: customer_id, signal_at  (your chosen signal: how-to clicks,
--            check-in replies, logins, QR scans, reviews)
CREATE VIEW customer_stages AS
WITH paid AS (
  SELECT order_id, customer_id, created_at,
         ROW_NUMBER() OVER (PARTITION BY customer_id
                            ORDER BY created_at) AS n
  FROM orders
  WHERE cancelled_at IS NULL
    AND source NOT IN ('exchange','replacement')
),
firsts AS (
  SELECT customer_id, order_id, created_at AS first_at
  FROM paid WHERE n = 1
)
SELECT f.customer_id,
       date_trunc('month', f.first_at)                  AS cohort,
       f.first_at,
       (SELECT MIN(fl.delivered_at) FROM fulfillments fl
         WHERE fl.order_id = f.order_id)                AS delivered_at,
       (SELECT MIN(u.signal_at) FROM first_use u
         WHERE u.customer_id = f.customer_id
           AND u.signal_at >= f.first_at)               AS first_use_at,
       (SELECT p.created_at FROM paid p
         WHERE p.customer_id = f.customer_id AND p.n = 2) AS second_at,
       (SELECT p.created_at FROM paid p
         WHERE p.customer_id = f.customer_id AND p.n = 3) AS third_at
FROM firsts f;

Exclude exchange and replacement orders, or every exchange looks like a second order. If your returns app creates orders with a normal source, flag them by tag or by a zero-value total.

The cohort table, nested

-- a customer counts at a stage only if they passed every earlier stage
WITH s AS (
  SELECT cohort,
    delivered_at <= first_at + INTERVAL '14 days'    AS d,
    first_use_at <= first_at + INTERVAL '30 days'    AS u,
    second_at    <= first_at + INTERVAL '120 days'   AS o2,
    third_at     <= first_at + INTERVAL '12 months'  AS o3
  FROM customer_stages
  WHERE first_at < CURRENT_DATE - INTERVAL '12 months'  -- windows closed
)
SELECT cohort,
       COUNT(*)                                            AS first_orders,
       COUNT(*) FILTER (WHERE d)                           AS delivered,
       COUNT(*) FILTER (WHERE d AND u)                     AS first_use,
       COUNT(*) FILTER (WHERE d AND u AND o2)              AS second_order,
       COUNT(*) FILTER (WHERE d AND u AND o2 AND o3)       AS third_order,
       COUNT(*) FILTER (WHERE o2)                          AS second_any
FROM s
GROUP BY cohort
ORDER BY cohort;

Comparisons with a null timestamp come back null, which the filters treat as false. The last column tests your first-use signal: if second_any is much larger than second_order, many customers reorder without the signal, and the signal is too weak to use as a stage.

Time in stage

-- median and 75th-percentile days, among customers who moved on
SELECT cohort,
  percentile_cont(0.5) WITHIN GROUP (ORDER BY
    EXTRACT(EPOCH FROM first_use_at - delivered_at) / 86400)
    FILTER (WHERE first_use_at >= delivered_at)       AS use_median,
  percentile_cont(0.75) WITHIN GROUP (ORDER BY
    EXTRACT(EPOCH FROM first_use_at - delivered_at) / 86400)
    FILTER (WHERE first_use_at >= delivered_at)       AS use_p75,
  COUNT(*) FILTER (WHERE first_use_at >= delivered_at) AS timed
FROM customer_stages
GROUP BY cohort
ORDER BY cohort;

Repeat the pattern for each pair of stages.

Collision days

-- email_events: customer_id, sent_at, channel ('email','sms'),
--   kind ('transactional','flow','campaign','review','problem')
--   Union in SMS, tracking-app and review-app logs before running.
WITH m AS (
  SELECT e.customer_id,
         (e.sent_at::date - cs.first_at::date) AS day,
         e.kind
  FROM email_events e
  JOIN customer_stages cs USING (customer_id)
  WHERE e.sent_at >= cs.first_at
    AND e.sent_at <  cs.first_at + INTERVAL '30 days'
    AND cs.first_at >= CURRENT_DATE - INTERVAL '90 days'
),
d AS (
  SELECT customer_id, day,
         COUNT(*) FILTER (WHERE kind IN ('flow','campaign')) AS marketing,
         COUNT(*) FILTER (WHERE kind = 'problem')            AS problems
  FROM m
  GROUP BY customer_id, day
)
SELECT COUNT(DISTINCT customer_id)
         FILTER (WHERE marketing >= 3
                    OR (problems > 0 AND marketing > 0)) * 1.0
       / COUNT(DISTINCT customer_id)        AS share_with_a_collision_day
FROM d;

Tag delay, failed-payment and open-ticket notices as problem. Customers who received no messages drop out of the denominator; count them separately, because they’re your silent stretch.

