Robotics customer-side deployment funnel v1: from KPI pilot to rollout proof

Date: 2026-06-15 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: SYNTHESIS_CANDIDATE Visibility: PUBLIC Public-safety: public-safe evidence map; no trade recommendation; no Hugo portfolio context; no private channel checks; no paid-report excerpts.

0. One-line answer

The most valuable current stale-knowledge update is that humanoid robotics now has a clearer customer-side deployment funnel: BMW/Figure shows measurable line-side work, BMW/Hexagon shows a second-plant / second-vendor evaluation mechanism, and Amazon logistics shows what multi-site rollout proof looks like; this strengthens S4 deployment engineering evidence but still does not prove S5 scaled economics because customer ROI, intervention distribution, robot count by site, repeat-order economics, revenue, gross margin, and support burden remain undisclosed. 🟢 BMW / Figure / Amazon primary sources; 🟠 Charlie stage classification.

1. Core question

What public evidence should separate “humanoid deployed in production” from “humanoid economics proven”?

Working answer: use a customer-side deployment funnel, not a demo/funding headline.

1. Lab / controlled evaluation.
2. Customer-site test deployment.
3. Line-side KPI pilot with task, runtime, parts, cycle-time target, accuracy target, intervention target.
4. Second site / second vendor / center-of-competence replication.
5. Multi-site rollout with explicit robot count, uptime/intervention, safety case, support model, ROI/payback, and contract economics.
6. Financial proof: repeat order, revenue, gross margin, backlog/RPO, service-cost burden.

Current humanoid evidence is strongest around steps 3–4. It is not yet step 5–6 in public sources.

2. What changed / why now

Existing robotics artifacts already cover Q0–Q5, Tesla/Figure/Leaderdrive, China-vs-US, logistics benchmarks, safety standards, and capital formation. This artifact adds a narrower customer-side deployment test: what does a real buyer/operator do after a successful humanoid pilot?

The freshest source-backed signal is BMW’s 2026 Leipzig announcement:

• BMW says it is bringing Physical AI to Europe through a humanoid pilot at Plant Leipzig. 🟢
• BMW says a theoretical evaluation phase and laboratory tests were followed by an initial test deployment at Leipzig in December 2025, with another test deployment planned from April 2026 and the actual pilot phase starting in summer 2026. 🟢
• BMW says the Leipzig work uses Hexagon Robotics / AEON, not Figure, creating a second-vendor comparator. 🟢
• BMW says the Spartanburg Figure deployment proved humanoids can function under real-world automotive manufacturing conditions, but BMW still frames Leipzig as pilot / step-by-step integration, not broad rollout economics. 🟢/🟠

Interpretation: this is a stronger adoption-process signal than a single customer logo. It shows an industrial customer building an internal evaluation and replication mechanism. But it remains S4 until BMW or vendors disclose repeat fleet economics.

3. Evidence table

4. Stage classification

Current classification: S4 customer-side deployment engineering.

Why S4:

• Real industrial customer site: BMW Spartanburg. 🟢
• Quantified task evidence: 90,000+ parts, 1,250+ hours, 30,000+ vehicles, 10-hour weekday shifts. 🟢
• KPI framework disclosed: cycle time, placement accuracy, interventions. 🟢
• Second-site / second-vendor evaluation appears at BMW Leipzig with Hexagon / AEON. 🟢
• Vendor-side manufacturing and fleet/service infrastructure is improving. 🟢

Why not S5:

• BMW/Figure public evidence does not disclose number of robots by site, paid contract value, customer ROI/payback, full uptime/intervention distribution, safety incident data, repeat order, fleet expansion economics, robot revenue, gross margin, or field-service cost. 🟠
• BMW/Hexagon Leipzig is explicitly a pilot / test-deployment process, not a disclosed scaled procurement. 🟢/🟠
• Figure BotQ output improves supply, but production cadence and delivered units are not the same as utilized customer fleets or profitable deployments. 🟠

5. Signal vs noise

Signal

• BMW’s second-plant / second-vendor process is a meaningful adoption-system signal: a large manufacturer is not just watching a demo; it is building evaluation infrastructure. 🟢
• Figure’s BMW KPI disclosure is unusually concrete for humanoids because it names runtime, parts loaded, vehicle contribution, cycle-time target, accuracy target, and intervention target. 🟢
• Figure’s Figure 03 changes from BMW learnings show a field-feedback loop from deployment to hardware revision. 🟢
• Figure’s BotQ and service/fleet tooling show the bottleneck is shifting from “can a humanoid perform once?” to “can hundreds of robots be built, updated, serviced, and deployed?” 🟢/🟠
• Amazon’s fleet and rollout disclosures set the proof-quality yardstick humanoids should eventually resemble: site count, fleet count, efficiency delta, rollout schedule, capex envelope, workforce impact. 🟢

