A million warehouse robots, and what happened to injury rates

The two sets of figures come from parties with opposite interests, and both are worth reading for that reason. The fleet numbers are Amazon's own disclosures in its own announcements. The injury figures come from a Senate committee majority staff investigation, which is an adversarial document. Neither is neutral, and the honest thing is to attribute each rather than blend them.

The comparison group matters more than the headline gap. Amazon's 2023 rate of 6.54 sits against 5.4 for large warehouses and 4.8 for all warehouses, so the size of the gap depends on which comparison you pick. Large warehouses are the fairer benchmark, and they are also the tougher one to look good against.

The trend and the level point different ways. The rate fell from 9.01 in 2019 to 6.54 in 2023 while the robot fleet grew toward a million, but it stayed above the comparison group throughout. A reader can honestly say the rate improved, and can honestly say it remained higher than peers, and both from the same series.

Nothing here establishes what robots did to the injury rate. Amazon changed enormously over the period in headcount, facility design, product mix and scrutiny. A side-by-side shows two things moving; it does not isolate one cause.

Attribute each figure to its source: the fleet numbers to Amazon, the injury rates to the Senate committee investigation.

Name the comparison group with the injury rate. 6.54 against 5.4 at large warehouses is a different claim from 6.54 against 4.8 for all warehouses.

Give the trend and the level together. The rate fell from 9.01 in 2019 to 6.54 in 2023 and remained above the comparison in every year examined.

Avoid causal language. The data shows a fleet growing and an injury rate falling but staying elevated; it does not show what caused either.

Presenting the injury figures as neutral government statistics. They come from a congressional committee's majority staff investigation, which is an adversarial document with a position.

Using the whole-warehouse average as the comparison when discussing a large operator. Large warehouses are the like-for-like group.

Claiming robots caused the injury rate to rise or fall. The period contains too many simultaneous changes for the comparison to isolate automation.

Quoting the fleet size as a measure of automation intensity per worker. Amazon also grew its workforce substantially over the same period.

Generalising the Shreveport staffing figure. Amazon gives it for one next-generation facility, not for its network or for advanced robotics as a whole.

The fleet figures come from Amazon's own published announcements, retrieved and confirmed. The injury figures come from the Senate committee investigation, retrieved as a PDF with a text layer and confirmed to contain each rate quoted, including the year-by-year comparison series.

We deliberately used the company's own numbers for its own fleet and an external investigation for injuries, and we label both. Using a company disclosure for a safety claim, or an adversarial investigation for a fleet count, would be the wrong way round.

We present the trend and the level together because the series supports two true statements that sound opposed, and reporting only one of them would be selective.

We make no causal claim. We looked for a study isolating the effect of warehouse automation on injury rates and did not find one that could be verified; what exists is a fleet count and an injury series covering the same years.