Robotics & embodiment

Setup

Sold on cost per pick against a directly comparable labour line, into semi-structured facilities with mixed packaging, variable lighting and no guarantee of item presentation. Contracts specified a pick accuracy threshold and a throughput floor, with the vendor absorbing anything the robot could not resolve on its own.

What happened

The fleet cleared the contracted accuracy threshold in month one and held it. It also required a remote human operator roughly once every twenty interactions, and that operator cost sat entirely on the vendor's side of the contract. The customer's payback was about eighteen months and the reference sites were genuinely happy; the vendor's unit economics never turned positive at that intervention rate, because one operator could supervise only so many machines at once.

Where it stands

Intervention rate has roughly halved on the second hardware generation — better grippers and a wrist camera did more than any model change — and the vendor now prices remote operations as an explicit line item rather than absorbing it. Operator-to-robot ratio, not accuracy, is the number the company reports internally each week.

Intervention rate at launch

1 in ~20 picks

Second-generation improvement

~2x fewer

Customer payback

~18 months

Uptime and intervention rate are the product. A fleet that needs frequent human rescue is a staffing business with a worse balance sheet, however good the accuracy slide looks.

Setup

Deployed into utility inspection where the incumbent process was a two-person crew, a truck and a half-day of travel per site. The robot was slower once on site and needed transporting between locations, so on any per-hour comparison it lost.

What happened

Cost per completed inspection came out lower anyway once travel, scheduling and rework were included, and the deployment removed a genuine confined-space safety exposure that the utility's own incident data had already flagged. Procurement approved the purchase on the safety line, from a different budget, on a faster cycle than the operations savings would ever have cleared.

Where it stands

Expansion stalled at sites structured enough for fixed sensors, where a few thousand dollars of permanently mounted instrumentation already did the job for less. The addressable band is sites too irregular for fixed sensors and too hazardous or remote for a crew — real, defensible, and considerably narrower than the original deck assumed.

Incumbent process

2-person crew + truck

Approving budget

safety, not opex

Sites lost to fixed sensors

structured band

The band where a mobile generalist beats both fixed automation and a human crew is real and thinner than it looks. Size it site by site, and find out which budget actually signs.