Noise unless upgraded

• “BMW is using humanoids” is too broad; the public signal is a bounded sheet-metal loading task plus a pilot/evaluation process. 🟢/🟠
• “Figure produced 350+ robots” is not the same as 350 customer-deployed revenue-generating robots. 🟠
• “Zero interventions target” is not the same as achieved zero-intervention fleet performance unless outcome distribution is disclosed. 🟢/🟠
• “Second vendor” is not a winner ranking; it is a customer-side procurement/evaluation signal. 🟠
• “Amazon-scale robotics validates humanoids” is too strong; Amazon validates logistics automation proof quality, not humanoid economics directly. 🟠

6. What would change our mind

Upgrade toward S5 if primary/customer sources disclose at least two of these:

1. BMW or another customer discloses robot count, paid deployment scope, multi-shift utilization, and repeat expansion.
2. Customer-side ROI/payback or productivity metrics tied to labor hours, throughput, quality, injury reduction, or scrap reduction.
3. Fleet-level uptime, intervention rate, safety incidents, and maintenance/service hours across multiple months.
4. Vendor discloses robot revenue, gross margin, backlog, service/support cost, or RaaS economics.
5. A second independent customer repeats a similar workflow with comparable economics.
6. BMW Leipzig moves from pilot to multi-line or multi-plant deployment with robot count and KPI outcomes.

Downgrade if:

1. BMW/Figure or BMW/Hexagon remains at pilot/test stage for 12–24 months without disclosed expansion.
2. Figure 03 production scale grows but customer-site deployment metrics do not follow.
3. Service/maintenance or intervention burden prevents repeat deployment.
4. Customer language shifts from operational value to exploration / learning only.
5. Safety or integration constraints keep humanoids in heavily supervised narrow cells.

7. Public-safe site draft section

The next robotics question is customer replication, not another demo

Humanoid robotics has crossed an important line: some robots are no longer just performing staged demonstrations. They are entering customer production environments and generating measurable work data.

The clearest public example remains BMW and Figure. Figure says Figure 02 ran 10-hour weekday shifts at BMW Spartanburg, loaded more than 90,000 parts, accumulated more than 1,250 runtime hours, and contributed to production of more than 30,000 BMW X3 vehicles. Just as important, Figure disclosed the operational KPIs it tracked: cycle time, placement accuracy, and human interventions.

But the more current customer-side signal is BMW’s next step. BMW says it is bringing Physical AI to Europe through a humanoid pilot at Plant Leipzig with Hexagon Robotics / AEON. Before the pilot phase, BMW describes theoretical evaluation, lab testing, an initial Leipzig test deployment, and a further test deployment planned before a summer 2026 pilot. BMW also created a Center of Competence for Physical AI in Production.

That matters because it changes the unit of evidence. The question is no longer simply “can a robot do the task once?” The better question is: can a customer evaluate, replicate, integrate, maintain, and scale humanoid robots across plants and workflows?

The answer is still not S5. Public sources do not yet disclose robot count by plant, paid contract value, ROI/payback, uptime, intervention distribution, repeat orders, robot revenue, gross margin, or service cost. But the customer-side funnel is becoming visible. For public research, that is the right next gate to track.

8. Slide-ready compression

Title: Humanoids are entering customer adoption funnels, not yet scaled economics

Four cards:

1. BMW/Figure measured work

   • 90,000+ parts loaded.
   • 1,250+ runtime hours.
   • 30,000+ X3 vehicles supported.
   • KPI frame: cycle time / placement accuracy / interventions.
   • Source: Figure 🟢.

2. Figure feedback loop

   • Figure 02 BMW deployment identified forearm as top failure point.
   • Figure 03 redesigned wrist electronics.
   • BotQ: 350+ Figure 03 delivered; 1/day to 1/hour; 24x in <120 days.
   • Source: Figure 🟢.

3. BMW replication process

   • Leipzig pilot with Hexagon / AEON.
   • Lab tests + initial Leipzig test deployment in Dec 2025.
   • Further test planned from Apr 2026; pilot starts summer 2026.
   • Center of Competence for Physical AI in Production.
   • Source: BMW 🟢.

