When does a cobot actually pay back?
The two published figures we can point at disagree: an automation integrator puts most deployments at 12 to 24 months, while a market analyst reading from April 2022 puts average RoI at 2 to 3 years. Neither is a best case. The answer is application-dependent, and below it runs on inputs you can see and change.
What an hour of operator labour actually costs.
US manufacturing production wage (BLS Current Employment Statistics, series CES3000000008) was $30.37/hr in August 2026 — up 4.4% year on year. That figure understates true displacement cost by the employer-benefit load.
BLS Employer Costs for Employee Compensation (ECEC), Q2 2026, released 2026-09-09, puts manufacturing total compensation at $48.62 per hour worked: $32.50 of wages and salaries plus $16.12 of benefits, which covers payroll tax, insurance, retirement and paid leave.
The two series count different people — ECEC covers all manufacturing workers, CES only production and nonsupervisory staff — so we take the benefit load from ECEC rather than its dollar level: $48.62 ÷ $32.50 = 1.496x. Apply that to the production wage and a fully-burdened operator hour is $45.43. Over a 2,000-hour year that is $90,860 per full-time role. That last figure is our arithmetic, not a BLS publication — every input is above, so you can redo it.
Why arm price is not installed cost.
Our own model, not a published figure. AMD Machines puts a complete installed cobot cell at $75,000-$200,000; the arm prices in the table above run $23,000-$85,000. Those two published ranges put the bare arm at roughly a quarter to a half of the installed cell - a 2x to 4x multiplier. The twelve-category bill of materials below brackets the same span from the other direction.
So a $30K UR5e lands somewhere around $60K-$120K installed, depending on how much tooling, vision and safety hardware the cell needs. Most first-time buyers see only the arm price. Payback math built on the arm-price denominator understates the investment by 2x to 4x, which is the single most common reason a cobot business case misses.
Published ranges from integrator data.
Transcribed from AMD Machines' published payback scenarios (an automation integrator, March 2026). Quality inspection has the tightest band of the four it covers, and the slowest payback, because the case rests on quality-escape avoidance rather than labour. Palletizing is fastest. Welding carries no figure: that source does not cover it, and we would rather leave it blank.
Two published figures, three years apart.
AMD Machines, March 2026: “most cobot deployments achieve payback in 12 to 24 months. Some high-utilization applications break even in under 10 months.” This is an integrator writing about cells it has deployed, with per-application figures behind it.
Interact Analysis, April 2022: “we now calculate the average RoI on a cobot is just 2-3 years, down from 5 years.” That is 24 to 36 months — slower than the integrator figure, not faster. It is also four years old, and arm prices and software maturity have both moved since.
Neither is a best case; both are whole-market averages. If you see a cobot payback under a year quoted without a named cell behind it, ask what installed cost it used as the denominator. On the arm price alone, almost any cobot pays back inside twelve months. On the installed cell, very few do, and that gap is where most optimistic payback claims come from.
Installed cell = arm + integration ($25K) + EOAT ($5K) + safety ($8K). Annual labour displaced = operator wage × 1.496 burden × 2,000 hours × FTEs displaced. Payback months = installed cell ÷ (annual labour ÷ 12).
Those three add-ons total $38K and do not scale with the arm, so the calculator models a minimal single-machine cell: no fixturing, no vision, no conveyor, one scanner. That is why the implied-multiplier readout runs below the 2-4x model above on the pricier arms — 1.8x on a $45K arm, 2.7x on a $23K one. We show the multiplier rather than hide it so you can see where your configuration sits. Add fixturing and vision from the bill of materials and it climbs into, then through, that band.
FTEs displaced is the input that moves the answer most, so it is a slider rather than a buried constant. AMD Machines publishes what its cells actually displace: 0.75 to 1.5 FTE for machine tending, 1 to 1.5 for palletizing, and 0.5 to 1 for assembly. A cobot running across two shifts does not displace two operators.
The check we hold the model to: set FTEs to AMD's published displacement and the payback should land in AMD's published band. At 0.75 FTE, the bottom of AMD's machine-tending displacement, a $45K arm in this minimal cell pays back in 14.6 months — the bottom of AMD's 14-22 month machine-tending band. At 1.5 FTE it falls to 7.3 months, which is the territory AMD describes as high-utilisation. The model and the published data agree, which is the point of showing both.