The demo worked. Now make the robot work on Tuesday.
The robot crosses the floor, finds the pallet, moves the box—and the room exhales. Then the next run stops with a fault no one can reproduce before lunch. The customer asks the harder question: will it do the same thing all day, with a different box, a low battery, and no expert technician nearby?
A robotics company has to turn an impressive run into a dependable service. That takes more than better code. It takes evidence that earns each next test, safety decisions with clear owners, and a business plan that counts the work around the robot as carefully as the work it performs.
Replace calendar milestones with proof that changes the next decision
A milestone should say what capability the robot must demonstrate, under which conditions, with what evidence, known exclusions, and what commitment the result permits. A date can schedule the work; it cannot establish readiness.
— Use a one-page milestone brief to define the claim, operating envelope, baseline, evidence, exclusions, cost, and decision that follows.
— Keep verification (“built to specification”) distinct from validation (“works for its intended use in context”).
— Separate prototype learning, technical validation, and commercial adoption so a successful demo does not quietly become a claim of field readiness.
Read the intervention, not just the success rate
Eight completed runs out of ten can conceal two different problems: a false obstacle stop and a failure to yield near a person. The average does not tell you what to fix—or whether exposure should expand.
— Record interventions and near misses by scenario, consequence, mechanism, and exposure; use logs, video, and operator accounts to investigate what happened.
— Sort results into supported claims, weakened claims, open questions, and confounds. Use each category to decide whether to proceed, narrow, redesign, or retest.
— Track hours, distance, cycles, loads, conditions, and human interventions alongside incident counts. A quiet log is useful only when you know what the robot encountered.
— Set stop rules and name who can pause or change a trial before customer pressure turns a narrow approval into a broader promise.
Count the service behind the machine
Task completion alone can make a fragile business look efficient. An attendant who rescues the robot hourly, a hand-fitted gripper, or a pilot dependent on one enthusiastic manager changes the economics.
— Count recovery, charging, maintenance, onboarding, safety controls, and support when judging whether a robot can deliver value.
— Bring safety, operations, and manufacturing expertise into decisions while design and scope can still change; assess a hire by the decisions and evidence they improve.
— Tie technical proof to customer workflow, procurement, supplier lead times, and cash forecasts. Payroll and obligations keep arriving while a replacement part is delayed.
The hard road to a working robot is a sequence of bounded claims, honest evidence, and commitments sized to what the system has actually proved. That discipline helps a team learn faster without mistaking activity for readiness—and build promises it can keep on the customer’s floor.