"Boring" robots versus flashy videos: why you start from the task when choosing a robot and which four numbers to ask the vendor for

Short answer: Ron Zukerman, innovation director at Vishay Precision Group (VPG), a company that makes sensors and measurement systems, wrote a column for The Robot Report on October 11, 2026. His idea: choosing a robot should start from a task that makes money, rather than building an impressive platform and then looking for where to use it. He proposes judging a robot not by videos but by four numbers: how often a person has to intervene, how much a unit of work costs, how long it takes to reconfigure for a new task, and how many safety incidents occur per hour of operation. This is the opinion of one author, and his company has its own interest: it sells the sensors he writes about. But the questions he suggests asking are useful to anyone thinking about a robot for their own production floor or warehouse.

Who is writing and why it matters

The author of the column works for a company that sells sensors and measurement systems, and in the text he repeatedly advises building a specific type of sensor, strain gauges, into the robot's design from the start. This does not make his arguments wrong, but they are worth reading with that in mind. The most concrete number in the column, savings of about $450,000 a year on one assembly line, comes from his own company's tests and is called an estimate in the column. How it was calculated is not disclosed in the text.

The main idea: task first, then the robot's form

Zukerman writes that the industry often first builds an outstanding platform and then looks for a task for it at industrial sites. He proposes the reverse order:

1. Find a task that makes money.
2. Assess whether a robot can do it better than a person.
3. Only then determine what mechanics, sensors and form are needed for it.

To show why the order matters, the author takes the same bipedal humanoid robot and two tasks:

• Inserting a pin connector on a car assembly line. According to him, the clearance there is usually about 20 micrometers and the mating force about 5 newtons, so the robot must detect changes in force smaller than 0.1 newton and notice jamming before the thin pin bends.
• Packing goods onto a pallet in a warehouse. Here, according to the author, such fine sensitivity is almost unnecessary. What is needed is payload capacity, structural rigidity and the ability to withstand a million cycles without gradual drift in the mechanics' settings.

The author's conclusion: a "universal" middle-of-the-road sensor may turn out not precise enough for the first task and not durable enough for the second.

Humanoid or specialized robot

The author divides sites into two kinds. New ones, built from scratch, can be designed for the task from the start. Existing ones, like an old refinery or chemical plant, are built for people: narrow aisles, stairs, grated floors, doorways. In his words, the real question is not what the robot should look like, but what is cheaper to change: the robot or the site.

• At a new site, specialized robots that do not look like people are, in the author's view, almost always better in cost, speed and energy use. As an example he cites Equinor's Northern Lights site, which could be built around the task from the start.
• At an existing site, a form adapted to human aisles and stairs becomes a major advantage: it avoids spending millions of dollars on rebuilding the facilities.

The test he proposes for any humanoid robot: if you were starting the project over, would you choose a humanoid form again? And does the robot really need to fit into a space made for people?

A video is not a deployment

Zukerman writes that flashy videos show what is technically possible, but the real test is a second order from the same buyer: only a repeat purchase proves business value. In his words, a demonstration of a robot calibrated in a room with constant temperature proves only that the robot can do the task once. And to work as a whole fleet, according to the author, a robot must withstand at least 10,000 cycles per shift amid constant temperature changes and with almost no drift.

The reasons deployments fail, as the author names them: gradual drift in readings, sensor wear and difficulties with recalibration. Not errors in the robot's "reasoning."

Another of his remarks: software can be multiplied by copying, while hardware has to be bought, built and waited for. Heat, power use, mechanical wear and safety requirements have to be accounted for all the time.

