Ask what a robot costs and you get a machine price. Ask what it costs to have one doing useful work and the figure is several times higher — and the difference is where budgets run out at the least convenient moment.
The one-time costs
The machine. The quoted number, and the smallest line in most deployments once everything else is counted.
Content and configuration. Loading your information, writing scripts, setting boundaries, testing. For any robot that interacts with people, this determines the quality customers experience and it is the most commonly under-budgeted item.
Integration. Connecting to existing systems, machines or processes. Where relevant, frequently the largest single line.
Site preparation. Power, charging position, network, floor work, sometimes rearranging a space.
Mapping and setup. For mobile robots, building the map and defining routes.
Training. Delivery cost plus the time of people attending.
Transport and installation.
Initial spares. Consumables and any long-lead item worth holding.
A useful planning ratio: expect the machine to be a minority of the delivered cost. Budgets built on the quoted price stall part way through with predictable regularity.
The recurring costs
The part that continues after the project closes.
Content maintenance. For interactive robots, information changes and must be updated. This is a monthly commitment, not a one-time task.
Consumables. Brushes, pads, filters, cups, gripper fingers — whatever the application wears through. Small individually, continuous in aggregate.
Maintenance and service. Scheduled work plus repairs.
Spares holding. Capital sitting on a shelf, which is a cost even though it is not an expense.
Software and licensing. Some products charge periodically for the AI service or the management platform.
Power. Modest but present.
The owner's time. Typically a few hours a week for a service robot. Not an invoice and unquestionably a cost.
Battery replacement. After a few years. Ask the price at purchase, because it is a significant expense arriving in year two or three and it should be in the plan from the start.
The costs that appear on no quotation
Real expenses that no supplier lists.
Lost output during installation and commissioning. Particularly where the robot is being fitted into a running operation.
Internal engineering time. Your own people, in meetings, answering questions, solving problems. Genuine cost even without an invoice.
The learning period. Weeks where everything is slower while people adjust.
Process changes. If the operation must change to suit the robot, that change has a cost.
Space. Floor area the robot and its charging position occupy, valued at whatever that area is worth.
Opportunity cost of attention. Management time spent on this rather than something else.
Contingency. A fifth to a quarter of the estimate is normal on a first project. Organisations with several completed projects can work with less; a first one encounters more unknowns than the estimate anticipates.
Reducing it to a usable number
A calculation that produces something comparable.
One. Total the one-time costs and divide by the expected useful life. Three to five years is a reasonable assumption for service robots; longer for industrial arms.
Two. Add the annual recurring costs.
Three. Convert the owner's time into money and add it.
Four. That gives annual total cost.
Five. Divide by expected annual output — interactions, deliveries, units, whatever the application produces.
The result is cost per unit of output, and it is the only figure that compares meaningfully against the current method, against alternative equipment, and against doing nothing.
What it usually reveals. Applications with high volume look far better than the purchase price suggested; applications where the robot sits idle look far worse. And a high figure caused by low volume points at placement or task selection rather than at the machine.
Reducing cost sensibly
Where savings are real and where they are false economies.
Real: rent before buying. A month of rental costs a fraction of a wrong purchase and answers the question with real experience.
Real: buy the appropriate form factor. A wheeled robot with a screen delivers most of the value of a humanoid at a fraction of the cost for most applications.
Real: prepare content yourself. Time-consuming, considerably cheaper, and it produces better understanding of the system.
Real: choose products with locally available spares. This affects year-two costs more than most people anticipate.
Real: reduce task variation. Often eliminates the need for expensive sensing entirely.
False: cutting factory testing. The time reappears as extended commissioning at your site, where it costs more.
False: minimal training. Reappears as poor availability and eventual abandonment.
False: skipping documentation. Reappears at every future change and at every staff turnover.
False: grey-market purchase for equipment you depend on. The price difference is usually smaller than the risk of no local support.
Frequently asked questions
How does the machine price relate to the total?
It is typically a minority of the delivered cost once content, integration, site preparation, training and spares are counted. Budgets built on the quoted price stall part way through with predictable regularity.
Which cost is most under-budgeted?
Content and configuration for any robot that interacts with people. It determines the quality customers experience, and skipping it produces a machine that looks right and cannot answer anything specific.
What is the only figure worth comparing?
Cost per unit of output — annual total cost divided by annual interactions, deliveries or units. It compares meaningfully against the current method, alternative equipment, and doing nothing.
Which cost reductions are false economies?
Cutting factory testing, minimal training, skipping documentation, and grey-market purchases for equipment you depend on. Each reappears later at higher cost than was saved.
More in Local ecosystem and Voice and language.
