Document status: Published. Version 1.0.
Date of issue: 11 September 2026.
Issued by: Confi Technologies, Inc., San Francisco Bay Area, United States.
Correspondence: section 14.
1.1 This document sets out the derivation, the underlying sources, the assumptions and the limitations of every quantified figure that Confi publishes in its marketing and public communications.
1.2 Two figures fall within scope:
(a) approximately $340 per year, published as the annual amount lost by a person who buys online regularly and pays for one or more subscriptions; and
(b) $500 to $740 per year, published as the annual amount lost by a person on whom all of the categories below fall within the same year.
1.3 Four categories of loss are within scope. They correspond to the circumstances Confi's product concerns itself with.
| # | Category | Counted in the figures? |
|---|---|---|
| (i) | Subscriptions paid for and not used | Yes |
| (ii) | Returns not made because the return period expired | Yes |
| (iii) | Orders that did not arrive and for which the buyer was not made whole | Yes |
| (iv) | Refunds owed and never received | No. Named, and counted at zero. See section 7. |
1.4 A fifth matter — having every order visible in one place — is a convenience. It is not quantified in this document and Confi does not attach a monetary value to it in any public statement.
1.5 Both figures are estimates of consumer loss. Neither is a statement of savings, recovery or benefit attributable to Confi or to any product. No inference of that kind is intended by this document or supported by the evidence cited in it. Confi does not claim that any amount described here is recoverable, by its product or by any other means.
1.6 This document does not describe Confi's products or how they operate. It addresses the figures only.
1.7 This document is maintained as a live record. Where a superior source becomes available, or an error is identified, the figure is amended and the amendment recorded at section 12 with the superseded value left visible.
2.1 This is the most important section of this document, and the one most often got wrong elsewhere.
2.2 The figures in this document are not population-wide averages across all United States adults. They describe a defined population: a person who buys online regularly and pays for at least one subscription.
2.3 That distinction is material in both directions. A population-wide average divides the loss across everyone, including people to whom none of it happens, and produces a smaller number. A figure stated for the population actually affected produces a larger one. Confi publishes the second, and states which it is publishing.
2.4 Each input below is stated in the units of the population it was measured in, and section 6 sets out where those units do not align.
2.5 Confi's published wording names the population before the number, and does so in terms of behaviour rather than demography. The canonical form is: for someone who shops online and keeps a few subscriptions running. Confi does not publish either figure in a form that could be read as "the average American."
3.1 Three classifications are applied to every quantity in this document.
| Classification | Meaning |
|---|---|
| Measured | A study was conducted, a sample drawn, and the quantity published by the party that conducted it. |
| Reported | Respondents were asked to recall events affecting them and answered from memory. The quantity aggregates self-reported recall, not observed transactions. |
| Derived by Confi | The quantity appears in no published study. Confi performed arithmetic on quantities published by third parties. The arithmetic is set out in full. |
3.2 Both published figures are classified derived by Confi. Every input is classified reported. No input to either figure is classified measured. Confi states this plainly rather than leaving it to be discovered.
4.1.1 The figure is the sum of three published quantities drawn from separate third-party surveys, with a fourth category named and counted at zero. No study has produced this figure; it appears in no source cited at section 13.
| # | Component | Conservative | Centre | Source | Denominator as published |
|---|---|---|---|---|---|
| (i) | Subscriptions paid for and not used | $204 | $250 | [1], [5] | Per subscriber |
| (ii) | Returns not made, period expired | $15 | $23 | [2], [3], [4] | Per online shopper, per year |
| (iii) | Orders not arrived, not made whole | $101 | $101 | [6] | Per household |
| (iv) | Refunds owed and never received | $0 | $0 | — | No published measurement exists |
| Total, derived by Confi | $320 | $374 |
4.1.2 The published figure of approximately $340 sits between the conservative construction of $320 and the centre construction of $374. It is not the highest value the sources support.
4.2.1 Reference [1] reports that subscribers spend an average of $17 per month on subscription services they do not use. $17 × 12 = $204 per year. Sample 2,440 United States adults, nationally weighted, fielded 2025. The denominator is subscribers. The population described at section 2 is a subscriber by definition, so the quantity applies without adjustment.
4.2.2 Reference [5] reports $26.79 per month among adults holding at least one unused subscription, equal to $321 per year. Sample 1,272 United States adults, fielded 2026. This is the upper value and applies to a person who currently carries a subscription she does not use.