4. Benchmark to watch

   • Amazon: 1m+ robots across 300+ facilities; DeepFleet 10% travel-time improvement.
   • Amazon Europe: >€10bn modernization; STARK to 15 sites by 2027; Proteus H1 2027 plan.
   • Humanoids need similar site-count / fleet-count / economics disclosure.
   • Source: Amazon 🟢.

Footer: Evidence map only. No winner ranking. No trade recommendation. Customer pilot ≠ scaled economics; production cadence ≠ utilized profitable fleet.

9. Common misconceptions

1. Misconception: “A humanoid at BMW means humanoid economics are proven.”

   • Correction: BMW/Figure proves a stronger S4 deployment KPI. S5 needs robot count, ROI/payback, repeat order, uptime/intervention, revenue, margin, and service cost. 🟢/🟠

2. Misconception: “If Figure can build one robot per hour, commercialization is solved.”

   • Correction: manufacturing supply is necessary but not sufficient. The customer-side gate is deployed utilization and economics. 🟠

3. Misconception: “Second-vendor BMW testing means one vendor is losing.”

   • Correction: second-vendor testing mainly shows BMW is creating an evaluation funnel; it is not a public procurement ranking. 🟠

4. Misconception: “Amazon’s robot fleet means humanoids will scale the same way.”

   • Correction: Amazon is a proof-quality benchmark, not a direct humanoid proof. Humanoids still need their own multi-site utilization and economics. 🟢/🟠

10. Think Deeper questions

• Is the first S5 humanoid proof more likely to come from one vendor’s broad fleet, or from one customer’s repeatable deployment playbook?
• Does the customer-side Center of Competence become a value-capture layer in robotics adoption, similar to internal cloud/data platforms in AI adoption?
• Which metric should matter most after a pilot: robot count, utilization hours, intervention rate, task expansion, repeat order, or ROI/payback?
• If customers evaluate multiple humanoid vendors in parallel, does value migrate away from OEM hardware toward integration, fleet management, safety case, and service?
• Is Amazon’s logistics benchmark too high for humanoids, or is it the right standard for avoiding demo-cycle overreach?

11. Source list

1. BMW Group PressClub, “BMW Group to deploy humanoid robots in production in Germany for the first time,” 2026. URL: https://www.press.bmwgroup.com/global/article/detail/T0455864EN/bmw-group-to-deploy-humanoid-robots-in-production-in-germany-for-the-first-time?language=en 🟢
2. Figure AI, “F.02 Contributed to the Production of 30,000 Cars at BMW,” 2025-11-19. URL: https://www.figure.ai/news/production-at-bmw 🟢
3. Figure AI, “Ramping Figure 03 Production,” 2026-04-29. URL: https://www.figure.ai/news/ramping-figure-03-production 🟢
4. Amazon, “Amazon launches a new AI foundation model to power its robotic fleet and deploys its 1 millionth robot,” accessed 2026-06-15. URL: https://www.aboutamazon.com/news/operations/amazon-million-robots-ai-foundation-model 🟢
5. Amazon, “Amazon announces new robots, faster delivery, and 25,000 jobs in Europe,” accessed 2026-06-15. URL: https://www.aboutamazon.com/news/operations/amazon-europe-robotics-delivery-investment 🟢
6. Amazon, “Amazon unveils next-gen Proteus robot as part of €10 billion European investment in its fulfillment network,” accessed 2026-06-15. URL: https://www.aboutamazon.com/news/operations/amazon-proteus-robot-europe-investment-employee-support 🟢
7. Charlie derived calculations from Figure BMW metrics: 90,000 parts / 1,250 runtime hours = 72 parts per runtime hour; 90,000 parts / 30,000 vehicles = 3 parts per vehicle. 🟠

12. Public-safety flag

PUBLIC-safe with these exclusions:

• No Hugo private portfolio weights, watchlist logic, purchase prices, tax context, trade rationale, private channel checks, paid-report excerpts, or rumors.
• No buy / sell / hold language for BMW, Figure, Hexagon, Amazon, Tesla, Unitree, suppliers, or any related public/private company.
• Do not imply BMW has placed a scaled fleet order or proven ROI/payback unless future primary sources disclose it.
• Do not imply Figure 03 production cadence equals customer-deployed profitable fleet.
• Do not imply Amazon logistics scale directly proves humanoid commercialization.