
← ResearchResearch note11 min read
Bundles on repeat
A subscription is a bundle that has to be chosen again every month. What the research says about why people stay.
Abstract
A subscription box is sold once and then has to survive a quiet vote every cycle: is this still the right set of things for me? Published research suggests that people sign up partly out of optimism and a liking for flat prices, that a large share leave within months, and that the survivors of any group of subscribers are a different mix of people from the group that started. Work on variety seeking explains why a box that never changes wears out even when every product in it is liked. Taken together, and with regulators now making it easier to cancel, retention comes down to whether the box still fits, which is the question of what sells together asked again every month.
Key findings
- 01In McKinsey’s 2018 survey of US shoppers, nearly 40% of e-commerce subscribers had cancelled, most of them within six months.
- 02A falling churn rate does not prove subscribers are growing more loyal: it also appears when the least committed simply leave first.
- 03People choose subscriptions partly because they overestimate their own future use and like flat prices, so a sign-up is weak evidence of fit.
- 04Repetition wears out even favourite products, and people will give up some enjoyment to get variety.
- 05Easy cancellation is becoming law in the UK and is back under review in the US, which leaves fit as the main thing holding a subscriber.
A one-off bundle is chosen once. The shopper builds the box, pays, and the decision is over. A subscription bundle is chosen once and then quietly put to the vote again before every delivery. Nothing has to go wrong for the answer to change. The subscriber only has to feel, one month, that the box is no longer quite the right set of things.
That makes retention the number a subscription lives or dies by, and it raises a practical question for any merchant running one: why do people stay, and what can be done for the ones who are about to leave? This note goes through what the published research says. Our reading of it is that most of the answer is about fit. A subscription is a "what sells together" question that never stops being asked.
Three kinds of subscription
The most widely used map of the market comes from McKinsey, which surveyed 5,093 US consumers in November 2017 and published the results the following February.1 It found that 15% of online shoppers had signed up for at least one e-commerce subscription, and sorted those subscriptions into three types. Replenishment sends the same staples on a schedule: razors, vitamins, coffee. Curation sends a selection chosen to surprise or suit the subscriber: meal kits, beauty boxes, clothing. Access charges a fee for members’ prices or perks. Curation was the largest, at 55% of subscriptions, with replenishment at 32% and access at 13%.1
The same survey gave the industry its most quoted churn figures. Nearly 40% of e-commerce subscribers had cancelled a subscription. Cancellation came early: more than a third of people who signed up left in under three months, and over half within six.2 Replenishment held on best, with 45% of its subscribers staying at least a year. The reasons people gave for leaving were ordinary ones: the quality of the products, dissatisfaction with the assortment, and a sense that the box was not worth the money.2
Two of those three reasons are about what was in the box. This is self-reported survey data from one country and one moment, taken before the pandemic changed how people shop, and it should be read that way. But the shape it describes, a steep early loss followed by a slower one, turns up almost everywhere subscriptions are measured.
Why people sign up, and what that hides
Before asking why subscribers leave, it helps to ask why they joined, because the research on that is unflattering. In a well-known study, Stefano DellaVigna and Ulrike Malmendier followed 7,752 members of three American health clubs over three years.3 Members who chose a flat monthly fee of more than $70 went, on average, 4.3 times a month, which works out at over $17 a visit when a pay-as-you-go pass cost $10. The typical monthly member gave up about $600 over the course of a membership. The authors’ explanation is overconfidence: people sign up for the person they intend to be.
There is a second finding in the same paper that matters for subscriptions. Members on a monthly contract that renewed automatically were 17% more likely to still be enrolled after a year than members who had committed to a full year up front, even though the monthly members were paying more for the freedom to leave.3 They overestimated how readily they would cancel, as well as how often they would go.
Anja Lambrecht and Bernd Skiera found a similar pull in internet access plans, and gave its parts names.4 People choose flat rates even when paying by use would be cheaper, which is known as the flat-rate bias. They do it because a fixed price insures them against a surprising bill, because use is more enjoyable when no meter is running, and because they overestimate how much they will use. Interestingly, customers who overpaid on a flat rate were not significantly more likely to leave.
For a subscription box the lesson is double-edged. Optimism and a taste for fixed prices help people sign up, so a sign-up is weak evidence that the box suits them. And some of the retention a merchant sees is inertia. Neither is a foundation to build on, for a reason we come to under regulation.
How a cohort decays
A cohort is simply everyone who started in the same period, followed over time. Plot the share still subscribed each month and you get a retention curve. The figure below shows the typical shape.