4.2.3 Reference [5] further reports that 59.9% of adults hold at least one unused subscription. Applying that proportion to the upper value yields $321 × 0.599 ≈ $192, which converges on reference [1]'s $204. Two independently conducted surveys estimating the same quantity differ by approximately 6%. That convergence is the reason the $204 figure is treated as solid.
4.2.4 The centre value of $250 is Confi's midpoint between the two published figures.
4.2.5 Disclosed weakness. Reference [1] measures subscriptions paid for and not used, which is broader than forgotten. Some unused subscriptions are retained deliberately. The strictly forgotten subset is smaller than $204, and Confi has not been able to quantify by how much.
4.2.6 Context, not an input. Reference [7] found that respondents estimated their total monthly subscription spending at $86 against an itemised actual of $219 — a 2.5× underestimate. That study measured total rather than unused expenditure and is not an input here, but it establishes that self-reported recall in this category errs downward.
4.3.1 This component was previously derived incorrectly and the error is corrected in this version. The prior construction multiplied a lifetime incidence (44% of shoppers have at some point missed a return window, reference [3]) by a lifetime average value (reference [2]) and presented the product as an annual figure. Two lifetime rates multiplied do not produce an annual rate. The error is recorded at section 10.
4.3.2 The corrected construction uses an annual-framed source. Reference [4] reports that 15% of United States shoppers were left with unwanted merchandise in the preceding twelve months because the return deadline had expired, and that the average abandoned return was worth $50. Sample 1,000 United States adults, census-weighted, fielded August 2025.
4.3.3 Reference [4] also reports that shoppers return approximately 13 online purchases per year, rising with income; reference [2] reports approximately 9 per year across a broader sample. The population described at section 2 buys online regularly and therefore sits at the higher end of return volume, which raises annual exposure proportionally.
4.3.4 Confi therefore estimates annual incidence for this population at 0.25 to 0.50 occurrences per year, against a population-wide rate of 0.15. Value per occurrence is taken at $50 (reference [4], annual-framed) to $78 (reference [2], among shoppers who kept an item).
Conservative: 0.25 × $50 = $13, stated as $15
Centre: 0.35 × $65 = $23
Upper: 0.50 × $78 = $39
4.3.5 Disclosed weakness. The incidence range is Confi's estimate, extrapolated from a published population-wide rate and a published return-volume differential. It is not a measured quantity for this population and no survey has measured it.
4.4.1 Reference [6] reports aggregate United States consumer losses from deliveries stolen or not received of $12.8 billion per year, net of amounts reimbursed by retailers, and a per-household equivalent of approximately $101. Method: consumer panel of 1,000 United States respondents blended with Federal Bureau of Investigation larceny statistics.
4.4.2 The implied household count is $12.8bn ÷ $101 ≈ 126.7 million, consistent with published United States household counts.
4.4.3 The quantity is taken directly from the source without adjustment. Reference [6] additionally reports that 29.7% of households were affected in the reference period and that 23.5% of affected households received no reimbursement at all.
4.4.4 Disclosed weakness. This is a per-household quantity used alongside per-person quantities. Section 6.2 sets out the effect.
4.5.1 Counted at zero. No published survey or study produces a dollar figure or an incidence rate for this category. Section 7 sets out the search conducted and what does exist.
4.5.2 The category is named in Confi's published wording because it is one of the four things Confi's product concerns itself with. It contributes nothing to either figure.
4.6.1 Conservative: $204 + $15 + $101 + $0 = $320
Centre: $250 + $23 + $101 + $0 = $374
4.6.2 Confi publishes approximately $340, which lies within that range and below its midpoint of $347.
4.6.3 The rounding convention is at section 11. Rounding is applied once, after summation, never to a component beforehand.
5.1.1 Figure 2 addresses a different question. Where Figure 1 describes a person who buys online and holds subscriptions, Figure 2 describes a person on whom all three counted categories fall within the same year, at the upper end of each.
| # | Component | Low | High | Source |
|---|---|---|---|---|
| (i) | Unused subscriptions, among those holding one | $321 | $321 | [5] |
| (ii) | One return not made, among those who kept the item | $78 | $78 | [2] |
| (iii) | Orders not arrived, not made whole | $101 | $340 | [6] |
| Total, derived by Confi | $500 | $739 |
5.1.2 $739 rounds to $740 under the convention at section 11.
5.1.3 The high value of component (iii) is derived as $101 ÷ 0.297 = $340.07, converting the per-household figure to a per-affected-household figure using reference [6]'s reported 29.7% affected rate. It is the average across all affected households, including those that received partial reimbursement. It is not the loss of a household that received nothing, and is not presented as such.
