We want to make the world better with AI.
Starting with the work nobody wants.
Lifting, palletising, picking, the night shift on a machine that runs unattended. Robotics in the mid-market is not about replacing people; there are not enough people to replace. It is about the shifts that go unfilled and the orders that get declined because of them.
Most robotics business cases are written to pass, not to hold.
The standard calculation compares a robot against a shift of labour and stops there. It leaves out the gripper development, the fixtures, the safety assessment, the six weeks of commissioning during which the line runs worse, and the fact that at small batch sizes the changeover time eats a third of the gain.
We put those in the model before you decide, because a cell that pays back in three years on paper and never on the shop floor damages the case for the next one too.
From feasibility to a cell that still runs in year three.
Feasibility and simulation
We simulate the cell against your real part mix and cycle times before anyone quotes hardware. Digital twins are cheap; discovering that the gripper cannot handle variant seven is not.
Vision and bin picking
Machine vision for quality inspection and for picking parts out of unstructured bins. This is where AI has changed what is possible in the last three years, and where most of the new cases come from.
Cobots and safe collaboration
Where the cell shares space with people, the safety assessment shapes the layout from the first sketch rather than after. Speed and separation monitoring done properly is what keeps the cycle time you budgeted for.
Retrofit before replacement
Older machines can often be instrumented and integrated rather than replaced. Sensors, an edge box and a connection into the planning system frequently deliver more per euro than a new asset does.
Foundation models moved the line between programmed and taught.
Classic industrial robotics needs the world to hold still: fixed positions, fixed parts, fixed lighting. Every deviation is a programming task. Vision models trained on physical interaction shift part of that burden from programming to demonstration, which is what makes small batch sizes viable at all.
This is genuinely new and it is also over-promised. Humanoids are not going to run your line next year. Bin picking on an unsorted tote, quality inspection with a part mix that changes weekly, and a cell that recovers from a fault without an engineer walking over: those work now, and they are worth building.
Questions
What batch size does a robot start making sense at?
There is no single number, which is why we simulate. Roughly: below a few hundred parts per variant, changeover dominates and the case depends on how much of it can be automated. Above that, the classic calculations start to hold.
Buy or lease?
Lease or robotics-as-a-service is usually right for a first cell, because it converts a capital decision into an operating one and the internal argument is easier. Buying wins once you are on your third cell and the team can maintain them.
Do you supply the hardware?
We are vendor-neutral and integrate what fits the case. We have no reseller margin to protect, which is the reason we can tell you a cell does not pay off.
Can you work on a site outside Germany?
For robotics, physical proximity matters more than it does for software. We take UK and Benelux sites; further afield we work with a local integrator and own the AI and simulation side.
Bring us the part and the volume.
With a drawing, a part mix and a cycle time we can tell you within a call whether this is a cell, a retrofit, or a case to revisit in two years.