Humanoid robots have real warehouse jobs now, but the ROI math only works for narrow tasks
Humanoid robots stopped being a demo trick sometime in the past year. Agility Robotics' Digit is running a revenue-generating deployment at a Spanx warehouse in Georgia. Figure's 02 has logged more than 1,250 hours on BMW's Spartanburg production line, loading over 90,000 parts across more than 30,000 vehicles with better than 99% placement accuracy. These aren't pilots running for a press release — they're recurring shifts with measurable output.
The thesis worth stating plainly: humanoid robots now clear a real ROI bar for a specific, narrow category of warehouse and manufacturing work. But the same data that proves this also shows exactly where the economics stop working, and companies extrapolating from "Digit works at Spanx" to "humanoids are ready for general warehouse labor" are getting ahead of what the numbers actually support.
The numbers that actually clear
Agility prices active pilot deployments of Digit around $250,000. Run against a warehouse labor benchmark of roughly $30/hour, Agility's own modeling — and independent calculations against it — put payback at 18 to 24 months, with 120–180% ROI over a five-year horizon, provided the robot runs a two-shift operation. That last condition matters more than it sounds: a single-shift deployment roughly doubles the payback period, because the capital cost is fixed but the hours of offsetting labor cost are cut in half.
Figure hasn't published official Figure 02 pricing, but industry estimates place early manufacturing and logistics deployments between $30,000 and $150,000 depending on configuration and support contract. At BMW, the deployment metrics are specific enough to be useful for anyone modeling their own case: 37-second load cycles at 5mm tolerance, sustained across tens of thousands of parts. That's not a robot doing a flashy one-off task — it's a robot doing the same motion thousands of times with industrial-grade consistency, which is precisely the kind of work the ROI math favors.
Why the task profile matters more than the robot
Both deployments share a structural feature that's easy to miss if you're focused on the humanoid form factor rather than the job: the tasks are repetitive, spatially predictable, and tightly scoped. Digit moves totes. Figure 02 loads parts into a fixture at a fixed station with known geometry. Neither is doing open-ended, variable-dexterity work — picking mixed, unknown SKUs from a cluttered bin, handling irregular packaging, or adapting to a warehouse layout that changes week to week.
That distinction is the entire ballgame for ROI. The payback math above assumes a robot running near-continuously on a task it's good at. The moment a deployment requires the robot to handle task variability — the kind that's trivial for a human worker and still genuinely hard for current humanoid platforms — utilization drops, error-driven downtime rises, and the payback period stretches well past the 18–24 month range that makes the pilot deployments look compelling.
What buyers should actually check before signing
- Shift count, not just hourly rate. A vendor's ROI model built on two-shift operation doesn't transfer to a single-shift warehouse. Ask for the assumption explicitly and rebuild the math for your actual operation.
- Task structure, not task category. "Warehouse work" isn't specific enough. Ask whether the deployment is closer to Digit's tote-moving or Figure's fixed-station part loading — both are narrow and repetitive — versus mixed-SKU picking, which neither platform has proven at comparable scale.
- Real uptime data, not demo footage. BMW's disclosed metrics (hours logged, placement accuracy, cycle time) are useful precisely because they're operational numbers, not marketing reels. Ask any vendor for the equivalent before assuming their platform performs the same way.
- Maintenance and support costs baked into payback. Pilot-stage pricing often understates ongoing service costs once a robot is out of a vendor-supported showcase deployment.
The realistic read for 2026: humanoid robots are commercially viable today for a genuinely useful but narrow slice of warehouse and manufacturing work — repetitive, structured, two-shift-friendly tasks where the ROI math has been proven with real operational hours, not projections. General-purpose warehouse labor replacement is still a different, harder problem, and the companies with the most credible deployments right now are the ones being explicit about that boundary rather than blurring it.
Sources: Agility Robotics pilot deployment data and ROI modeling, Figure AI BMW Spartanburg production disclosures, and industry reporting on 2026 humanoid robot commercial deployments.