5.2.1 An individual on whom all three counted categories fall within one year. It is not a typical outcome and is not presented as one. Most people experience fewer than three of these in a year, and some experience none.
5.2.2 The range is bounded by a single variable. Components (i) and (ii) are held constant; only component (iii) moves between the low and high ends. The range therefore expresses uncertainty in one of three inputs and is not a confidence interval across the estimate as a whole.
5.2.3 The figure assumes the events are independent and nets nothing beyond the reimbursement already deducted by the publisher of reference [6].
6.1 Denominators are not uniform across the inputs.
| Input | Published denominator |
|---|---|
| Subscriptions ($204) | Adults holding a subscription |
| Returns ($15) | Online shoppers, annual, adjusted by Confi for return volume |
| Orders not arrived ($101) | All households |
6.2 Households are not individuals. Component (iii) is per household and is used alongside per-person quantities. There are materially fewer households than adults in the United States. Where a household contains more than one online shopper, the per-person figure is lower than $101; where a household contains one, it is the same. Confi does not adjust for this, and states here that not adjusting causes the figure to be larger than a strictly per-person construction would produce.
6.3 All inputs are self-reported. None derives from transaction records, bank data, card data or retailer records. Self-reported recall of financial loss is subject to error in both directions; reference [7] measured that error in one of these categories and found it substantial and downward.
6.4 Reference periods are not aligned. Sources were fielded between September 2024 and March 2026. No adjustment for inflation or for changes in retailer policy has been applied.
6.5 No combined error term can be stated. Because the figures are sums of point estimates from unrelated studies, the margins of error published by the individual studies do not compose into a margin for the sum.
6.6 Reference [6] is not purely a survey. It blends a consumer panel with Federal Bureau of Investigation larceny statistics. The result is neither a pure survey estimate nor a pure administrative statistic.
6.7 One component is a Confi estimate rather than a published quantity. The annual incidence range at section 4.3.4 is extrapolated, not measured. It contributes between $15 and $39 to Figure 1.
7.1 Refunds owed and never received. A refund is confirmed as due and does not reach the consumer. The consumer forgets, gives up, or never notices.
7.2 Search conducted. Confi searched for a population-wide dollar figure or incidence rate: the Federal Trade Commission's Consumer Sentinel Network published category data; the Consumer Financial Protection Bureau's complaint database and reports; Better Business Bureau published statistics; chargeback-industry reporting including references [8] and related publications; reverse-logistics and returns-industry research including Narvar, Loop Returns, Optoro, goTRG and Happy Returns; consumer surveys by finance media, market-research firms and trade bodies; academic consumer-redress literature; and class-action databases.
7.3 No consumer survey located anywhere publishes a dollar figure or an incidence rate for this category. Confi does not assert that no such figure exists anywhere; it asserts that none was located in the sources searched, and invites correction under section 14.
7.4 What does exist is a court record, and its scale is instructive. In In re: Amazon Return Policy Litigation [9], a settlement filed in January 2026 established a $309.5 million fund, in addition to approximately $570 million Amazon had already refunded, covering United States consumers who between September 2017 and the class data date did not receive a refund, received an incorrect refund, received a late refund, or were erroneously recharged. The largest subclass covers returns that were never processed because the item was lost in transit or the review never completed.
7.5 That is approximately $880 million of consumer money, at one retailer, over roughly eight years, for precisely this failure — and it covers only the cases that retailer's own records could identify. It excludes every instance a customer never raised.
7.6 Confi does not use this as an input. It is a fact about one company and a settlement, not a population estimate. It is recorded here because it establishes that the category is neither hypothetical nor small.
7.7 The asymmetry of measurement. The corresponding cost on the retailer side of the same transactions is estimated annually and published in detail. Reference [10] estimates that 15.8% of United States retail sales will be returned in 2025, equal to $849.9 billion, and 19.3% of online sales; the 2024 estimate was 16.9% and $890 billion. It is broken out by channel, by category and by the share assessed as fraudulent, and it is produced every year.
7.8 The retailer-side cost of returns is therefore among the more closely estimated quantities in United States retail. Confi located no comparable recurring estimate of the corresponding cost to the consumer in the category at section 7.1.
