What 435 print-on-demand orders actually measured

Most writing about print-on-demand quality is guesswork dressed as advice — someone who ordered one mug telling you what a platform is “like.” We had a different option: pull our own fulfilment history and count. This is what 435 print-on-demand orders across our physical storefronts actually measured, read off the fulfilment records on 18 July 2026. Where the data can’t answer a question, we say so rather than guessing.

The sample

435 fulfilled-through-print-on-demand orders across our two physical-product shops. Our digital-download shop is not print-on-demand and does not appear here. This is the production layer’s own record — not the full marketplace order history, and not a survey of anyone else’s experience. One studio, one supply chain, counted honestly.

Orders reach the customer 95% of the time — and the other 4% had to be made twice

Of the 435 orders, 95.4% completed and shipped. 4.1% — eighteen orders — were cancelled in production.

Here is the part most reviews skip. The records mark all eighteen as “unspecified” — a reason field with nothing in it. But we handled every one of these orders ourselves, and we can say first-hand what the empty field cannot — including the surprise: almost none of them were printing errors. The redos were parcels damaged in transit, parcels lost in transit, and wrong items shipped. The failure sellers fear most — a bad print — was nearly absent from the eighteen. That makes 4.1% — about one in twenty-five orders — our measured fulfilment-failure-and-redo rate: not a customer-return rate, not a refund rate, and mostly not a printing problem either. The chain broke after the press, not at it.

Two things follow. First, this is the hidden cost line no platform pricing page shows you — a failed order costs days of delivery time and sometimes the redo itself. Second, if the losses live in packaging, carriers and picking rather than in print quality, then judging a fulfilment partner by print samples alone measures the wrong link. Ask how they pack, who they ship with, and what happens when a parcel disappears — that is where our one-in-twenty-five actually happened.

Production takes about three days — but which facility you draw decides that

Across the orders we could time from placement to shipment, production averaged roughly 74 hours — about three days before a parcel leaves the plant. Total time to delivered averaged 8 days.

Those averages hide the real story. The same catalogue, ordered by the same customers, was produced across six different facilities — and they do not perform alike:

• Our highest-volume facility handled 287 of the orders and was also the slowest: about 94 hours to ship, 8.7 days to delivered.
• A second facility, on 97 orders, shipped in about 41 hours and delivered in 6.7 days — better than half the production time of the busy one.
• A third, on 32 orders, sat in between at 51 hours and 7.7 days.
• A fourth, on 17 orders, was the fastest measured: 37 hours to ship, 5.7 days to delivered.

Read that again: production time ranged from 37 to 94 hours depending on nothing the customer chose. Same product, same design, same price — a two-and-a-half-times difference in how long it took to make, decided entirely by which facility in the network picked up the job.

The lesson: you are not buying from a platform, you are buying from a plant

This is the single most useful thing 435 orders taught us, and it is the thing the platform-comparison articles get wrong. When you route a catalogue through a print network, “the platform” is not one quality level or one speed. It is a set of facilities with genuinely different output, and which one makes any given order is often not yours to choose. The busy default is not always the good one — ours was the slowest of the four we could measure. And a facility does not just print the order: it packs it and hands it to a carrier, which is exactly where our failures lived.

If you sell through a print network, the practical moves follow directly: sample per facility, not per platform; know which provider your bestsellers actually route to; and when a provider lets you pin production to a specific facility, treat that control as worth more than a small base-price saving. Speed and consistency compound over thousands of orders in a way a few cents per unit never will.

What this data cannot tell you

Two honest gaps, so you can weigh the rest. First, the failure classification above comes from having handled those eighteen orders ourselves, not from the records — the reason field was empty on every one, which tells you something about how much diagnostic data a seller gets from a print network by default. We can name the failure categories first-hand — transit damage, transit loss, a wrong item shipped, and almost never a misprint — but not the exact count of each, because we did not log them as they happened; the per-category tally starts now, not retroactively. Second, customer-side quality complaints — the ones that arrive as messages and reviews rather than production cancellations — live in a different system and have not been counted rigorously enough to publish. When they have been, they will get their own page — not a paragraph of guesswork here.

The operational lesson underneath both gaps: count as you go. Reconstructing categories from an empty reason field a year later is not measurement, and we will not pretend otherwise.

Everything above is a count from our own records, taken on one day. The numbers will drift as the catalogue grows; the method won’t. If a claim about print-on-demand can’t be traced to a number someone actually counted, it is worth exactly what it cost to write.