By Sarah Xiao 
The question arrives in almost the same words every time: should this line have a robot on it? And nine times out of ten the arithmetic behind the answer is identical — how many operators does it replace, what does the robot cost, divide, and see how many years to payback.
That calculation is not wrong. It is just incomplete in a way that reliably produces the wrong decision, because it cannot explain the thing everyone has seen: two factories buy comparable cells in the same quarter, and one is running three months later while the other is still waiting for “the right process” two years on.
The difference is rarely the equipment. It is which sums were done.
Sum one: lifecycle, not purchase price
A six-axis robot body might be RMB 150,000–200,000. That is the entry ticket, not the cost. What actually lands on the capital request is:
- The cell: robot, controller, teach pendant.
- Integration: fixture, end effector, safety fencing, infeed and outfeed.
- Programming and commissioning: teaching, offline programming, on-site integration — during which the line is not producing.
- Running costs: spares, lubrication, recalibration, power, service contract.
- The hidden line: rework when the process does not match, downtime at changeover, and the queue that forms while nobody on site can fix a fault.
Add these up and the true figure is routinely two to three times the body price. The number that decides the project is not what you spend; it is whether the cell returns that spend through takt time and yield, reliably, for years.
Which suggests a better unit of comparison than payback period: cost per productive hour. Take the full lifecycle cost, divide by the hours the cell actually works over its life — not the hours it exists — and compare that with the loaded labour cost per hour plus whatever throughput gain the cell delivers. Two projects with the same payback can have very different risk once you do this, because one assumes 6,000 productive hours a year and the other assumes 2,000.
Sum two: stability, not the demo
A demonstration runs in ideal conditions. Your line has incoming variation, dust, temperature drift, and workpieces that sit half a millimetre off location.
A robot that trips faults and stops is more expensive than the operator it replaced, because when it stops the whole line stops, and someone has to stand there and clear it. So the second sum has three variables:
- Yield. Does the cell hold tolerance across the shift, or does someone rework the output?
- Availability. What fraction of the day is it actually working, after alarms, waiting and changeover?
- Flexibility. How often does your product change, and how long does a changeover take? Frequent changeover eats the advantage that justified the cell.
The general rule: simple process, large batch, stable incoming quality — robots win easily. Complex process, small batch, variable quality — the robot becomes expensive furniture. Test on your actual incoming parts and your actual takt, not on the integrator’s sample.
Sum three: people, not equipment
This is the sum most often skipped entirely, and it is the one that kills projects after the invoice is paid.
Who runs it? Who programs it? Who reads the alarm log? Who repairs it? Industrial robotics has a persistent skills gap in exactly these roles — operators, programmers, maintenance technicians. Which means the realistic constraint is not “can we afford the robot” but “can we staff it.”
Three questions:
- Can you hire? In your area, what does it cost to recruit someone who can operate and do simple programming?
- Can your existing team make the move? Will the operators who currently do the job by hand adapt to supervising a screen and adjusting programs?
- What happens when the one person who can set it up leaves? If the answer is “the line stops”, you have built a single point of failure into a capital asset.
Equipment can be bought in a quarter. The people who can run, repair and improve it take longer.
A screening sheet
Before the detailed business case, score the candidate line 1–5 on each row. Anything scoring 1 or 2 on two or more rows is a project that will disappoint.
| Criterion | Score 5 looks like | Score 1 looks like |
|---|---|---|
| Repetition | Same motion thousands of times per shift | Different every cycle |
| Batch size | Long runs, stable product | One-offs, frequent changeover |
| Incoming consistency | Fixtures locate within tolerance | Part variation exceeds the gripper’s compliance |
| Environment | Dirty, hot, heavy, unsafe for people | Clean, light, comfortable |
| Staffing | Hard to hire; high turnover | Easy to hire; stable team |
| Payback on lifecycle cost | Under two years | Beyond the product’s life |
Then apply three kill criteria. Do not proceed if: the lifecycle cost is more than three times the body price and the payback still exceeds the programme; the process is not yet stable enough that a human does it the same way twice; or nobody on site can be nominated as the owner of the cell before it is ordered.
That last one sounds bureaucratic. It is the single best predictor of whether the cell is running in a year.
The filter, not the verdict
There is a category of process where robots are close to automatic: high repetition, heavy or hazardous, large batch, tight takt, hard to staff. Conveying, palletising, welding, spraying, machine tending.
And a category where they are usually a mistake: single-piece or small-batch work with frequent changeover, processes not yet defined, incoming variation the cell cannot sense, and sites with nobody to program or maintain it.
The purpose of the three sums is not to produce a number. It is to move the decision from “is this advanced” to “is this appropriate” — and to do it before the purchase order, rather than two years after.
Related reading
- Robotics shutdowns 2026: funding is not delivery capability — what happens when a supplier’s process was never under control.
- LV switchgear auxiliary material takeoff: per-panel rules — the same discipline applied to a bill of materials.
- Robot component sourcing: from prototype to production — qualifying the vendor before the capital is committed.
Sources: the three-account framing, the lifecycle multiplier and the fit/no-fit lists follow 智造日记 (12 October 2026). The cost-per-productive-hour method, the screening sheet and the three kill criteria are Latest Robot News editorial.