Reliability is often discussed as a percentage. The more useful question is what produces that percentage: component choice, assembly quality, environmental design, software stability, and how quickly a fault is seen and repaired.
Design for the environment
A machine in a climate-controlled lobby and a machine on a street corner are not the same product in practice. Dust, temperature, vibration, and public exposure all change the failure modes. Hardware has to be specified for the harsher case if the fleet includes outdoor units.
Test what the field will do
Laboratory testing should include the mechanical cycles, power events, and network interruptions the machine will actually see. A unit that works on a bench and fails after a month on site was not finished.
A reliable ATM is one that can be diagnosed remotely and serviced locally without taking the rest of the network with it.
Reliability continues after install
Preventative maintenance, software updates, and cash operations all affect whether the public still sees a working machine. Hardware quality sets the ceiling. Operations determine whether the fleet lives up to it.