DSTL - Audio in. Alpha out.

dstl

TERMINAL

LIBRARY

From Burned Sensors to AI-Powered Factories: How Theotics Monitors the Unmonitorable

Join

From Burned Sensors to AI-Powered Factories: How Theotics Monitors the Unmonitorable

Apr 9, 2026

Human By Design

1:04:04

THESIS

Theotics is replacing physical vibration sensors with electrical signature analysis, unlocking condition monitoring for the 30-40% of industrial machines operating in environments too harsh for any sensor to survive.

ASSET CLASS

SECULAR

CONVICTION

HIGH

TIME HORIZON

5 to 10 years

PREMISE

A persistent blind spot in industrial monitoring: 30-40% of critical machines operate where sensors cannot survive

Industrial reliability depends on condition monitoring, and for decades vibration sensors have been the dominant method. However, roughly 30-40% of machines in heavy industry operate in environments where installing a vibration sensor is either physically impossible or prohibitively expensive — steel blast furnaces at 1500°C, submerged sewage pumps, explosion-risk chemical zones, electric submersible pumps 3,000 meters underground. These are often the most critical assets in a facility, meaning the highest-consequence machines have the least visibility. Existing process data is volumous but was designed for process optimization, not reliability prediction, and performs poorly when repurposed for condition monitoring. This structural gap between monitoring capability and monitoring need has persisted for decades because solving it requires simultaneously building custom high-frequency sensors, signal processing pipelines, and AI models — a combination that pure software startups and traditional sensor companies have not attempted.

MECHANISM

Electrical signature analysis at 20,000 Hz transforms every motor into a virtual vibration sensor for its entire drivetrain

Theotics discovered — initially by chance through a Dutch railway project constrained from installing physical sensors — that mechanical faults detectable by vibration sensors are also visible in high-frequency current and voltage sine waves. By installing a dedicated sensor at the electrical supply point (not on the machine itself) measuring at 20,000 samples per second, they capture electrical signatures that encode mechanical condition, electrical faults, energy efficiency, and operational state simultaneously. The critical technical insight was that both current and voltage are required; an initial 300-sensor deployment at Heineken with current-only data failed, and the team had to pivot with limited remaining capital to add voltage measurement. The system layers signal processing to handle noisy industrial environments, anomaly detection using reconstruction-based models that learn healthy operating patterns across variable set points, fault classification matching fingerprints to specific failure modes like bearing degradation or cavitation, and a human-in-the-loop verification layer where mechanical and electrical engineers review alerts before they reach clients. This human layer served a dual strategic purpose: it compensated for undertrained algorithms in early deployments while building the data flywheel — better product leads to more clients, more clients generate more training data, more data improves algorithms. The company now monitors tens of thousands of machines across steel, wastewater, oil and gas, chemicals, and mining, with each engineer managing approximately 2,000-3,000 assets, and claims 90%+ detection accuracy and roughly 50% market share in the hard-to-reach asset monitoring category.

OUTCOME

Theotics is positioning as the essential data infrastructure layer for autonomous smart factories

The immediate outcome is a globally unique, scaled condition monitoring service for previously unmonitorable industrial assets, delivered as a managed service with 24/7 coverage across Asia, Europe, and the Americas. But the larger strategic outcome is a transformation from condition monitoring provider to foundational data infrastructure for smart factories. As industrial facilities move toward digital twins and autonomous operations, those abstraction layers require real-time machine-level data to function — digital twins without data fall flat. Theotics's electrical signature analysis is positioned as the most scalable and elegant method to make traditional machines smart, connected, and context-aware. The company is expanding into on-premise deployments for regions and industries requiring local data processing, and broadening from condition monitoring into energy and performance data provision. The competitive moat combines hardware IP, proprietary signal processing tuned to custom sensors, trained AI models built on a dataset no competitor can replicate at scale, and deep domain expertise from years of in-factory iteration — a full-stack approach that pure software competitors cannot easily replicate.

NECESSARY CONDITION

Regulatory frameworks must remain permissive to innovation (avoiding the 'European' model) and open source development must remain unencumbered by downstream liability.

If we would have had an alternative, we would have chosen another path quite a few times, I reckon. So, burning bridges.

30:45

RISK

Steel Man Counter-Thesis

Somotics has built a genuinely innovative product addressing a real gap in industrial condition monitoring. However, the strongest counter-thesis is that the company is constructing a category-of-one that validates a market opportunity without possessing the structural defenses to capture it long-term. The underlying science of electrical signature analysis has been publicly documented since the 1970s and is not patentable in its fundamental form. The company's differentiation lies in engineering integration (sensor + signal processing + AI + service), but this is an execution moat, not a structural one. Industrial automation giants like Siemens, ABB, and Schneider Electric possess orders-of-magnitude advantages in hardware engineering, factory-floor distribution, existing customer trust relationships, and capital for R&D. Once Somotics proves the ESA-at-scale category is commercially viable — which it is now actively doing through public case studies and client testimonials — these incumbents can build or acquire equivalent capabilities and bundle them into existing monitoring platforms at near-zero marginal distribution cost. The data flywheel advantage is real but time-limited: industrial fault signatures are finite in variety, and a well-resourced competitor with access to even a fraction of the deployment base could train equivalent models within 2-3 years. Meanwhile, the company's managed service model, while currently a quality moat, creates a structural margin ceiling that limits the capital available for the R&D arms race that will inevitably follow category validation. The historical precedent is instructive: companies like Nest (smart thermostats), Fitbit (wearable health), and Blue River Technology (agricultural AI) all validated new categories only to be acquired by incumbents (Google, Google, John Deere respectively) who recognized that distribution and integration advantages ultimately trump first-mover innovation. Somotics may be building toward a successful acquisition rather than an independent platform — which is a fine outcome for founders but a fundamentally different risk-return profile for growth-stage investors betting on platform-scale returns.