Contacts by stage

-- support_tickets: ticket_id, customer_id, order_id, created_at, reason
SELECT CASE
         WHEN cs.delivered_at IS NULL OR t.created_at < cs.delivered_at
           THEN '1 before delivery'
         WHEN cs.first_use_at IS NULL OR t.created_at < cs.first_use_at
           THEN '2 delivered, no first use'
         WHEN cs.second_at IS NULL OR t.created_at < cs.second_at
           THEN '3 after first use'
         ELSE '4 after second order'
       END                AS stage,
       t.reason,
       COUNT(*)           AS tickets
FROM support_tickets t
JOIN customer_stages cs USING (customer_id)
WHERE t.created_at >= CURRENT_DATE - INTERVAL '90 days'
GROUP BY 1, 2
ORDER BY 1, 3 DESC;

Divide by orders in the same 90 days, times 100, for contacts per 100 orders. For repeat contacts, count tickets on the same order within seven days of an earlier one.

Stage owner brief

STAGE          [First use]
STARTS WHEN    [Carrier's delivered scan]
ENDS WHEN      [First-use signal: how-to click, check-in reply or review]
WINDOW         [30 days from first order]
TARGET         [Median within 7 days of delivery]
OWNER          [Name]
THE NUMBER     [Share of delivered customers with the signal by day 30]
GUARDRAIL      [Contacts per 100 orders at this stage]
TODAY          [61%] of the [March] cohort; median [6] days, p75 [18]
TOUCHES        [List every message and page in this stage, by day,
                with the system that sends it]
FAIL POINTS    [From the blueprint: each F, its system, who can fix it]
MEASURED /     [What we know from data / heard from customers /
HEARD /         still believe without checking]
INFERRED
LIVE FIX       [Name, start date, holdout share, decision date]
NEXT FIX       [Top of the ranked list]

Interview script (30 minutes)

BEFORE     Recruit from the stage, within two weeks of the moment:
           e.g. delivered 10 to 20 days ago, no first-use signal.
           Offer a thank-you that doesn't depend on what they say.
           Write down what you think the problem is.

OPEN       Thanks for making time. There are no wrong answers, and
           nothing you say will hurt our feelings. I'm trying to
           understand what happened, not to sell you anything.
           May I take notes?

BEFORE     Take me back to when you ordered. What was going on?
THE ORDER  What were you hoping it would do for you?

ARRIVAL    Walk me through the day it arrived. Where were you?
           What did you do with it first?
           Did anything from us arrive around then? What did you
           make of it?

FIRST USE  When did you first use it, if you have? What happened?
           If not yet: what's in the way?

SINCE      What would make you order it again? What would stop you?

CLOSE      Anything I should have asked? Thank you. Here's how
           you'll get the thank-you: [how].

AFTER      Within the hour: three quotes, the circumstance, the
           progress they wanted, what got in the way. Tag each
           point measured, heard or inferred.

Monthly map review (45 minutes)

ATTENDEES   Every stage owner. One chair. One note-taker.
PRE-READ    The scorecard (chapter 13), sent two days ahead.

10 MIN      Cohort table: newest closed cohort vs the three before.
            Has the stall moved? Customers lost there this month?
 5 MIN      Time in stage vs target. Any tail over 2x not yet split?
 5 MIN      Touches: every new or changed automated message, from
            any system. Collision days found. Longest silence.
15 MIN      Fixes in flight: each with its holdout read so far.
            Decision for each: stop, continue or roll out.
10 MIN      Next: top of the ranked list, owner, start date,
            holdout size, the result that would make us stop.

OUTPUT      Decisions and owners, sent the same day.

Email: first-week check-in

SUBJECT      How's It Going With Your [Product]?
PREVIEW      one tip for your first week, and a quick question

Hi [first name],

Your [product] arrived [on Tuesday]. Most people get the best
results when they [one specific tip]. Here's a 60-second guide:
[link to how-to]

Quick question: have you tried it yet? Just reply with
1 if yes, 2 if not yet, or 3 if something isn't right.
A person reads every reply, and a 3 comes straight to our team.

[Name], [Brand]

[Brand], [postal address]
You're receiving this because you ordered from us and agreed
to hear from us by email. Unsubscribe: [link]

Send it at your first-use target, counted from the delivered scan, and only to customers without a first-use signal yet. The click and the reply both become the signal. Pause it for anyone with an open ticket, a delay or a return in progress.

SMS: first-week check-in

[Brand]: Hi [first name], your [product] arrived [Tuesday].
Tried it yet? Reply 1 for yes, 2 for not yet, or 3 if
something's not right and we'll help. Quick start: [link]
Reply STOP to opt out

Send only to customers who gave SMS consent, within quiet hours for their time zone, and never on the same day as the email. Consent and quiet-hour rules differ by state; have counsel review your SMS program.

