For equity investors in hardware-enabled, software-dependent companies, a quality-of-earnings report tells you what happened. It cannot tell you whether the asset stack works, whether the unit economics hold at scale, or whether the intelligence layer is real.
Intelligent hardware is physical equipment whose productive value depends not only on the asset itself, but also on software, connectivity, hosted services or data. Companies built on it are unit-economic engines. Their performance depends not on what is in the financials but on what is happening at the asset level: the goodput of a GPU rack, the uptime and redeployability of a robotics fleet, the CMS reliability of a charging network. Financial and commercial diligence tell you what the business has earned. They don't tell you whether the asset stack will keep earning it.
The companies that underperform post-close in this category rarely fail on the commercial thesis. They fail because operational and technical realities that were knowable at the time of investment were not examined with the right framework.
Across all five asset classes, the underlying technology stack shares a common structure. Risk at one layer typically manifests as a symptom at another. Understanding the anatomy is the precondition for diligence that finds the real exposure rather than the visible one.
The deployed unit that generates value: GPU server, robot, vehicle, charger, vending machine. Slow to change. Often supplier-dependent. Capex-heavy.
Networking, edge compute, local intelligence, protocol stacks. Standards-driven but fragile at integration points. Often the invisible dependency no one models.
Fleet management, dispatch, billing, monitoring, customer-facing software. Typically proprietary. Fast-changing. The layer where technical debt accumulates and scaling bottlenecks form.
Algorithms, ML models, optimisation logic that drives revenue or cost outcomes. Fastest-changing. Often claimed as competitive differentiation. Rarely examined.
Every intelligent-hardware company, regardless of vertical, can be underwritten across the same ten pillars. The pillars are universal. The specific questions, KPI thresholds, red flags, and structural mitigants within each pillar are calibrated to the asset class.
"Do the deployed assets actually perform at the throughput, uptime, and quality levels the unit economics depend on?"
"At the level of one unit, is the business profitable on a fully-loaded basis after maintenance, depreciation, and software cost?"
"Is the orchestration layer purpose-built and scalable? Does the telemetry drive decisions, or is it window dressing?"
"Where AI/ML is claimed as differentiation, what is the measurable performance lift versus a rule-based approach? Where is the moat, and how durable is it?"
"What is the bill-of-materials risk, sole-source dependency, lead-time exposure, and geopolitical sensitivity in the supply chain?"
"Who keeps the assets working once deployed? What does field economics look like at 3x current scale?"
"What is the regulatory perimeter? What pending regulation could expand or contract the addressable market before the investment horizon ends?"
"Connected devices in the field are an attack surface. What is the security architecture, OTA update capability, and incident response track record?"
"Who pays, on what contract structure, with what stickiness, what concentration, and what unit-level retention behaviour over time?"
"Does the team have the cross-functional depth to operate at the next level of scale? Where are the key-person dependencies and operational chokepoints?"
Across growth-stage and late-stage deal teams active in intelligent-hardware sectors, the differentiating diligence questions are remarkably consistent. Answered well, they lead to conviction. Left unresolved, they become the surprises that arrive in year two.
Not a portfolio average. A single fully-loaded unit P&L: one GPU, one charger, one robot, one device. Every cost allocated. Every assumption made explicit. This one question separates companies with genuine unit economics from those with a compelling story about them.
For deployed hardware, cohort 1 at 24 months is the most honest data the company has. Utilisation trajectory, maintenance cost curve, revenue retention. Companies that cannot produce this are either too early or not measuring what matters.
Every deployed-asset business has had a bad day. The response to that question tells you more than the incident does. The companies worth backing have written retrospectives, traceable corrective actions, and typically a stronger system as a result.
The loop from anomaly through triage to field resolution is a cultural fingerprint. Best-in-class companies measure it in days. The companies that will disappoint their investors measure it in months, or have no systematic way to measure it at all.
In almost every intelligent-hardware vertical, a 24-36 month generational change is coming: GPU architectures, robot platforms, charger power electronics, vehicle compute. Companies that have not modelled the upgrade path, the residual value curve, and the financing requirements are carrying undisclosed risk.
