Every research method your team owns has the same hidden requirement. Surveys, interviews, conjoint, concept tests, product detail page studies, choice architecture experiments - all of them assume there is a person who made a decision, at a moment, for reasons they can be asked about.
Replenishment revenue does not satisfy that assumption. And it is the fastest-growing kind of revenue in consumer commerce.
The short answer
In a replenishment subscription, one decision produces an unbounded number of purchases - and every purchase after the first has no decider, no moment, and no reason attached to it. Your instruments have nobody to interview.
This is not the disclosure problem, which is about whether people understood what they agreed to at signup and is covered properly in auto-renewal and cancellation research. Assume perfect disclosure. Assume an enthusiastic, fully informed customer who read every word. Fourteen months later that customer is still being charged, and there is no longer any act of choosing to study. The revenue is real and the decision is historical.
That gap has a consequence teams feel without naming: satisfaction becomes unobservable at the point of sale, and the only signal the customer ever sends is cancellation - which is binary, lagging, and arrives after the opinion that caused it has already hardened.
The scale of the silent half
Replenishment is no longer a niche. Recharge's Subscription Trend Report 2026, drawn from 20,000 brands on its platform, found that subscribers placed nearly 3x more orders than one-time shoppers and drove a growing share of revenue across every vertical it covers - health and wellness, beauty and personal care, food and beverage, home, and pet. Subscription checkouts rose 16% year over year, and same-day cancels fell 35%.
Recurly's 2026 State of Subscriptions, covering 76 million subscribers, adds two numbers that matter for this argument specifically. Pause usage among top merchants rose 337%, and 1 in 4 new sign-ups are now returning subscribers. Both describe a market where the relationship is increasingly managed by settings rather than by decisions.
Now the customer side, and it is the single most important statistic in this article. West Monroe surveyed 2,500 US consumers across 21 subscription categories in June 2021, asking them to estimate their monthly subscription spend and then itemise it. Average actual spend was $273 a month, up from $237 in 2018. And:
100% of respondents were unaware of their actual spend.
66% were off by more than $200 - against 24% in 2018. Not a few people, not most people. Everybody, and the error was growing.
That figure is now several years old, and the honest caveat is that it predates the current wave of subscription-management tooling. It also almost certainly understates the problem, because the number of recurring relationships per household has gone up since, not down. Either way, the mechanism is what matters: people cannot report their own recurring spending, which means they cannot be relied on to report their own recurring decisions either.
The instrument problem, stated plainly
Here is what a normal purchase gives a researcher, and what a replenishment charge gives instead.
| Property of the purchase | One-time purchase | Replenishment charge |
|---|---|---|
| A person chose it | Yes | Once, months or years ago |
| The chooser remembers choosing | Usually | Frequently not |
| There is a moment to sample around | Yes | No |
| Dissatisfaction has an outlet | Abandon, return, complain | Cancel, eventually |
| The signal arrives near the cause | Yes | Delayed by one or more cycles |
| You can run a test on it | Yes | Not without an impression to vary |
Every row is a "no" or a "once." That is why the standard research toolkit produces so little here: it is not that the methods are bad, it is that there is no respondent in the role the method requires.
The last row deserves attention because it is the one teams try hardest to work around. You cannot A/B test a purchase that has no page view. There is no impression, no variant, no exposure. The only things you can vary are the messages around the charge - the reminder email, the pause offer, the skip prompt - and varying those changes cancellation behaviour, which everybody then reads as a satisfaction result. It is not. It is a measurement of how easy you made it to leave.
Silent revenue, and why it is dangerous
Call the portion of your revenue that arrives without a decision silent revenue. It has three properties that make it behave unlike anything else on your P&L.
It looks like loyalty and may be inertia. A retention curve cannot distinguish a customer who loves the product from one who has not looked at the charge. Both are active subscribers. Only one of them is going to survive a price increase, a competitor's launch, or a household budget review.
It accumulates unexpressed opinion. In a one-time purchase, dissatisfaction gets expressed immediately - a return, a bad review, an abandoned cart. The NRF's 2025 Retail Returns Landscape put the online return rate at 19.3% on $849.9 billion of total 2025 returns, which is a large, fast, honest feedback channel. Replenishment has no equivalent. The dissatisfied subscriber does not return anything. They accumulate a view, quietly, across several cycles, and then act once.
Its failure is discontinuous. Because opinion accumulates without an outlet, the observable series looks flat right up until it does not. Teams describe this as "churn came out of nowhere." It did not come out of nowhere; it came out of a period during which the customer had no channel through which to tell you anything, and you had no instrument pointed at them.
Recurly's 337% rise in pause usage is worth re-reading in this light. A pause is the first moment in the entire lifecycle where a silent customer does something legible. It is a research event, and most teams treat it as a retention loss to be countered with an offer.
The method: the re-decision probe
If there is no decision to study, manufacture one - not in the product, in the research.
Ask active subscribers to choose again from scratch, as though they were not currently subscribed, and record the answer alongside their actual status. The measurement you want is the gap between the two.
The latent churn rate is the share of active subscribers who would not choose the subscription again today. Every one of them is currently counted as retained. Every one of them will eventually appear in a cancellation cohort, and by then it will be too late to ask why, because you will be interviewing somebody who has already rationalised the decision.
This is a genuinely different instrument from a cancel-flow exit interview, which is excellent but structurally late - it samples people at the moment they leave, so it can only ever explain departures you have already suffered. The re-decision probe samples people who have not left yet. It is the only forward-looking instrument available on revenue that generates no decisions.
