State of Physical AI Product Management 2027: Research Protocol

How do organizations assign product ownership, make evidence-gated decisions and move intelligent machines from technical capability into repeatable field and commercial operation?

The proposed study covers industrial and collaborative robotics, autonomous and mobile systems, intelligent machines, inspection and control, energy and charging infrastructure, and embodied or edge-AI products with material operator, field or service consequences. It includes the platforms and product families that support them.

An operated product is the complete promise a buyer can justify, people can use, sites can deploy, specialists can assure and an organization can support at repeatable economics. The study asks who actually owns integrated product choices; a job title alone does not establish authority.

Software-only AI without a material physical operating system, fixed automation without material AI behavior, and pure research without a declared user or productization intent would be excluded or analyzed separately. The instrument must not require respondents to use the label “Physical AI.”

This is a proposed mixed-method study, not a representative census or a causal experiment. Purposive sampling would seek variation in company maturity, role, product, sector, size, geography and success state, including stalled and stopped work. A European or Netherlands weighting would be disclosed if present.

A later validation phase would return anonymized themes for error checking and invite a disclosed practitioner review panel. Reviewer comments would remain distinct from participant data, and disagreements would be recorded. No reviewer participation is claimed.

The existing Product System provides an explicit interpretation lens across Strategy, Discovery, Productization, Launch, Scale. It must not dictate respondents’ language or force their accounts into a preferred framework.

• Customer, operator and product value
• Economics and business viability
• AI, data, evaluation and HMI
• Complete system architecture
• Safety, cybersecurity, compliance and governance
• Deployment, operations, service and ecosystem

The interview guide asks for a specific product, operating conditions, consequential decision, actual decision rights, alternatives, evidence and unresolved gaps. It examines product families, variants, build/buy/partner decisions, useful-work baselines, lifecycle economics, field support, failure and handover.

Reliability questions name exposure and denominators, intervention and recovery burden, and operator workload. Model questions name the role and maximum authority of world models, VLAs or policies; evaluator dependencies; task memory and fleet learning; and the capability, complexity, migration and rollback consequences of a technology change.

Comprehension would be piloted with at least five people outside the design team before the main instrument is frozen. Exclusions and analysis rules must be recorded before analysis, with traceable reasons for duplicate, ineligible, empty, machine-generated or unresolved contradictory responses. Inconvenient responses cannot be silently removed.

Raw data, cleaned data, analysis code and publication tables would be separate and versioned. A second reviewer would independently code a sample. Counts and denominators would accompany percentages; skewed numeric responses would use distributions and medians.

Cross-tabulated cells with fewer than 10 respondents would be suppressed. Sectors, roles and companies would not be ranked from small or non-comparable groups. Associations would remain descriptive; negative cases and framework-mapping choices would be visible. Design-partner interviews would remain separate from the main sample.

Before any collection, a dedicated notice and explicit consent process must explain the organizer and funding, purpose, time, voluntary participation, recording, uses, quote options, retention, withdrawal and contact procedures. Data locations and retention/withdrawal arrangements still require preparation and review before recruitment.

Quote choices would be named attribution, role/sector only, or no quotation, with no public person or company identification by default. Raw notes, contact details, disclosive small cells, customer secrets and identifiable incidents would not be published. An anonymized dataset would require a separate assessment of consent, contracts and re-identification risk.

Hyperion may benefit commercially from category visibility. Participation must not imply a customer relationship, endorsement, logo permission, testimonial, training reuse or commercial follow-up permission. Marketing and commercial follow-up require separate explicit opt-in. No hidden sales scoring, preferred recommendation of Hyperion, or sales-call gate for aggregate findings belongs in the study.

• Pending:Owner approval of scope and budget
• Pending:Completed privacy and legal review
• Pending:Defined conflict and relationship labels
• Pending:Tested consent language and withdrawal path
• Pending:Secured recruitment ledger and data locations
• Pending:Tested design-partner guide
• Pending:Frozen main instrument version
• Pending:Preregistered exclusion and analysis rules with an immutable timestamp
• Pending:Assigned publication and disclosure review owners
• Pending:No advertised result, achieved sample or participant logo before it exists and is authorized

A “State of” report must not proceed if eligibility or provenance cannot be checked, an undisclosed relationship dominates the sample, instrument versions cannot be separated, privacy or contract obligations cannot be met, the evidence supports only anecdotes, or the analysis cannot be reproduced.

The appropriate outcome may be a clearly labelled interview synthesis, a methods note, or no publication. Any eventual report would need its achieved sample, collection dates, instrument, analysis notes, disclosure controls, limitations and correction history. The year in this protocol’s title is not a promised release date.

Download the complete design protocol · Markdown

Protocol owner: Mohammed Cherifi / Hyperion Consulting. AI tools supported preparation; no completed human editorial or privacy/legal review is asserted.