The standard reference on reading such curves is a 2007 paper by Peter Fader of Wharton and Bruce Hardie of London Business School.5 They start from real data on a subscription business, shown in the table. Notice what happens to the renewal rate. In the first year, 63% of the regular customers renew. Among those who remain, 74% renew the following year, then 82%, then 85%. Churn falls every year.
| Year | Regular: still subscribed | Regular: renewal rate | High end: still subscribed |
|---|---|---|---|
| 1 | 63.1%5 | 63% | 86.9% |
| 2 | 46.8%5 | 74% | 74.3% |
| 3 | 38.2%5 | 82% | 65.3% |
| 4 | 32.6%5 | 85% | 59.3% |
| 5 | 28.9%5 | 89% | 55.1% |
| 12 | 17.3%5 | 95% | 39.4% |
The natural reading is that customers grow more loyal the longer they stay. Fader and Hardie show that the curve can be reproduced without assuming anything of the kind. Their model, the shifted beta geometric, is built on a deliberately plain story. At the end of each period every customer flips a coin to decide whether to cancel. Each customer’s coin keeps the same odds for ever. But the odds differ from one customer to the next.5 That is all. The customers with the worst odds leave first, the ones who remain are those who were always likely to stay, and the average renewal rate rises although no individual has changed. The paper calls this, after earlier demographers, a ruse of heterogeneity.
This has practical consequences. Projecting a cohort forward at a constant churn rate is wrong in a predictable direction: Fader and Hardie show a constant-rate curve underestimating twelve-year survival of the regular segment by 54%.5 And it changes what a retention chart can tell you. A merchant who sees churn fall between month two and month six has not necessarily improved anything. The real questions are how many of each kind of subscriber were acquired, and whether anything done to the box moved a person from one kind to the other.
The box that stops fitting
Why would someone who liked a box in January tire of it by April, when the box has not changed? Consumer research has a long-standing answer: that is exactly why.
In 1982 Leigh McAlister proposed that what people choose depends on what they have recently had.6 Products are bundles of attributes (sweet, bitter, rich, light), and as a person consumes more of an attribute its appeal fades for a while. This is satiation. Tested against diaries of soft drink consumption, her model predicted people’s next choice better than one that treated each choice as independent. On this view, switching is a predictable response to having had enough of something, and need not mean the person likes it any less.
Rebecca Ratner, Barbara Kahn and Daniel Kahneman pushed the finding further in 1999.7 In their experiments people chose to switch away from their favourite option to ones they liked less, even though repeating the favourite would have given them more enjoyment in the moment. Part of the reason is memory: looking back, people rated a varied sequence more highly than a run of their favourites, although they had enjoyed it less at the time.
A subscriber does not have to stop liking the box to stop wanting it.
This is the awkward truth for replenishment-style bundles. The promise is "the same again", and the research says that the same again wears thin. It also suggests why curation is the largest part of the market and still loses so many subscribers: variety is what people want, and a stranger’s guess at variety often misses. The middle path is a box with a stable core and room to change at the edges, and the question of which change suits which box is a pairing question. Given what this subscriber keeps, what belongs beside it next month?
Skip, swap, pause
If a box can stop fitting without anything being wrong, the alternative to cancelling should not be all or nothing. The industry’s answer has been to give subscribers controls: skip a delivery, swap a product, pause for a while.
The best public data on how much these are used comes from Recharge, a subscription platform, whose 2023 report covered more than 15,000 merchants over 2022.8 It found that 35% of subscribers adjusted an order during the year, and that 39% of those who did skipped a delivery. Across the same merchants, 45% of subscribers were still active after six months and 33% after twelve.
We want to be careful about what that shows. It shows that a third of subscribers will use controls when they are offered. It does not show that the controls caused anyone to stay, and the source is a vendor with an interest in the answer. We looked for a published, controlled study that isolates the effect of skip or swap on cancellation in physical subscription boxes and did not find one we could verify. The argument for flexibility rests on the behavioural research above, and on a simple piece of logic: a subscriber with only one button will press it.
A box, worked through
Take an invented tea subscription: six teas a month, chosen by the subscriber at sign-up. In month one the box is exactly what they asked for. By month four, satiation predicts that the two teas they drink most are the two they are most tired of, while a tea they rarely finish is piling up in a cupboard.
With no controls, the subscriber has one way to express any of this, and the merchant sees a cancellation with "too much product" as the reason. With a skip, the cupboard problem is solved for a month and the relationship continues. With a swap, the subscriber replaces the tea they never finish, and in doing so tells the merchant something no survey would: which product does not belong in this box, and which one they think does.
Now put that against the cohort picture. Some subscribers were always going to leave in month two. Controls will not change that. The ones worth attention are those in the middle, for whom the box nearly fits. Every swap they make is the same kind of evidence as a shopper building a bundle for the first time, with the difference that it arrives every month.
The law is moving towards the exit
Retention built on difficulty is running out of road. The details differ by country and are changing, so we give them as they stand in mid-November 2026.