7.9 Consequence. Both published figures are built only from categories somebody has measured. One of the four categories Confi's published wording names contributes zero. Confi's position is that the figures therefore understate total loss in this subject area, and that Confi cannot state the magnitude of the understatement. The direction of the omission is known; its size is not. Confi does not treat the omission as grounds for increasing either figure and has not done so.
8.1 The figures do not measure, and must not be read as measuring:
(a) savings, recovery or benefit obtainable by any means, including by use of any product;
(b) amounts recoverable in a dispute, chargeback or claim;
(c) time, inconvenience or any non-monetary cost, including the convenience described at section 1.4;
(d) losses from fraud, identity theft or unauthorised transactions;
(e) losses outside the United States;
(f) any individual's actual loss. Both figures are estimates across populations, and no individual's experience is predicted by either.
9.1 Selection criteria, applied in order: the largest sample; the most clearly disclosed method; the publisher with the least direct commercial interest in the size of the result; the most recent field date.
9.2 On that basis the subscription component is taken from reference [1] — the largest sample located (n=2,440), nationally weighted, published by a media organisation rather than a vendor of a related product. The returns component is taken from reference [4] because it is the only annual-framed source located with a census-weighted sample.
9.3 Commercial interests, disclosed.
| Reference | Publisher | Position |
|---|---|---|
| [2] | CardRates | Operates a credit-card comparison service |
| [3] | LendingTree | Operates a consumer lending marketplace |
| [5] | Self Financial | Sells a consumer credit-building product |
| [6] | Omnisend | Sells software to online retailers |
| [10] | National Retail Federation with Happy Returns | Trade association; co-publisher sells returns-processing services |
9.4 Confi makes no assertion about the intentions of any of these parties. Their positions are recorded so that a reader may weigh the sources accordingly.
9.5 Sources located and not used. A package-theft estimate of $15 billion with an average stolen-package value of $144, published by a home-security vendor, and a second estimate of $8.2 billion gross with an average of $222 per package [11]: both excluded because reimbursement is not netted out, which makes them non-comparable with reference [6] and would have produced a larger figure. A 2021 survey reporting that 71% waste $50 or more monthly on subscriptions: excluded because the amount was self-estimated rather than itemised. Aggregated user-savings claims published by subscription-cancellation applications: excluded because the user base is self-selected and the figures unaudited.
9.6 Used as corroboration only, contributing no quantity: references [7] and [12].
10.1 The following were present in Confi's prior working materials and are corrected here.
| # | Error | Correction |
|---|---|---|
| 1 | The returns component multiplied a lifetime incidence by a lifetime average and presented the result as annual. | Rebuilt on an annual-framed source [4]: 15% in twelve months, $50 average. Section 4.3. |
| 2 | The $78 average was attributed to reference [3]. | It is published by reference [2]. Reference [3] publishes the 44% lifetime incidence and no dollar figure. |
| 3 | A retail-returns total of $890 billion was carried without its year. | $890 billion is the 2024 estimate. The 2025 estimate is $849.9 billion. Both attributed with their years. |
| 4 | The returns component was stated at approximately $36 and the total at $337. | Superseded entirely by the rebuild at section 4.3. |
| 5 | The upper bound of Figure 2 was stated as $750. | The sum is $739, which rounds to $740. |
| 6 | The high value of component (iii) was described as the loss of a household that received nothing. | It is the average across all affected households. Section 5.1.3. |
| 7 | The subscription component was described as a per-adult quantity. | It is published per subscriber. Section 4.2.1. |
10.2 The corrections at items 1, 2 and 3 were identified by an adversarial review commissioned by Confi of its own published material.
11.1 Rounding. To the nearest ten, applied once after summation, never to a component beforehand.
11.2 Direction. Standard arithmetic rules, never applied selectively in the direction favourable to Confi.
11.3 Attribution. No quantity is published without an identified publisher, year, sample size and method. The formulation "studies show" is not used.
11.4 Classification. No derived quantity is presented as a measured one. Where Confi performed the arithmetic, Confi says so.
11.5 Population. No figure is published without naming the population it describes, before the number.
11.6 Currency and period. United States dollars, annual rates, no inflation adjustment applied.
| Version | Date | Change | Superseded value |
|---|---|---|---|
| 1.0 | 11 September 2026 | First published version. | Not applicable. |
12.1 Subsequent revisions record the figure changed, its previous value, the source or correction causing the change, and the date.
[1] CNET, survey conducted by YouGov. Subscription Survey 2025. Published 2025. Sample: 2,440 United States adults, online panel, weighted nationally representative. Relied upon: subscribers report $17 per month, equal to $204 per year, on subscriptions they do not use; 80% of United States adults paid for at least one subscription in the twelve months to April 2025.
