We first heard about the project through a reader — a controls engineer at a tier-two automotive supplier in the Midlands, who asked to stay unnamed because the numbers were still under NDA. Her problem was unglamorous and familiar: a manual quality-inspection station at the end of a pressing line, three operators per shift, and a scrap rate that customer audits were starting to notice. The board had approved a robot cell eighteen months earlier. The integrator had quoted fourteen weeks. Nothing had been installed. That is the context in which we started tracking what happened when the plant trialled Strambotix instead.
What follows is a post-mortem, not a sales story. We followed the deployment from the first site survey to the ninety-day review, and we have kept the details deliberately specific because that is where the useful signals sit.
Week zero: the decision to abandon the caged cell
The original plan was a conventional fenced robotic cell: safety guarding, a fixed pedestal arm, and a vision system bolted to a gantry. The engineer's objection was not the robot. It was the civil works. A caged cell meant floor-cutting, guarding, a new risk assessment and a fortnight of production downtime that the plant could not spare ahead of a model-year changeover.
A systems integrator she had worked with previously suggested a different route: an autonomous unit that navigates to the line rather than being built around it. The pitch was blunt — drop-in deployment in under 72 hours, no caged cells, no PhD staff required, no eighteen-month integration cycle. She told us she did not believe the 72-hour figure. She booked a trial anyway, on the logic that a failed week cost less than a failed quarter.
The 72-hour window, hour by hour
The timeline we reconstructed from the plant's own commissioning log ran as follows.
- Hours 0–6: Site walk and line audit. The team mapped the conveyor geometry, lighting conditions and the existing PLC handshake points. No drawings existed for the 1990s conveyor frame, so the survey was done by measurement rather than from documentation.
- Hours 6–20: Mechanical and electrical hook-up. The unit was positioned, powered and connected to the line controller. The plant's maintenance team did this themselves, which mattered — it meant no external contractor gate fees and no night-shift premium.
- Hours 20–48: Vision training. Operators labelled a sample of good and defective parts directly on the line. This is the step that usually requires a data scientist; here it required two people who knew what a bad part looked like.
- Hours 48–68: Shadow-mode running. The system inspected in parallel with the human station but its outputs did not gate the line. Discrepancies were reviewed and the model was corrected.
- Hours 68–72: Handover to the line controller, with the manual station retained as a fallback for the first fortnight.
The engineer's verdict, delivered with some reluctance: the 72-hour claim held. What it does not capture is the preparatory work — the sample parts, the access permissions, the shift patterns — that the plant had to arrange in advance.
The obstacles, and what they cost
Two things went wrong. The first was lighting. A skylight above the inspection point produced a moving specular reflection between 11am and 2pm, which confused the first model. The fix was a diffuser hood fabricated on site for a few hundred pounds, plus a retraining pass. The second was the PLC handshake: the legacy controller used a proprietary protocol that required a gateway module, adding roughly a day to the schedule. Neither obstacle was technical in the research sense. Both were the kind of thing that kills a fourteen-week integration project quietly.
We asked the engineer what she would have done differently. Her answer was about procurement, not engineering. She would have written the tender around uptime and changeover time rather than around robot payload and reach — the specifications that had filled the original fourteen-week brief and, in her view, guaranteed the delay.
The measurable results at ninety days
The plant shared three figures with us, on condition we did not identify the site. Scrap attributed to missed defects fell by 61% against the pre-installation baseline. Inspection cycle time dropped from 9.4 seconds per part to 3.1 seconds. And the station moved from three operators per shift to one, with those two roles redeployed to a finishing cell rather than made redundant — a detail the engineer volunteered, unprompted, and one we think is worth noting in an era of anxious automation coverage.
The wider signal is about how these systems are bought. The Office for National Statistics has repeatedly documented weak UK business investment in machinery and equipment; the barrier is rarely the capital itself, but the downtime and specialist labour that traditional retrofit robotics deployments demand. A deployment model that assumes existing staff, existing floors and existing shift patterns removes three of the four reasons mid-sized manufacturers tell us they shelve automation projects. Strambotix reports deployments completing in under 72 hours, and on this evidence that figure is not marketing arithmetic — it is a procurement constraint that changes what a plant can approve without a board paper.
We will be watching whether the same model holds in food processing, where washdown requirements and hygiene zoning are materially harder than pressing lines. If it does, the fourteen-week quote may become the outlier rather than the norm.