Every record your business keeps about a bad purchase requires a customer willing to accuse somebody. A return accuses the product. A dispute accuses you. A review accuses you in public. A support ticket accuses a process. But the most common form of post-purchase failure accuses nobody at all, because the customer has quietly concluded that the mistake was theirs. They keep the item, they say nothing, and they do not come back. There is no row in any table where that customer appears as anything other than a success.
This is the blind spot at the end of the ecommerce funnel, and it is structural rather than accidental. Your data does not miss these people because your instrumentation is poor. It misses them because every instrument you own is triggered by an accusation, and they are not making one.
The short answer
Post-purchase regret only becomes data when the customer blames someone else. Regret that the customer blames themselves for produces no return, no dispute, no review and no ticket, and is therefore absent from every dataset in the business while remaining fully present in their future purchasing behaviour. The only way to see it is to sample the population that took no action at all and ask them directly, which almost nobody does.
The sample frame gives it away
The clearest evidence for this is sitting in the methodology of the most authoritative returns study in the industry.
The National Retail Federation's 2025 Retail Returns Landscape, published with Happy Returns on 15 October 2025, surveyed 2,006 consumers who had returned at least one online purchase in the previous 12 months, alongside 358 ecommerce professionals at US merchants above $500 million in revenue.
Read that sample definition again. The definitive national study of what goes wrong after a purchase is, by construction, a survey of people who did something about it. Everyone who bought the wrong thing and kept it was screened out before the first question. This is not a criticism of the NRF, whose report is excellent and whose numbers we rely on throughout this cluster. It is an illustration of how completely the field has equated post-purchase failure with the return.
The same screening happens inside your own company, in a dozen places, without anyone deciding it. Voice-of-customer programmes sample reviewers. Win-loss research samples deals. Churn surveys sample the churned. Every one of these is a study of people who declared something, and the declaration is the eligibility criterion.
The accusation each artefact requires
Here is the post-purchase stack, sorted by what the customer has to be willing to assert in order to generate a record.
| Artefact | Who the customer accuses | Cost to the customer of producing it | Which team sees it |
|---|---|---|---|
| Return | The product or the listing | Packaging, a trip, sometimes a fee | Operations, merchandising |
| Chargeback or dispute | You, formally | A call to the bank, some confrontation | Payments, fraud |
| Public review | You, in front of everyone | Time, plus willingness to be seen complaining | Marketing, brand |
| Support ticket | A process or a person | Time, and a conversation they may dread | Support, CX |
| Survey response | Whoever the question names | A minute, if they open it at all | Research, CX |
| Silent keep | Themselves | Nothing | Nobody |
The bottom row is the cheapest thing a disappointed customer can do, which is precisely why it is the most common. It is also the only row with an empty right-hand column.
The distinction from the row above it matters. A customer who disputes a charge has escalated past you, and that is a genuinely under-researched population with its own diagnostic value, which we cover separately in chargeback research. But a dispute at least generates a record, a counterparty and a team who has to respond. The silent keeper generates none of those. This article is about the customer who does nothing, not the customer who goes around you.
Why self-blame is the expensive failure mode
The failures that customers attribute to themselves are, with grim reliability, the ones you caused.
- "I ordered the wrong size." The size chart was wrong, or absent, or contradicted by the photography.
- "I should have read the description more carefully." The critical spec was below the fold, or in a specification table nobody opens, or absent.
- "I didn't really need it." The listing was written to make the purchase feel necessary.
- "I misunderstood what it came with." The bundle contents were ambiguous.
- "I didn't realise it wouldn't arrive in time." The delivery estimate was optimistic.
Every one of these is a merchandising or content failure that presents to the customer as a personal failing. The customer's explanation is internal, so no accusation is made, so no artefact is created. The category of error most likely to be your fault is the category least likely to reach you, because embarrassment suppresses reporting more effectively than any policy could.
This is why "we barely get returns in this category" is entirely compatible with widespread regret in that category. A low return rate in a low-value category usually means the item was not worth the trouble of sending back, and we work through the general version of that ambiguity in why a falling return rate is two different stories.
The industry is manufacturing this population
Three verified numbers from the same NRF report, read together, describe a machine for converting returns into silent keeps.
- 72% of US merchants now charge for at least some return options, up from 66% the year before.
- 82% of consumers say free returns are an important purchase consideration, up from 76%.
