We read 1,019 warehouse websites. Half show no sign of any software.
Original data on software adoption in small US warehousing, including the part where we published the number wrong the first time and it was off by a factor of three.
Every vendor in this category will tell you the market is saturated, or that it is wide open, depending on what they are selling you. Nobody shows their working. So we went and looked: 1,019 warehouse and third-party logistics websites across 20 US metro areas, every one of them opened and read, with the text kept so the result could be checked later.
What this does not prove, before the numbers
A website is not an operation. This measures what a warehouse says about itself in public, which is a proxy for what it runs, and a loose one.
- It does not say these warehouses have no software. It says their site shows no sign of any.
- Plenty of operations run a system and never mention it. Nobody puts their WMS on the homepage unless they are selling to brands who care.
- The bias runs one way: this overstates the gap. Read it as a ceiling, not a measurement.
- The sample comes from Google Maps in 20 metros, so it is not a random sample of the country.
- Sites with less than 2,000 characters of readable text are excluded, because you cannot conclude anything from a page that says "coming soon".
The number
Of the 1,019 sites, 895 had substantial text, with a median of 11,710 characters. These are real pages, not fragments. Of those 895, 428 mention no vendor by name and show no technical signal of any inventory system at all. That is 47.8%.
Signal here means the things a warehouse writes when it has software and wants you to know: a client portal, live inventory, EDI, an API, barcode scanning, or the phrase "warehouse management system" itself. Generic WMS mentions appear on 279 sites, a customer portal on 180, live inventory on 170, EDI on 166.
The control, which is the only reason to believe any of this
A count of what is missing is worthless without a count of what is present, because if the detector is broken both numbers move together. Named vendors across the 895: NetSuite 25, Extensiv 15, ShipHero 12, Deposco 4, Cin7 4, 3PL Central 3, Linnworks 3, SkuVault 3.
Those are plausible numbers for this segment. If Extensiv had come back at 268, something would be wrong. Which brings us to the interesting part.
We published this number wrong the first time, and it was wrong by a factor of three
The first run of this reported that 17% of warehouses showed no software, and we wrote it down as the honest rate of the market. It was not. The detector searched for vendor names as plain substrings, and the string "extensiv" appears inside the ordinary English word "extensive", which turns up on the site of nearly every logistics company alive: extensive experience, extensive network, extensive capabilities.
That single missing word boundary tagged 268 sites as already running Extensiv. The real figure is 15. Every one of those false positives was a warehouse we had crossed off as already served.
The corrected figure is 47.8%, not 17%. A one-character bug in a discard filter shrank our view of the market to a third of its size, and it took us four days to notice because a number that says "the market is small" does not feel like a bug. It feels like bad news.
Worse, the first run saved the verdicts and threw away the page text, so there was no way to audit the decision without crawling all 1,019 sites again. Which is what we did. If you run anything like this: keep the raw text. The verdict is the cheap part.
How to check a claim like this, including ours
Any adoption statistic you are shown, including this one, should come with three things, and if it does not, it is a marketing number:
- The denominator, stated plainly. Ours is 895 sites with substantial text, out of 1,019 read.
- A positive control. If someone tells you how many companies lack a thing, ask how many have it and whether that distribution looks sane.
- The method, reproducible. Ours is a script that recalculates every figure in this piece from the saved text. No number here was typed by hand.
The third one matters more than it sounds. We do not write these figures into the page by hand. A script reads the raw data and prints them, and if the data changes the page is wrong until someone reruns it. That is not diligence, it is self defence: we already published one number we had not recalculated.
What it means if you run a warehouse
Mostly that you are not behind. If roughly half the operations in your metro show no system at all, the spreadsheet you are apologising for is the median, not the exception. The useful question is not whether you are late, it is whether the specific thing that goes wrong in your building is the kind that software fixes, and for a lot of small operations the honest answer is not yet.
We wrote separately about where that line falls, and it is lower than anyone selling software will tell you: below a few hundred SKUs a well kept spreadsheet beats most of what you could buy, and we say so on our own pricing page.
Use the data
The counts in this piece are free to reuse with a link back. The limitations above travel with the numbers. We would rather be cited accurately than cited often.
And the disclosure that belongs at the top of anything like this: we sell warehouse software. That is a reason to check our arithmetic, which is why we published the arithmetic.
The bias, stated up front
We sell the software described above, so treat everything here as coming from someone with a stake in it. Several of these notes can be applied in full with a label printer and a weekend, and we’d rather you did that than bought something you don’t need.