Three design rules make it work:
1. Do not ask about satisfaction. "How satisfied are you?" is answered by people who have not thought about the product in months, and it produces a flattering number. Ask them to re-choose.
2. Force the reconstruction. Ask what else they would consider, what they would pay, and what would have to be true for them to sign up today. A held preference has content. Inertia does not, and it collapses immediately under a follow-up.
3. Ask when they last thought about it. The answer is the single best predictor in the study. A subscriber who last considered the subscription two weeks ago is in a relationship. One who cannot remember is in a standing order.
| What you need | Question type | Example |
|---|---|---|
| Whether the decision is live | yes_no | Have you thought about this subscription in the last month? |
| Re-decision under a clean slate | single_choice | If you were not subscribed today, would you sign up? |
| The reason, or its absence | open_ended | What would you be replacing it with, and why that? |
| Strength of the standing preference | scale | How much would the price have to rise before you cancelled? |
| Where the value actually sits | ranking | Rank convenience, price, quality, habit, and avoiding the reorder |
| Triggers you can watch for | multiple_choice | Which of these would make you review it? |
The open_ended row is the one that separates a real preference from a standing order, and it is the row a static survey handles worst. "It's just easier" is where a form stops and where the conversation should begin: easier than what, tried when, and what happened.
Two more things worth building in permanently:
- Instrument the pause, not just the cancel. A pause is a customer volunteering that something changed. Interview at pause, and you get the reason while it is still recoverable.
- Use your own price changes as natural experiments. You know the exact date every price moved. The share of subscribers who noticed is a direct measure of how much attention your revenue is receiving - and attention, not satisfaction, is the variable that predicts what happens next.
Where Koji fits
Everything above has one operational requirement: talking to a lot of currently-happy customers about a decision they did not make, at a cadence that catches drift before cancellation. That is exactly the study legacy tooling makes uneconomic.
SurveyMonkey and Typeform can reach the volume but cannot follow up, so "it's just easier" is where your data ends. UserTesting and dscout can get the depth but require recruiting and scheduling, which puts the cost per conversation far above what a routine subscriber check-in can justify - so it never gets run. Qualtrics gives you a tracker that measures satisfaction, which is the question that does not work here. Dovetail organises transcripts somebody else still has to produce.
Koji runs AI-moderated voice interviews at survey scale and interview depth, so the re-decision probe can run quarterly across a live subscriber base without anybody scheduling a call. The AI consultant is customizable to your category, so the follow-ups are specific rather than generic. Thematic analysis runs automatically, so the latent churn rate and the reasons behind it come back as a ranked report rather than a pile of recordings - from question to insight in hours, not weeks, with no research background required and no moderator whose phrasing drifts between interview 5 and interview 200.
Pair it with structured questions to quantify the re-decision gap, cancel-flow exit interviews for the departures you do suffer, win-back interviews for the ones already gone, and the churn hazard curve to see which tenure bands your latent churn is concentrated in.
What to do in the next quarter
- Calculate what share of your revenue arrived this month without any customer action. That is your silent revenue, and most teams have never put a number on it.
- Run the re-decision probe on 100 active subscribers and publish the latent churn rate next to your retention rate. Expect them to disagree.
- Start interviewing at pause rather than only at cancel. Pauses are the earliest legible signal you will ever get.
- Add "when did you last think about this?" to any subscriber research you already run. It costs one question and it sorts your base into relationships and standing orders.
Silent revenue is still revenue. It is just revenue you know nothing about - and the first quarter you find that out should not be the quarter it leaves.
Related reading
- Choice Architecture and Defaults (2026)
- Product Detail Page Research (2026)
- Agentic Commerce Research (2026)
- Shopper Mission Research (2026)
- Customer Research for Ecommerce Brands
- The Regret That Never Files a Return (2026) - measuring customers who keep what they did not want.
Frequently Asked Questions
What is latent churn?
Latent churn is the share of currently active subscribers who say they would not choose the subscription again if they were deciding today. They are counted as retained in every dashboard you have, and they represent cancellations that have effectively already happened but have not yet been executed. Measuring it converts a lagging metric into a leading one.
How is this different from a cancel-flow exit interview?
A cancel-flow exit interview samples people at the moment they leave, so it explains departures you have already suffered and can sometimes save the individual customer. The re-decision probe samples people who have not left, which is the only way to see the problem while it is still fixable at the cohort level. Run both; they answer different questions at different points in the lifecycle.
Isn't this just a satisfaction survey?
No, and the difference is the whole method. Satisfaction questions are answered fluently by people who have not thought about the product in months, and they produce reassuring scores. Asking somebody to re-choose from a clean slate forces a reconstruction that inertia cannot fake - a real preference names an alternative and a reason, while a standing order collapses under one follow-up question.
Why can't we A/B test a replenishment purchase?
Because there is no impression to vary. A test requires an exposure that can differ between arms, and a scheduled charge involves no page, no offer, and no moment of choice. You can only test the messages around the charge - reminders, skip prompts, pause offers - and those measure how easy you made it to leave, not whether the customer still wants the product.
How often should we run the re-decision probe?
Quarterly is a sensible default for most replenishment categories, with an additional run whenever you change price, formulation, delivery cadence, or packaging. The cadence matters more than the sample size, because the value comes from watching the latent churn rate move against known changes rather than from any single measurement.
Does this apply to B2B subscriptions too?
Yes, and the decider problem is usually worse. In B2B the person who signed up has often changed roles or left the company, so the renewal is approved by somebody who never made the original decision and has no reason attached to it. The re-decision probe works the same way, but you should identify who currently holds the mandate before you decide whom to interview.