- 71% say they are less likely to shop with a retailer again after a poor returns experience, up from 67% in 2024.
Every incremental unit of return friction moves some customers from the returning population, where they are visible, expensive and recoverable, into the silent keeping population, where they are invisible, apparently free and quietly gone. The finance line improves. The research pipeline degrades. Neither effect is attributed to the policy that caused it.
There is a defensible reading of the meta-analytic evidence here too. Janakiraman, Syrdal and Freling's review in the Journal of Retailing (2016) found that effort leniency, meaning how much hassle a return requires, increases purchases. Adding effort therefore does not only suppress returns; it suppresses future buying. The mechanism that hides your bad purchases is the same mechanism that reduces the good ones.
A note on the evidence you will find if you search for this
Search for buyer's remorse statistics and you will find a series of confident percentages, most tracing back to brand-commissioned online panel surveys issued as press releases, with no published methodology, no sample frame and no field dates. We are not citing them, and neither should you. They are marketing collateral wearing the costume of research, the same category error we described about vendor-reported feature impact in product detail page research.
The honest position is that there is no reliable public benchmark for the silent keep rate, precisely because measuring it requires sampling a population defined by inaction, and no commercial incentive exists to fund that. Which means the number is unknown for your business specifically, and knowable only by you.
The non-returner probe
The instrument follows directly from the problem. Sample the population everyone treats as success.
Population: orders delivered 30 to 60 days ago, not returned, no support contact, no review. In most catalogues this is 75 to 90 percent of orders, and it is the group no research programme ever touches.
Do not lead with satisfaction. A satisfaction question is answered fluently and positively by people who have not thought about the product since it arrived, which makes it the single worst instrument for this population. It will tell you everything is fine.
The discriminating question is about a road not taken: "Did you think about sending it back? What stopped you?"
That splits the kept population three ways, and the split is the finding.
| Response | What it means | Currently counted as | Actually |
|---|---|---|---|
| Never crossed my mind | The purchase was right | Satisfied customer | Satisfied customer |
| I thought about it, but the process put me off | Returns friction worked | Satisfied customer | A returns failure that cannot appear in returns data |
| I thought about it, but it wasn't worth it for the price | The item is below the complaint threshold | Satisfied customer | A lost repeat customer at zero recorded cost |
The second and third groups are, by construction, absent from every returns dataset in existence. They are the reason your return reasons and your repeat purchase rate seem to belong to two different companies.
The measure: the kept-regret rate. The share of non-returning customers who say they would not buy the item again. Every one of them is currently counted as a successful order and most are counted as retained customers.
This is a different instrument from the re-decision probe we described for subscription revenue in replenishment and subscription research, and the difference is worth stating precisely. That probe addresses recurring charges where no decision was ever made after the first one. This one addresses a single purchase where a decision was made, at a knowable moment, and has since been privately reversed. The subscription case has no decider; this case has a decider who has already delivered a verdict to an empty room.
Three design rules:
- Ask when they last thought about the item. It sorts the base more sharply than any satisfaction scale. Recent thought means it is in use. No thought in six weeks means it is in a cupboard.
- Force the reconstruction. "What did you expect it to be?" before "what was it?" Asking for an evaluation invites a summary; asking for the expectation retrieves the detail.
- Ask what would have changed the purchase. Non-returners who regret the purchase can usually name the exact sentence that was missing, because they have had a month of living with its absence.
Where Koji fits
This population is defined by having done nothing, which makes it hard to reach in exactly the ways traditional research is bad at. There is no complaint to respond to, no ticket thread to continue, no reason for them to accept a scheduled call. Response rates to conventional surveys from this group are poor, and the ones who do respond are the engaged minority, which reintroduces the selection problem you were trying to escape.
Koji is built for the case where the population is large, low-salience and unreachable on a researcher's calendar. AI-moderated voice or text interviews run on the customer's own schedule, with real-time follow-up on every answer. When somebody says "it was fine, I suppose", the AI asks what "fine" is doing in that sentence. That probe is the entire study, and it is the one a survey cannot perform and a human moderator performs inconsistently across two hundred conversations.
The structured question types matter here more than in most studies, because you need the split and the story from the same person:
- yes_no: did you consider returning it? This is the variable everything else is analysed against.
- single_choice: what stopped you? Fee, hassle, time, not worth it, could not find the policy.
- scale: how likely are you to buy this product again?
