Compute has a unit of work. A token, a FLOP — a defined quantum of processing that lets you set any two chips, or any two models, side by side and say which does more per dollar. The unit is what makes the comparison honest. It is why a benchmark means something and why "faster" is a claim you can check.
Robots have no such unit. Physical work does not tokenize: moving a tote is not the same quantum of work as welding a seam is not the same as picking an item off a shelf, and no conversion turns one into another. This is not a gap someone forgot to close. It is a structural absence — and the absence is not neutral. It is exactly where the industry's productivity claims live.
Three metrics, none of which compose
Walk through what actually circulates, and you find three different measures of robot "work," each native to a different world.
The first is cycle time and throughput — units per hour, seconds per operation. It is the manufacturing floor's own language, precise and real, but comparable only within one task on one line. A throughput number for palletizing tells you nothing about a picking arm.
The second is dollars per hour of labor substituted — the framing behind robotics-as-a-service pricing, which sells a machine as a cheaper worker. It converts the robot into a wage, which is legible to a CFO but says nothing about how much the machine actually does; two robots at the same hourly rate can do wildly different amounts of work.
The third is task-success rate on a benchmark suite — the academic measure, a percentage of trials completed on a standardized set of tasks. It is rigorous within a lab and has essentially zero economic content: a success rate does not price out, and it does not survive being moved to a different embodiment or a different bench.
Cycle time, dollars per hour, success rate. Three numbers, three worlds, and none of them converts into either of the others. There is no exchange rate. A robot described as "3x" on one and "10x" on another has not been measured twice; it has been measured on two things that do not share a scale.
No standards body ever built the missing unit — and not by oversight
You might expect the standards apparatus to have closed this. It has not, and the shape of the not-closing is informative.
ISO's robotics committee, TC 299, runs sixteen working groups. They cover vocabulary, safety across service and industrial and medical and mobile robots, modularity, management systems, biomechanical data, electrical interfaces, mobile-robot interoperability, energy consumption, and humanoid datasets. Read the whole roster and not one group targets an economic productivity metric or a cross-vendor unit of work. The nearest by name — the working group on robot performance — is scoped to the technical performance of service robots, not the economic output of industrial ones.
The one performance standard that does exist for industrial robots, ISO 9283, measures pose accuracy, repeatability, path accuracy, cycle time as a matter of motion timing, velocity, and static compliance. It is motion metrology end to end — how precisely and how fast the arm moves through space — with no dollars-per-hour, no task-success rate, and no cross-task measure of productivity anywhere in it.
Read together, the two facts say the same thing from two directions: the standards world has never attempted an economic unit of work for robots. Every existing working group and the one performance standard on the books is structurally scoped somewhere else. We report that no such standard exists. We do not claim to know why, and we draw no inference about anyone's motive from its absence.
Into the vacuum: four "productivity" multiples, no metric named
Absent a shared unit, "productivity" becomes a word a vendor can attach a number to without ever saying what was counted. Two named vendors, from archived public marketing, show the pattern.
Zebra Technologies, announcing its Fetch Robotics line, headlines that customers "increase productivity by 2.5x, double throughput and reclaim 15% of their space." Productivity at 2.5x and throughput at 2x are presented as two separate figures — with no stated basis for either, and no explanation of how a "productivity" that is distinct from throughput was measured. The same materials say Fetch's FlexShelf "increases productivity up to three times by augmenting workers with robots that are 60% faster and carry larger and heavier payloads" — a productivity multiple pinned loosely to speed and payload specs, still with no defined unit of work underneath it.
inVia Robotics, on a single marketing page, adds two more. Its automation will "boost your productivity 2-3x ... and 5x when you add inVia Picker autonomous mobile robots." A Gnarlywood case "reduces labor cost by 65% and boosts productivity 6x." Two-to-three-x, five-x, six-x — and in none of them is the productivity metric named.
The tell is on the very same page. Two case studies do name the metric — and both name the same one. SICK, using inVia's PickerWall, "boosts pick rates ... 10x." And a Scholastic Canada case reports a "300% increase in pick rates at unit level," with the customer's own words describing how they used existing floor space to "triple our pick rates." Pick rates — a specific, checkable quantity. The same vendor names the unit precisely when a concrete metric is flattering and defensible, and drops it to a generic "productivity" everywhere else. That contrast is the whole argument on one page: the vocabulary exists; it is used selectively.
Money already changes hands against the undefined unit
The sharpest version is not a claim in an ad — it is a price. inVia sells its robotics as a subscription, and the page pegs the fee to the same undefined word: customers "pay only for what matters most: warehouse productivity," and "pay just for the productivity our automation delivers." A subscription fee is pegged to a term the same page never defines. The unit of work is missing precisely where a buyer is being asked to pay by it.
Why this is the mechanism, not another vaporware case
The Robot Ledger has documented robots announced but never shipped, and prices that appeared once and vanished. This is a level beneath those. Those were about whether the thing or its price was real. This is about whether the measure is real — and the finding is that even when a productivity number stands behind an actual invoice, there is no shared unit that makes it comparable across vendors or checkable by the buyer. A 6x from one warehouse vendor and a 3x from another are not two points on a scale. They are two different things wearing the same word.
We are not saying these figures are wrong or inflated; we have no evidence either way, and that is exactly the point — without a named metric there is nothing to check against. The SICK "pick rates" line proves the metric can be stated. Mostly it is not.
Until robotics has a unit of work — and no one has built one — "productivity" is a marketing term with a number attached, and a buyer has no way to price what it names. Other publications count the robots. Our franchise is to price them, and a thing with no unit of work cannot be priced. That is not a limitation we are conceding. It is the story.
The Robot Ledger is published by Pexara AI LLC. Figures may be cited with attribution; underlying archive available for verification inquiries.