Four numbers worth asking about

This is the main practical part of the column. The author proposes looking at the measures that matter to the finance department:

Measure	What it means	Why it matters, according to the author
Mean time between human interventions	how long the robot works before a person has to come over and fix something	the most honest measure and at the same time the least often published; if a person comes over every 20 minutes to pull a stuck part out of the gripper or reset a drifted sensor, high autonomy figures mean little
Cost per unit of work	how much one picked item, one meter traveled or one inspection costs	it determines whether the robot will replace people or help them
Reconfiguration time	how long it takes to move the robot to a new task	shows whether a "general-purpose" platform really retrains quickly or whether weeks of engineering work are needed every time
Safety incidents per hour of operation	how many dangerous events per hour of operation	it determines whether the robot will come out from behind the fence and be able to work next to people

How to apply this yourself

What follows is my advice, not data from the column. It is for an owner or manager of a production site or warehouse who is considering buying a robot. I do not cover renting here: it has different economics and needs to be calculated separately under the terms of the contract.

1. Describe the task before talking to the vendor. What exactly the person does, how many times per shift, what the defect rate is now, how much an hour of work costs.
   • Who does it: the site manager together with the process engineer.
   • How to check: a written description that includes the number of operations per shift and the current defect rate, backed by records for the last month.
2. Ask for the four numbers from the table above, with the measurement conditions. What matters is not only the number but where it was obtained: at a live site, for how many weeks, at what temperature.
   • Who does it: whoever runs the purchase.
   • How to check: the vendor's answer in writing; if some number is missing, that is also an answer.
3. Test at your own site, not by the video. Ask for a trial run on your task.
   • Who does it: the site manager.
   • How to check: a trial log in which every human intervention and the time it happened are recorded; the mean time between interventions is calculated from it.
   • Safety: for the trial, the robot is placed behind a fence or in a separate zone until the vendor shows safety data and until your occupational safety specialist approves work next to people.
4. Calculate the payback. First, net savings per month = the reduction in staff costs that will actually happen after deployment (cut positions or people not hired, not hours that were simply freed up) + the reduction in defect losses − maintenance − pay for people's time spent on interventions − losses from robot downtime − other costs (electricity, spare parts, insurance, space rent). If net savings are zero or less, the robot does not pay off, and the formula below cannot be applied. If they are greater than zero: payback period in months = (robot price + deployment) / net savings per month. An example with hypothetical numbers, not from the column: the robot with deployment costs 6 million rubles, after deployment staff costs actually go down by 300,000 rubles a month and defect losses by 100,000 rubles, maintenance is 60,000 rubles a month, and for simplicity there is no downtime and no other costs. If a person comes over to it on average once an hour, for 5 minutes, across two 8-hour shifts, that is about 1.3 hours a day, roughly 29 hours a month with 22 working days. At 600 rubles an hour, that is about 17,600 rubles. Net savings are about 322,000 rubles a month, payback about 19 months. If interventions happen every 20 minutes, there are already three times as many, and the payback period grows; if on top of that the robot sits idle while waiting for a person, savings fall even more.
   • Who does it: the finance manager together with the site manager.
   • How to check: plug numbers from the trial log into the formula, not numbers from marketing materials.

When this does not concern you

If you do not have a repetitive physical operation that is done many times per shift, a robot is most likely not the first thing worth automating. This is my estimate, not a threshold from the column: how many repetitions are enough can only be shown by the payback calculation above, using your own numbers.

Summary

Ron Zukerman's column in The Robot Report proposes a simple order: first a task that makes money, then the robot's form, and judge it by operating numbers rather than videos. This is the opinion of an employee of a company that sells sensors, and for its only concrete savings example in the column, the method of calculation is not disclosed. But the four questions for the vendor, about human interventions, cost per unit of work, reconfiguration for a new task and safety, are worth asking in any case, and the payback is worth calculating from the trial log at your own site.

I work on AI agents and automation. If you want to look at my projects or discuss your own task, visit my portfolio.

Sources

• The Robot Report, Ron Zukerman, 11.10.2026: Will 'boring' robots win in the real world? Why robotic form should follow workflow. https://www.therobotreport.com/will-boring-robots-win-in-the-real-world-why-robotic-form-should-follow-workflow/