Front and chapter 2: journeys and maps

• Rawson, Alex, Ewan Duncan and Conor Jones. “The Truth About Customer Experience,” Harvard Business Review, September 2013. McKinsey’s own survey research.
• Griffin, Abbie and John R. Hauser. “The Voice of the Customer,” Marketing Science 12(1), 1993, 1–27.
• Kaplan, Kate. “Approaches to Journey Mapping: 2 Critical Decisions to Make Before You Begin,” Nielsen Norman Group, June 14, 2020.
• Gibbons, Sarah. “UX Mapping Methods Compared: A Cheat Sheet,” Nielsen Norman Group, November 5, 2017; “Journey Mapping 101,” December 9, 2018.
• Lemon, Katherine N. and Peter C. Verhoef. “Understanding Customer Experience Throughout the Customer Journey,” Journal of Marketing 80(6), 2016, 69–96.

Chapters 3 to 5: stages, cohorts and time

• Christensen, Clayton M., Taddy Hall, Karen Dillon and David S. Duncan. “Know Your Customers’ ‘Jobs to Be Done’,” Harvard Business Review, September 2016; and Competing Against Luck, Harper Business, 2016, source of the job definition quoted in chapter 3.
• Fader, Peter S., Bruce G. S. Hardie and Ka Lok Lee. “Counting Your Customers the Easy Way: An Alternative to the Pareto/NBD Model,” Marketing Science 24(2), 2005, 275–284.

Chapters 6 and 7: touches and the backstage

• Dixon, Matthew, Karen Freeman and Nicholas Toman. “Stop Trying to Delight Your Customers,” Harvard Business Review, July–August 2010. Corporate Executive Board’s own research.
• Shostack, G. Lynn. “Designing Services That Deliver,” Harvard Business Review 62(1), 1984, 133–139.
• Bitner, Mary Jo, Amy L. Ostrom and Felicia N. Morgan. “Service Blueprinting: A Practical Technique for Service Innovation,” California Management Review 50(3), 2008, 66–94. ARAMARK figures from the authors’ working-paper version.
• Gibbons, Sarah. “Service Blueprints: Definition,” Nielsen Norman Group, August 27, 2017.

Chapter 8: what the data can’t see

• Griffin and Hauser (1993), above.
• Nielsen, Jakob. “Why You Only Need to Test with 5 Users,” Nielsen Norman Group, March 18, 2000.

Chapter 9: peaks, ends and effort

• Kahneman, Daniel, Barbara L. Fredrickson, Charles A. Schreiber and Donald A. Redelmeier. “When More Pain Is Preferred to Less: Adding a Better End,” Psychological Science 4(6), 1993, 401–405.
• Redelmeier, Donald A. and Daniel Kahneman. “Patients’ Memories of Painful Medical Treatments: Real-Time and Retrospective Evaluations of Two Minimally Invasive Procedures,” Pain 66(1), 1996, 3–8.
• Redelmeier, Donald A., Joel Katz and Daniel Kahneman. “Memories of Colonoscopy: A Randomized Trial,” Pain 104(1–2), 2003, 187–194.
• Alaybek, Balca, Reeshad S. Dalal, Shea Fyffe, John A. Aitken, You Zhou, Xiao Qu, Alexis Roman and Julia I. Baines. “All’s Well That Ends (and Peaks) Well? A Meta-Analysis of the Peak-End Rule and Duration Neglect,” Organizational Behavior and Human Decision Processes 170, 2022, 104149.
• Kemp, Simon, Christopher D. B. Burt and Laura Furneaux. “A Test of the Peak-End Rule with Extended Autobiographical Events,” Memory & Cognition 36(1), 2008, 132–138.
• Verhoef, Peter C., Gerrit Antonides and Arnoud N. de Hoog. “Service Encounters as a Sequence of Events: The Importance of Peak Experiences,” Journal of Service Research 7(1), 2004, 53–64.
• Dixon, Freeman and Toman (2010), above.
• de Haan, Evert, Peter C. Verhoef and Thorsten Wiesel. “The Predictive Ability of Different Customer Feedback Metrics for Retention,” International Journal of Research in Marketing 32(2), 2015, 195–206.

Chapters 10 to 12: owners, ranking and cases

• Edelman, David C. and Marc Singer. “Competing on Customer Journeys,” Harvard Business Review, November 2015.
• Warby Parker Inc. Form 10-K for the fiscal year ended December 31, 2024.
• Warby Parker Inc. “Warby Parker Announces Second Quarter 2025 Results,” earnings release, Form 8-K exhibit, August 7, 2025.
• Retail Dive. “Warby Parker to end home try-on program as it focuses on stores, digital experience,” August 7, 2025.
• Chewy, Inc. Form 10-K for the fiscal year ended February 1, 2026, filed March 25, 2026.
• Chewy, Inc. “Chewy Announces Fiscal Fourth Quarter and Full Year 2025 Financial Results,” March 25, 2026.