Physical AI Commercialization

PHYSICAL AI COMMERCIALIZATION

Physical AI
doesn't fail in the lab.
It fails on the P&L.

The model works. The sensor works. The commercial model — pricing, regulatory posture, edge/cloud split, service economics — is where physical-AI products stall. We've launched them from conception through clearance at Nokia, GE Healthcare, NavVis, and EY. Operators, not consultants.

OPERATOR PROOF
4×

Physical-AI products led from conception to market — across ultrasound imaging, connected vehicles, mobile spatial mapping, and industrial digital twin.

WHERE PHYSICAL AI STALLS

The model is the easy part.
Four problems
software playbooks can't solve.

SaaS instincts mislead in physical AI. Hardware is a cost structure, a regulator, and a field-service problem — not an update you can ship on Friday. The companies that scale solve these four as one architecture.

01 · PRICING IS COMPOUND

Device + subscription + services + data.

One product, four revenue layers, four margin profiles. Treat them as one SaaS tier and the board model breaks. Treat them as four unrelated SKUs and procurement walks.

02 · THE REGULATOR IS THE FIRST BUYER

Clearance timeline is the commercial plan.

FDA 510(k), EU MDR, EU AI Act Annex III, UNECE, ISO 26262. Each adds months and reshapes what you can claim — which reshapes what you can charge.

03 · EDGE / CLOUD IS A BUSINESS DECISION

Latency, privacy, TCO, channel.

Where inference runs determines margin, data rights, serviceability, and who owns the customer. Engineering answers it by default. That default usually loses the business.

04 · LAUNCH VELOCITY IS 18–36 MONTHS

The commercial model has to survive two hardware cycles.

Unlike SaaS, you can't re-price every quarter. The first contract sets precedent for a decade of installed base. Getting it right pre-launch is worth ten iterations post-launch.

THREE ANGLES · ONE PRACTICE

Same operators.
Three buyer realities.

Physical AI shows up in three different buying contexts. The technical problem rhymes; the commercial problem is distinct. We work all three.

01 · INDUSTRIAL & SPATIAL

SLAM works.
Your commercial model has to.

Mobile mapping, digital twin, industrial automation, robotics. Hardware + software + services pricing, enterprise field deployment, channel design.

NavVis · EY · AEC, manufacturing, energy

02 · HEALTHCARE IMAGING & DEVICES

Your AI feature
is a regulated claim.

FDA SaMD, clinical validation, evidence-backed pricing, reimbursement strategy, ultrasound and imaging modality AI.

GE Healthcare · Voluson · Class II devices

03 · CONNECTED VEHICLES & ROBOTICS

OEM procurement,
not SaaS procurement.

V2X, ADAS, autonomy stacks, robotics foundation models. Tier-1 OEM contracts, functional-safety-adjacent commercial models, multi-year platform revenue.

Nokia · Connected Vehicle · ADAS

NAMED · DATED · SPECIFIC

We've sat
in the seat.

Every claim below is a product shipped, not a deck written. Ask us for detail in the call.

GE Healthcare
Voluson ultrasound AI
Product leadership on clinical AI inside an FDA-cleared Class II imaging device. Owned the clinical evidence → regulatory → pricing → field deployment chain.
FDA SaMD, hardware-embedded AI, clinical validation, 510(k) discipline.
NavVis
VLX · MLX · Ivion
Product ownership on mobile mapping hardware and the spatial computing platform. SLAM, LiDAR fusion, digital-twin delivery into enterprise AEC and industrial buyers.
Spatial AI, hardware + SaaS pricing, enterprise field deployment, channel partners.
Nokia
Connected Vehicle · V2X
Productised V2X and edge inference for automotive OEMs. Designed pricing for tiered automakers and multi-year platform contracts.
Automotive-grade AI, OEM procurement, functional-safety-adjacent commercial models.
EY
Industrial digital twin
Commercial architecture for enterprise twin programs across manufacturing, energy, and infrastructure. Buy-side diligence and operating-model redesign for AI in the physical world.
Enterprise twin economics, industrial AI transformation, PE / corporate buy-side lens.

WHAT WE BUILD · 8–12 WEEKS

A commercial architecture
a hardware P&L can defend.

Three workstreams, one integrated system. Each produces a standalone artifact your CFO, regulatory lead, or board can use independently.

01 — COMPOUND PRICING MODEL

Device, subscription, services, data — one integrated frame.

Four revenue layers priced as one coherent offer. Margin stress-tested across BOM volatility, compute cost, and service-level scenarios.