[2] CardRates. Complicated Return Policies. Published September 2024. Sample: 1,006 United States adults. Relied upon: approximately three in four have kept an item rather than return it; average value forgone among those respondents, $78; approximately 9 returns per year. https://www.cardrates.com/studies/complicated-return-policies/
[3] LendingTree. Buy, Use, Return: Shoppers Admit to "Borrowing" Retail Goods. Published 2025; fielded February 2025 by QuestionPro, nonprobability with quotas. Sample: 1,715 United States consumers who have returned items. Relied upon: 44% have at some point missed the return window for an item they intended to return, of whom 14% frequently and 27% occasionally. This source publishes no dollar figure and states no timeframe. https://www.lendingtree.com/credit-cards/study/retail-returns/
[4] Savings.com. Online Shopping Returns. Published 2026; fielded August 2025. Sample: 1,000 United States adults, aligned to Census benchmarks. Relied upon: 15% were left with unwanted merchandise in the preceding twelve months because the return deadline had expired; average abandoned return worth $50; 27% of non-returners cite an expired or short deadline; approximately 13 online returns per year, rising with income. https://www.savings.com/insights/return-policy-study
[5] Self Financial. The Cost of Unused Paid Subscriptions. Published March 2026. Sample: 1,272 United States adults. Relied upon: $26.79 per month among adults holding at least one unused subscription; 59.9% hold at least one; 70% have forgotten to cancel a free trial at least once, at an average cost of $34.31. https://www.self.inc/info/cost-of-unused-paid-subscriptions/
[6] Omnisend. Package theft report. Published 2025. Method: consumer panel of 1,000 United States respondents blended with Federal Bureau of Investigation larceny statistics. Relied upon: $12.8 billion aggregate annual United States consumer loss net of retailer reimbursement; approximately $101 per household; 29.7% of households affected; 23.5% of affected households reimbursed nothing. https://www.omnisend.com/porch-pirates/
[7] C+R Research. Subscription Service Statistics and Costs. Published 2022. Sample: 1,000 United States adults, itemised recall. Corroboration only: respondents estimated subscription spending at $86 per month against an itemised actual of $219. https://www.crresearch.com/blog/subscription-service-statistics-and-costs/
[8] Chargebacks911. Cardholder Dispute Index 2025. Sample: approximately 1,200 United States and United Kingdom cardholders. Cited at section 7.2 as part of the search record: 84% report that filing a chargeback feels easier than contacting the merchant; 52% skip contacting the seller first. No quantity taken. https://chargebacks911.com/cardholder-dispute-index/
[9] In re: Amazon Return Policy Litigation, No. 2:23-cv-01372 (W.D. Wash.). Proposed settlement filed 23 January 2026: $309.5 million fund, in addition to approximately $570 million previously refunded, for United States consumers who did not receive a refund, received an incorrect or late refund, or were erroneously recharged, between 5 September 2017 and the class data date. Cited as a fact concerning one company; not an input to either figure.
[10] National Retail Federation with Happy Returns, a UPS company. 2025 Retail Returns Landscape. Published October 2025. Method: survey of more than 2,000 United States consumers and more than 350 retail professionals. Relied upon at section 7.7: 15.8% of 2025 sales estimated returned, equal to $849.9 billion; 19.3% of online sales; 2024 comparative 16.9% and $890 billion. https://nrf.com/research/2025-retail-returns-landscape
[11] Security.org. Package Theft Annual Report. Published October 2025. Sample: 3,307 United States adults, three-month recall annualised. $8.2 billion gross of reimbursement; average $222 per package. Located and not used; section 9.5. https://www.security.org/package-theft/annual-report/
[12] Einav, L., Klopack, B., and Mahoney, N. Selling Subscriptions. National Bureau of Economic Research Working Paper 31547, 2023; American Economic Review, 2025. Method: card-transaction records covering approximately 30% of United States subscribers, 2017–2021. Corroboration of mechanism only; no quantity taken. https://www.nber.org/papers/w31547
14.1 Confi maintains this document as a live record and treats an identified error as an amendment to be made rather than a matter to be defended. Where a figure is shown to be wrong it is corrected and the correction recorded at section 12.
14.2 Correspondence regarding this document may be sent to the address published at confi.ai.
End of document. Version 1.0, 11 September 2026.
← Back to founding users · Confi Technologies, Inc., San Francisco Bay Area