Proof of delivery is treated as settled evidence in customer disputes, marketplace appeals and payment claims. It is not. A delivery scan proves that a carrier recorded an event at a point in time, not that the intended person received the intended parcel. The gap between those two statements is where merchant losses on non-receipt sit.
What does proof of delivery actually mean in practice?
Proof of delivery, or POD, is the record a carrier creates when it considers a parcel handed over. That might be a signature, a photograph, coordinates, a scan into a controlled location, or a courier pressing a button on a handheld device. All surface to the buyer as the same word: delivered.
The distinction that matters is between a status and an artefact. A status is a claim made by one party; an artefact is something a third party can inspect afterwards. Disputes are lost when the merchant holds a status where it needed an artefact.
| POD form | What is captured | What it genuinely evidences | Where it breaks down |
|---|---|---|---|
| Status update only | A timestamped event code | A courier marked the stop complete | No artefact. Cannot separate premature scan, misdelivery and theft |
| Geolocation on the scan | Coordinates at scan time | The device was near an address | Accuracy varies, and proximity says nothing about who took the parcel |
| Photograph at the drop point | An image of the parcel in place | A parcel existed at a location at a time | Silent on what happened next, and on whether the address is right |
| Recipient signature | A name and mark on device | A person present accepted the parcel | Often illegible, signed by any adult present, sometimes entered by the courier |
| Access-controlled handover | A scan into a locker, store or mailroom, plus collection | Custody continuing past the doorstep | Custody may end at a mailroom, not with the buyer |
| Carrier-issued POD document | A consolidated record | A citable document a marketplace or bank will read | Content, availability and retention vary by carrier |
Practical read: the weakest forms are the most common, and the strongest are the least universally available.
Why does "Delivered" mean different things on different networks?
Because there is no cross-carrier standard for tracking event codes. Universal Postal Union (UPU) standards govern the postal network only: S10 defines the 13-character identifier for international items, and M40, known as EMSEVT V3, defines item-level tracking events. UPS, FedEx, DHL and Amazon operate proprietary code sets outside that framework.
So a merchant comparing carrier performance is comparing labels, not evidence. Two parcels can both read delivered when one is backed by a photograph and coordinates and the other by a keypress at the end of a route, and nothing in the feed distinguishes them.
Delivery-location data is also treated as sensitive by the platforms that surface it. Google's Package Tracking Early Adopters Program documentation, last updated July 14, 2026, explicitly refuses any personal or geographic data about sender or recipient. Merchants should not expect richer location evidence to become freely portable.
What this means in a dispute: you cannot argue from the status alone. You argue from what your carrier captured on that parcel, which means knowing in advance which carriers capture what.
What does the data actually tell us about parcels that never arrive?
Less than the volume of published statistics suggests. No carrier or regulator publishes a rate for parcels marked delivered that never reach the buyer, so every circulating figure comes from consumer self-report, which cannot separate theft, misdelivery, a premature scan or a parcel left somewhere unexpected.
The most transparently documented estimate is Security.org's 2025 annual package theft report, based on 3,307 US adults surveyed in October 2025 and census-weighted on age, sex and ethnicity.
- Incidence. One in four Americans, roughly 64 million people, have had a package stolen at some point. Three per cent, roughly nine million adults, reported a theft in the previous three months.
- Scale and value. An estimated 37 million packages stolen per year, worth roughly USD 8.2 billion, at an average value of USD 222, up around 10 per cent year on year.
- Housing type. Apartment and condominium residents are victimised 3.5 times more often than single-family homeowners.
- Who hears about it. Fewer than one in four report the theft to police, while 64 per cent report to the retailer and 62 per cent to the carrier.
That last point is the operationally important one. The merchant is the primary reporting channel for a loss it did not cause and cannot inspect, while the police report, the only independent record, is the least likely to exist.
Demand-side effects show up too. Narvar's State of Post-Purchase 2025, a vendor-run study of 3,461 US online shoppers published on November 6, 2025, found 41 per cent saying they have had a package stolen, 40 per cent abandoning a purchase over fear of theft, and 74 per cent experiencing a late delivery in the past year. That 41 per cent and Security.org's one in four answer similar questions with materially different results, so treat the direction as informative and the level as uncertain.
Why do the theft estimates disagree by so much?
Because the published 2025 estimates are irreconcilable, and the market rarely says so. Set them side by side.
| Source | Packages stolen | Value |
|---|---|---|
| Security.org, 2025 annual report | 37 million per year | USD 8.2 billion |
| Omnisend, 2025 | 228 million | USD 12.8 billion |
| SafeWise, 2025 | Not stated in these terms | USD 37 billion in total losses |
These differ by roughly four to six times. They cannot all be approximately right, and no reconciliation has been published.