  • Four-layer pricing architecture with tiered commercial bundles
  • Margin model with BOM, compute, and service-cost sensitivity
  • Channel and partner economics — direct, OEM, reseller
02 — REGULATORY-COMMERCIAL ARCHITECTURE

Clearance pathway mapped to what you can charge.

FDA, MDR, EU AI Act, UNECE, ISO 26262 — translated into commercial claims, pricing posture, and evidence-backed positioning.

  • Pathway selection with dual-market (US / EU) reuse
  • Clinical or safety evidence package mapped to pricing claims
  • Post-market surveillance as a commercial asset, not a cost
03 — EDGE / CLOUD BUSINESS MODEL

Where inference runs is a board decision.

Latency, privacy, TCO, data rights, serviceability, and channel implications modelled as a single trade-space. CTO and CFO see the same picture.

  • Edge / cloud TCO and margin model across three deployment scenarios
  • Data-rights and customer-ownership implications per topology
  • Serviceability and installed-base economics over a 10-year horizon

FREQUENTLY ASKED QUESTIONS

Questions from physical-AI
product leaders.

We have a working sensor and a working model. Why is the commercial side so hard?

Because physical AI isn't SaaS and isn't hardware. It's both, layered, with a regulator and a field-service obligation on top. Your pricing has to price four things at once — the device, the software, the service, and the data — without one layer cannibalising another. Software pricing instincts fail here. Hardware pricing instincts fail too. The architecture needs to be designed for the category, not borrowed from either parent.

Our AI runs at the edge for latency reasons. How does that affect our business model?

It affects everything. Edge inference changes who owns the data, who sees the usage telemetry, how you service the installed base, what your gross margin looks like, and whether you can upsell the fleet. A decision that looks like an engineering choice — "we need 50ms latency, so inference runs on the device" — is actually a commercial architecture choice worth millions over the life of the product. We force the CTO and CFO into the same room, model three scenarios, and pick the topology on business grounds, not default ones.

We're planning a US / EU launch. How do FDA, MDR, and the EU AI Act interact?

They don't cancel each other. Most healthcare and imaging AI products land in the EU AI Act's high-risk category alongside MDR, and the dual-compliance regime is in effect from August 2026. US pathways (510(k), De Novo, PMA) run in parallel. We scope the combined pathway in the first two weeks so submissions, evidence packages, and governance artifacts are reusable across markets — not duplicated. The commercial implication matters: what you can claim in each jurisdiction determines what you can charge, which determines whether the business model even works.

Our AI feature is embedded in a hardware platform. Can we price the AI separately?

Sometimes yes, sometimes no — and getting it wrong is expensive. Voluson-class ultrasound, NavVis-class mobile mapping, and automotive ADAS each have a different answer. We work through four questions: is the AI claim regulated separately? does the buyer recognise the AI as a distinct value layer? can the product run without it? and does the channel support an add-on tier? The answer determines whether AI is a feature, a SKU, or a subscription layer. Each choice changes the P&L by an order of magnitude.

We're in hardware due diligence for a PE deal. What do you bring that a generalist firm doesn't?

Operator lens. Generalist diligence quantifies the TAM and benchmarks the software stack. We've run the P&L on products like the target — ultrasound AI, mobile mapping, connected-vehicle — and we know where the commercial model leaks: BOM-to-ARR translation, service margin erosion, channel conflict, regulatory exposure in adjacent markets, installed-base economics at year 7. A PE partner preparing a 100-day plan wants someone who has shipped the thing, not someone who has studied it.

Our launch is 18 months out. Is it too early to engage?

That is the ideal moment. The commercial architecture — pricing posture, regulatory framing, edge / cloud topology, channel design — is an order of magnitude cheaper to get right pre-launch than to fix post-launch. The first contract sets precedent for the installed base. We prefer to work 12–18 months before launch on physical-AI products, so the first enterprise close locks in the right terms, not the wrong ones.

How is this different from your Regulated AI practice?

Regulated AI solves commercial architecture for software running inside regulated enterprises — healthcare software, legal AI, financial services. Physical AI solves commercial architecture where the AI is embedded in a device, sensor, or vehicle. Different BOM, different regulator, different procurement. There is overlap (FDA SaMD, EU AI Act), and we route engagements to the right practice during the first call.

Technically ready.
Commercially stuck.
25 minutes to diagnose which.

We'd rather spend 25 minutes finding out whether your problem maps to this solution than scope an engagement that doesn't fit. Most intros end with a clear next step.