Five stages, each ending in an explicit decision gate — evaluated through the same six lenses at every stage. Strategy, discovery, productization, launch and scale for robotics, autonomous systems, industrial vision and intelligent machines.
The lifecycle is cyclical, not a waterfall: field evidence, customer learning, incidents, economics and technical discovery can send a product back to an earlier stage.
Stage 1 of 5
Strategy
The decision
Is this the right product opportunity, for the right customer, with a credible path to value?
The gate
Opportunity approved for evidence-building — or stopped before avoidable engineering expenditure.
Representative work
Market and ecosystem analysis, opportunity selection and segmentation
Product thesis, vision, intended use and strategic boundaries
Autonomy strategy and product-versus-feature-versus-service decisions
Build, buy, partner or integrate decisions
Business model, revenue architecture and product economics
Product-family or platform hypothesis and investment/stop criteria
Representative outputs
Product strategy memorandum
Opportunity map and ICP with value proposition
Competitive-alternative analysis
Business-model architecture and product economics
Strategic roadmap with a go, redirect, partner or stop recommendation
Stage 2 of 5
Discovery
The decision
Is there enough customer, field, technical and economic evidence to justify product investment?
The gate
Validated problem, credible product concept, acceptable feasibility and sufficient evidence to invest — or a clear decision not to proceed.
Representative work
Buyer, user, operator and maintainer research; site and workflow observation
Jobs-to-be-done, current and target journeys, economic value to the customer
Data, sensor and AI feasibility assessment
Human-factors research and human-oversight concept
Design-partner development and procurement-path validation
Willingness-to-adopt and willingness-to-pay testing; prototype and simulation experiments
Representative outputs
Discovery evidence report with stakeholder and workflow maps
Intended use, non-goals and foreseeable misuse
Product concept and customer-value model
Data/AI feasibility assessment and assumption register
Experiment results and a development recommendation
Stage 3 of 5
Productization
The decision
Can this become a complete, dependable, supportable and commercially viable product?
The gate
A product candidate that meets explicit customer, technical, operational, trust and economic acceptance criteria.
Representative work
Concept of operations, product and system requirements, product boundaries
AI-versus-deterministic-control decisions and hardware-software-AI architecture
Data and evaluation strategy; intelligent human-machine interaction
Simulation, software-in-the-loop, hardware-in-the-loop, on-device and field-validation planning
Reliability and failure behaviour; modularity, commonality and product-family decisions
Roadmap, release strategy, requirements traceability and change governance
Representative outputs
ConOps and product requirements
Architecture and interface decisions with an AI/data strategy
Evaluation and acceptance matrix; human-oversight specification
Risk register and prototype or pilot plan
Product-candidate criteria and product-family decision
Stage 4 of 5
Launch
The decision
Can the product be sold, installed, accepted, supported and operated responsibly?
The gate
Controlled market or operational release with measurable acceptance and support readiness.
Representative work
Positioning, messaging, and pricing and packaging strategy (without necessarily publishing price)
Go-to-market planning, design-partner conversion and launch-customer selection
Site readiness, installation, commissioning and customer acceptance
Operator and technician training; product documentation; sales and partner enablement
Support and escalation model; telemetry and observability
Incident response; release, rollout and rollback
Representative outputs
Launch strategy and go-to-market plan
Pricing and packaging architecture
Deployment playbook and site-readiness checklist
Acceptance criteria, training and support model
Incident-response plan, launch dashboard and controlled-rollout recommendation
Stage 5 of 5
Scale
The decision
Can customer value, deployment and economics be repeated across sites, machines, customers and product variants?
The gate
Evidence that value, deployment, support and unit economics are repeatable enough to justify scale.
Representative work
Product and fleet analytics; adoption, utilisation, expansion and retention
Deployment-repeatability improvement and support-cost reduction
Reliability roadmaps, product-line extensions and pruning
Product-family commonality and shared platform architecture
APIs, standards, partner and ecosystem strategy
Model, firmware and configuration lifecycle; product operations and retirement
Representative outputs
Product KPI tree and fleet/deployment dashboard
Cost-to-serve model and expansion strategy
Product-family roadmap and platform architecture
API and standards strategy with a partner model
Product operating cadence and portfolio/sunsetting plan
The six decision lenses
Every stage and every engagement evaluates the product through the same six lenses — they are what make the system specific to Physical AI.
1
Customer and operator value
Who adopts it, who operates it, and why it is worth changing how they work.
2
Business model and product economics
How the product earns, what it costs to deliver and support, and when it repeats.
3
AI, data, evaluation and human-machine interaction
What the intelligence must do, how it is evaluated, and how people work with it.
4
Hardware, software and system architecture
The physical and digital system that has to hold together as one product.
5
Safety, cybersecurity, governance, compliance and human oversight
The boundaries, controls and oversight that make the product responsible to operate.
6
Deployment, operations, product family, partners and ecosystem
How it installs, runs, scales across variants, and lives inside a wider ecosystem.
Not generic digital product management. Digital product management iterates software behind a release flag. A Physical AI product couples hardware revisions, model behaviour, field conditions, safety cases and operator trust — the gates and the lenses exist because those cannot be managed as separate concerns.
Where is your product?
Most engagements start by locating the product honestly on this lifecycle — and naming the one decision that matters next.