- ranking: order the factors that made this purchase disappointing.
- multiple_choice: which sources shaped what you expected, from photographs to reviews to the specification table.
- open_ended: what did you think you were buying? The one that produces the fix.
All six run in one conversation, as described in structured questions, and Koji's automatic thematic analysis clusters the open responses into themes with quotes attached, so the result is a set of findings with evidence rather than a transcript pile. One-click reports mean the output is shareable the same day, which matters because the audience for this finding is merchandising and creative, not researchers.
The comparison with the alternatives is not close. Traditional platforms are built to capture people who raise their hand: feedback widgets, review solicitation, ticket analytics, panel recruitment. All of them sample declarers. The whole point of this study is to sample the people who declared nothing, and that only works if the interview is cheap enough to run at the scale of your quiet majority. That is the difference between a six-week, five-figure panel study and a programme you run every quarter against a moving catalogue.
What to do in the next quarter
- Count your silent population. Orders delivered, not returned, no ticket, no review. That percentage is the share of your customer base no research programme has ever spoken to.
- Run 60 non-returner interviews in your highest-volume category. Split by whether they considered returning, and expect the second and third groups to be larger than anyone predicts.
- Establish your kept-regret rate as a baseline. It is unbenchmarked publicly, so its value is entirely in its movement over time.
- Diff regret against the listing. For each recurring regret, find the sentence that should have prevented it. Missing is a content fix; present but unread is a hierarchy fix.
- Re-run it after any returns policy change. Fee increases move customers from visible to invisible, and this is the only instrument that watches the destination.
- Change one reporting rule. Delivered-and-not-returned is currently reported as success. It is an unmeasured category, and naming it that way is the first honest step.
Related reading
- Product Returns Research (2026): Why the Reason Code Is a Dropdown, Not a Finding
- How to Reduce Return Rate (2026): Why a Falling Return Rate Is Two Different Stories
- Replenishment and Subscription Research (2026)
- Product Detail Page Research (2026)
- Chargeback Research: Why Customers Dispute Instead of Contacting You
- How to Design Post-Purchase Surveys
Frequently Asked Questions
What is a silent keep?
A silent keep is a purchase the customer regrets but does not return, dispute, review or contact you about. Because every post-purchase record requires the customer to accuse someone, and a customer who blames themselves accuses nobody, the silent keep produces no artefact anywhere in your systems. It is counted as a successful order and behaves like a lost customer.
Why don't unhappy customers just return the item?
Often because returning costs more than the disappointment is worth. The NRF found 72% of US merchants now charge for at least some return options, up from 66%, while 82% of consumers say free returns matter to their purchase decision. Below a certain order value the rational move is to keep the item and never buy from you again, which is the outcome least visible in your data.
How do I measure post-purchase regret?
Sample orders delivered 30 to 60 days ago with no return, ticket or review, and ask whether the customer considered returning the item and what stopped them. The kept-regret rate is the share of these non-returners who say they would not buy the product again. Avoid leading with a satisfaction question, which is answered positively and fluently by people who have not thought about the product in weeks.
Is there a benchmark for buyer's remorse rates?
Not a credible public one. Most circulating figures come from brand-commissioned panel surveys issued as press releases without published methodology or sample frames. Since measuring the silent keep requires sampling a population defined by inaction, no widely trusted benchmark exists. The number is only knowable for your own catalogue, which makes its trend more useful than its level.
How is this different from a post-purchase survey?
A post-purchase survey measures satisfaction with an experience the customer has just had, and is answered mainly by the engaged. This study targets people who deliberately did nothing, and its central question is about an action they considered and declined. The selection problem is the point: conventional instruments sample customers who declared something, and the silent keeper's defining trait is that they declared nothing.
Can this be run without a research team?
Yes. The barrier has always been that this population is large, low-salience and will not schedule a call, which makes traditional moderated research uneconomic. Koji runs AI-moderated voice or text interviews at that scale on the customer's own schedule, probes each answer in real time, and returns themed analysis with supporting quotes, so a non-returner study fields in days rather than weeks.
Find the customers your dashboard calls successful
Somewhere between a tenth and a third of your delivered orders are sitting in a cupboard, unreturned and unregretted only in the sense that nobody was ever asked. They are counted as wins today and they will not appear in next year's revenue.
Run a non-returner study with Koji and hear the verdict they never filed.