In the United States, the Federal Trade Commission finalised a "click-to-cancel" rule in 2024, requiring that cancelling be as easy as signing up. It never took effect: a federal appeals court, the Eighth Circuit, struck it down in July 2025 on procedural grounds. On 11 March 2026 the FTC reopened the question with an advance notice of proposed rulemaking, an early step that asks for public comment and proposes no text. In announcing it, the agency said it had received more than 100,000 complaints about such practices in five years.9 So there is no federal click-to-cancel rule in force today, though older federal law on online subscriptions and a growing number of state laws still apply.
In the United Kingdom the direction is firmer. The Digital Markets, Competition and Consumers Act 2024 contains a chapter on subscription contracts that requires clear information before sign-up, reminder notices before renewal, cooling-off periods, and a way to end the contract that is straightforward.10 Those provisions are not yet in force. On 9 August 2026 the government announced that they would begin in January 2027, earlier than previously planned.11 The detailed regulations were laid before Parliament in draft on 19 October 2026, with a start date of 1 January 2027 for contracts entered into from that day, and were still awaiting approval when we wrote this.12
We are not lawyers and this is not advice. The commercial point is simpler. Recall the health club members who stayed because cancelling was a chore. As leaving becomes a single click, that source of retention goes, and what remains is whether the subscriber wants the next box.
What is not known
- •How much flexibility helps. As above, the public evidence that skip, swap and pause reduce cancellation is descriptive and comes largely from vendors. It is plausible, and it has not been established.
- •Whether a pause is a save. A paused subscriber may return, or may have found a politer way to leave. We have not found good public data on how often pauses end in resumption.
- •How far the laboratory travels. The variety-seeking experiments used snacks, drinks and songs over short periods. Monthly boxes of coffee or skincare are a different rhythm, and the size of the effect there is not known.
- •How current the numbers are. The McKinsey survey is from 2017 and covers the United States. The Fader and Hardie data are annual and from an unnamed business. Both give the shape of the problem; neither gives a benchmark for a particular store.
What it means for a subscription bundle
Read together, the research says three things. Look at cohorts, never at one average churn figure, because the average hides who is leaving. Expect the box to wear out, because repetition does that to anything. And give the subscriber something to do other than cancel, because the moment a box stops fitting is the moment they will reach for whatever control is nearest.
Kitenzo runs subscriptions on bundles, so the thing that renews is the box the shopper built. From their account a subscriber can skip an upcoming delivery and undo the skip, pause and resume, swap a product in an upcoming box for another from the same bundle without the price changing, and cancel, giving a reason if they wish. Merchants can follow how their bundles perform in reporting.
What we are working on next is the other half: using what subscribers keep and what they swap to suggest how a box should be refreshed before it goes stale. That work is in the lab and is not in the app. We would sooner say that plainly than let a research note imply otherwise.
References
- 1Chen, T., Fenyo, K., Yang, S. and Zhang, J. (2018). Thinking inside the subscription box: New research on e-commerce consumers. McKinsey & Company.
- 2Retail Dive (2018). Nearly 40% of subscribers ultimately cancel services (report of the McKinsey survey). Retail Dive.
- 3DellaVigna, S. and Malmendier, U. (2006). Paying Not to Go to the Gym. American Economic Review, 96(3), 694–719.
- 4Lambrecht, A. and Skiera, B. (2006). Paying Too Much and Being Happy About It: Existence, Causes, and Consequences of Tariff-Choice Biases. Journal of Marketing Research, 43(2), 212–223.
- 5Fader, P. S. and Hardie, B. G. S. (2007). How to Project Customer Retention. Journal of Interactive Marketing, 21(1), 76–90 (authors’ version).
- 6McAlister, L. (1982). A Dynamic Attribute Satiation Model of Variety-Seeking Behavior. Journal of Consumer Research, 9(2), 141–150.
- 7Ratner, R. K., Kahn, B. E. and Kahneman, D. (1999). Choosing Less-Preferred Experiences for the Sake of Variety. Journal of Consumer Research, 26(1), 1–15.
- 8Recharge (2023). The State of Subscription Commerce 2023. Recharge.
- 9Federal Trade Commission (2026). FTC Seeks Public Comment in Response to Advance Notice of Proposed Rulemaking Regarding Negative Option Plans. FTC press release, 11 March 2026.
- 10UK Parliament (2024). Digital Markets, Competition and Consumers Act 2024, Part 4, Chapter 2: Subscription contracts. legislation.gov.uk.
- 11Prime Minister’s Office, 10 Downing Street (2026). PM starts roll-out of everyday fixes on the cost of living, ending rip-off discounts and subscription traps. GOV.UK, 9 August 2026.
- 12CMS (2026). UK subscription contracts: the Government publishes the missing detail. CMS Law-Now, 23 October 2026.