Intelligent-hardware companies live on telemetry quality. Calibration drift, data dropouts, edge-cloud sync failures, and sensor degradation silently corrupt operational decisions. The quality of the answer to this question is one of the strongest leading indicators of platform maturity.
Three names, typically. The head of a critical engineering function, a critical commercial relationship, and regulatory or compliance. Key-person concentration is acceptable in scale-ups; undisclosed and unmitigated key-person concentration is a structural risk.
Billing systems, audit logging, observability tooling, compliance documentation, contract management. Companies that have invested in this unglamorous layer scale. Companies that have not hit a ceiling at around 5-10x current size. Deficiencies here are nearly invisible until they become catastrophic.
In an engagement, the full framework provides the asset-class-specific instantiation of each question, the KPI thresholds that correspond to best-in-class performance, and the red-flag patterns that have preceded the most common post-investment surprises in each vertical.
Talk through the frameworkSOMA's equity diligence engagement produces decision-ready outputs structured for an investment committee, not a description of what was reviewed. Every finding is rated, every risk has a proposed mitigant, and the deliverable can be put in front of an IC on the day it arrives.
4-6 page decision document. Leads with a recommendation. Five headline findings (three strengths, two concerns) as testable propositions. RAG rating per pillar in a single table. Written by someone who saw the asset in the field.
Single-unit P&L with every cost allocated. Cohort utilisation, performance, and economics at month 24. Sensitivity analysis on the two most material assumptions. Empirically grounded, not modelled from management projections.
A RAG rating for each of the ten pillars. For every Amber or Red finding: a proposed deal-structuring response, whether a board seat, milestone, information right, hire commitment, or monitoring covenant.
Where the company stands relative to its regulatory perimeter, with a view on pending regulation that could alter the investment thesis before the hold period ends. Cyber and resilience risk specific to deployed-asset operations.
A 30-60-90-180 day plan converting the diligence findings into a value-creation agenda. Specific KPIs to track. Escalation triggers. Milestone-driven re-assessment schedule. The diligence findings become the monitoring spine.
For portfolio companies post-close, a structured monthly or quarterly KPI surveillance cadence drawing on the asset-class-specific dashboard. Milestone-driven re-assessment and event-driven inspection as required.
A complete dashboard requires metrics from at least four of the six families. Companies reporting only Throughput and Economics while omitting Utilisation, Reliability, Quality, and Lifecycle are typically optimising for visible top-line at the expense of underlying asset health. This is the most common pattern in scale-up diligence that later generates IC-level surprises.
The framework is calibrated to each vertical: the dominant risk profile, the unit economics structure, the regulatory perimeter, the KPI benchmarks, and the typical failure modes that have preceded the most common post-investment surprises. Each asset class has its own chapter.
AI cloud and neoclouds. Goodput, utilisation, contract-backing, orchestration dependency, model generation obsolescence.
CPOs and integrated networks. Utilisation by site, CMS reliability, network effects, regulatory charging mandates.
Deployed automation fleets. Uptime at cohort level, redeployability, software-hardware integration risk, spare-part economics.
Connected unattended retail. Site economics, network density, payment software reliability, route-density unit economics.
Fleet telematics and AV platforms. OTA dependency, data monetisation, autonomy milestone risk, regulatory software certification.
The framework has been applied to battery storage, connected medical devices and drone networks; those assessments are available on request.
The framework assumes commercial traction: deployed assets and real unit economics. Typically Series B onward.
The SOMA diligence framework draws on direct operational experience building and scaling intelligent-hardware systems at board level, across four companies, in six countries, over fifteen years. It reflects what was learned when these systems failed, and what it took to hold them together through PE transitions, global rollouts, and market stress.
The framework was not written from a data room.
A first conversation typically takes 20 minutes and covers the asset class, the stage of the company, the specific operational questions your IC will face, and whether SOMA is the right fit. No commitment required.
Or read how SOMA approaches credit assessment: The Three Clocks Problem