The hardware usually works. The pilot usually fails anyway, because nobody agreed what "working" meant before the crate arrived.
Weeks 1–2: scope, in numbers
One task. Measured. A tote weighing 6.2 kg moving 14 m, 90 times a shift, with a 40-second takt. If the scope document does not contain numbers, the pilot will end in an argument about whether it succeeded.
Weeks 3–4: bench validation
The robot does the task in our lab, on a mocked-up version of your station. This is where you discover that your shelf is 3 cm too deep or your conveyor height was never actually measured. Cheap to fix here, expensive later.
Weeks 5–8: shadow mode
The robot runs alongside your staff without touching the real workflow. It performs the task into a holding area; your team ignores it. This is the week everyone wants to skip because "it obviously works". It is also the only week that produces honest data on interruptions, and interruptions are what kill utilisation.
Weeks 9–11: supervised operation
Real work, real consequences, one supervisor per shift. Speed limits stay low. Every stop is logged with a one-line cause. By week 11 the stop log should be boring.
Weeks 12–13: handover
Two staff trained as first-line operators, a printed runbook, a named escalation path, and a review meeting where we read the stop log out loud together. Only then does the pilot become a deployment.
Every week you cut from shadow mode is a week you will spend in month three explaining why utilisation is 40 %.
What good looks like at day 90
Utilisation above 70 % of scheduled hours, fewer than one unplanned stop per shift, and one person on your team who can restart the robot without calling us.