One methodological point, stated carefully. We located Security.org's sample size, weighting and field dates, and could not locate equivalents for the other two. That is a statement about what is findable in public, not a claim that no methodology exists, and it is why this article uses Security.org as its base case.
The honest conclusion: anyone quoting a single confident package theft number is quoting one estimate from a set that disagrees by a factor of several. Build loss assumptions from your own claim data.
What do marketplaces actually require, and how do they use scans?
Marketplace performance systems often rely heavily on compliant tracking events rather than independently determining what physically happened to a parcel. They then attach commercial consequences to those recorded events and to aggregate performance.
- eBay protects seller metrics only where valid tracking with an acceptance scan is uploaded within the stated handling time. The acceptance scan, not the delivery scan, is the gate.
- Amazon applies On-Time Delivery Rate to seller-fulfilled orders, and from February 28, 2026 a rate below 90 per cent triggers deactivation at listing level, removing the worst-contributing ASINs rather than the whole catalogue. The penalty lands on specific products, making carrier performance a merchandising problem.
- Walmart Marketplace requires 90 per cent or better on-time delivery under its Seller Performance Standards.
The pattern is consistent. Platforms measure the scan record, so the scan record becomes the product, and a seller whose parcels arrive but whose carrier scans inconsistently is penalised identically to one whose parcels are late.
On payment disputes, be careful. Card-scheme rules change and vary by scheme and region, so confirm current requirements with your acquirer. What is safe to say is that a file built on a bare status line is weaker than one built on a retrievable carrier document, and that what you can produce was fixed months earlier by carrier choice and data retention.
Why is the fastest-growing carrier segment also the weakest on proof?
Because growth in US parcel volume is concentrated in networks built for cost, not for evidence. The Pitney Bowes Parcel Shipping Index 2026, US edition, reports that alternative carriers including OnTrac, GLS, Veho, UniUni and SpeedX grew 127 per cent to 1.8 billion parcels, doubling revenue share from 3.4 to 7.2 per cent on Shein and Temu volume. Pitney Bowes explicitly identifies tracking and reliability as this segment's weak point.
That is the problem in one sentence. POD quality varies most in the segment growing fastest, so merchants moving volume there are trading evidentiary strength for rate without pricing the trade.
The answer is not to avoid those networks, but to treat POD capability as a line item in the carrier decision and reserve the weaker ones for order values where the dispute cost is tolerable.
How should a merchant build a defensible evidence trail?
Work backwards from what you would need to produce if the order were challenged in six months.
- Audit what each carrier actually captures, per service level, not what the tracking page displays: signature, photograph, coordinates or none, and how long each is retained.
- Store the raw event data, not the display status. Retain carrier codes, timestamps and any media reference at the point of receipt. Webhook-driven capture through a tracking API is the practical route, and one reason merchants run parcel data through a platform such as Ship24.
- Set POD requirements by order value. Define a threshold above which a signature or access-controlled delivery is mandatory, enforced at label generation rather than at claim time.
- Prefer custody continuation over doorstep completion. Lockers and store collection extend custody past the point where doorstep evidence goes blind, and address the apartment exposure Security.org identifies.
- Instrument the acceptance scan. Marketplace protection often turns on it, and unlike the delivery scan it is within your control.
- Log claims by carrier, service and postcode. Your own claim rate by lane beats any published theft figure.
One caution. None of this converts a probabilistic record into certainty. The goal is to move from a claim to an artefact, and to know which orders will only ever have a claim.
So what is proof of delivery worth?
It is worth exactly as much as the artefact behind the status, and no more. Treated as a binary, delivered or not, POD misleads. Treated as a spectrum, from a bare keypress to a documented collection, it becomes something an operations team can manage and price.
Our judgement is that the industry has the emphasis backwards. Considerable attention goes to how tracking looks to the customer and little to whether the record would survive scrutiny, while the estimates used to size the problem disagree by a factor of several. Both point the same way: reason from your own carrier mix.
The practical position for 2026 is unglamorous. Know what each carrier captures, keep the raw record, escalate evidence requirements with order value, and budget for the portion of non-receipt that is simply unresolvable. Proof of delivery proves a carrier recorded an event, and everything else is inference.
Sources & methodology
- Security.org Package Theft Annual Report 2025 Consumer research with published sample and weighting.
- Omnisend U.S. Porch Piracy Report 2025 Company research combining survey and FBI data.
- SafeWise 2025 U.S. Package Theft Report Company research.
- Universal Postal Union Standards programme Primary standards source.
- Google Search Central Package Tracking documentation Primary documentation.
- eBay Expected delivery dates and tracking guidance Primary marketplace guidance.
- Walmart Marketplace Pro Seller performance standards Primary marketplace guidance.
- Pitney Bowes Parcel Shipping Index 2026 Industry research.
- Narvar State of Post-Purchase 2025 Company research with disclosed sample.