RISK 01

Electrical Signature Analysis Accuracy Ceiling in Multi-Variable Industrial Environments

THESIS

The core thesis rests on the claim that current and voltage sine wave analysis can substitute for direct vibration measurement on hard-to-reach assets. However, the CEO himself acknowledges that most machines operate at a variety of set points and work points, creating a very unstable environment for anomaly detection. As factories increase automation complexity, variable frequency drives, regenerative braking, and power quality disturbances from adjacent equipment could inject noise patterns that mimic or mask genuine mechanical fault signatures. The fundamental physics of inferring mechanical degradation from an electrical proxy signal may hit an accuracy ceiling well before matching the reliability of direct vibration sensing, particularly for incipient faults with subtle signatures. The 90%+ detection accuracy cited is impressive but leaves a meaningful gap where critical failures could be missed on the most dangerous, hard-to-reach assets — precisely the ones where failure consequences are most catastrophic.

DEFENSE

The company partially addresses this through the human-in-the-loop model, where mechanical and electrical engineers review a subset of alerts before they reach clients, compensating for algorithmic limitations. The CEO also acknowledges that remaining useful life prediction is more art than science and relies heavily on engineer judgment. However, the defense is operational (human labor) rather than fundamental (technological breakthrough), meaning it does not scale frictionlessly and introduces its own failure modes around human fatigue and consistency.

RISK 02

Sensor Incumbents and Hyperscalers Closing the Moat via Acquisition or Replication

THESIS

The company's moat is described as a combination of integrated hardware-software stack, accumulated training data at scale, and first-mover advantage in the ESA category. However, major industrial automation incumbents (Siemens, ABB, Schneider Electric, Emerson) already have deep customer relationships, existing sensor infrastructure, and massive R&D budgets. Any one of these players could acquire an ESA startup or build an internal ESA capability and bundle it into their existing condition monitoring suites at marginal cost. Similarly, platforms like Palantir — which the interviewer explicitly mentions as a potential overlap — could partner with hardware providers to replicate the stack. The CEO's claimed 50% market share is of a nascent, still-small category; category creation risk means the moment the category becomes validated and large, the incumbents will enter aggressively. The very act of proving the market invites well-resourced competitors who can outspend on hardware R&D, sales, and distribution.

DEFENSE

The CEO frames competitors as fellow category builders who should collectively focus on quality and underpromise/overdeliver. This is a cooperative framing that entirely sidesteps the existential threat from industrial automation incumbents with 10-100x the resources. There is no discussion of defensibility against a Siemens or ABB entering the space, no mention of switching costs once deployed, no discussion of long-term contractual lock-in, and no acknowledgment that the integrated hardware-software approach that was hard for startups to replicate is trivial for vertically integrated industrial conglomerates. The trade secret vs. patent balance mentioned is also precarious — trade secrets offer no protection against independent development by well-funded competitors with access to the same ESA literature from the 1970s.

RISK 03

Scaling Economics and the Human-in-the-Loop Bottleneck

THESIS

The business model is presented as a managed service with 24/7 global engineering coverage. Currently, one engineer handles approximately 2,000-3,000 assets, and the company monitors tens of thousands of machines. As the company scales toward the ambition of being the data infrastructure layer for all smart factories globally, the number of monitored assets could grow by orders of magnitude. The human-in-the-loop model — which is described as essential for maintaining quality, compensating for undertrained algorithms, and handling the remaining useful life prediction problem — creates a fundamentally linear cost structure embedded within what is pitched as a scalable platform business. If the algorithms do not improve fast enough to reduce human dependency, gross margins compress as the company scales. If the company removes humans too aggressively to improve margins, detection quality degrades and client trust erodes. This tension between SaaS-like scalability aspirations and managed-service operational reality is a structural business model risk that directly affects valuation multiples and capital efficiency.

DEFENSE

The CEO describes the human-in-the-loop as a competitive advantage and a quality differentiator, but does not address the inherent tension between this labor-intensive model and the stated ambition to be the scalable data infrastructure layer for global smart factories. The flywheel argument — more clients yield more data yield better algorithms — is invoked but no concrete evidence is offered that the ratio of assets-per-engineer is meaningfully improving over time or that there is a credible path to dramatically reducing human intervention. The CEO explicitly states that remaining useful life estimation remains more art than science, suggesting the most valuable part of the service may resist automation indefinitely.

ASYMMETRIC SKEW

Downside is concentrated in two scenarios: (1) incumbent entry compresses margins and market share after the category is validated, capping returns at an acquisition-level outcome rather than platform-scale outcome, and (2) the human-in-the-loop model fails to scale, creating margin compression that makes the business increasingly capital-intensive at precisely the moment it needs to outspend larger competitors. Upside requires that the company's data and integration advantages compound faster than incumbents can replicate them, that the ESA category grows large enough to sustain an independent platform, and that the algorithm-to-human ratio improves dramatically. The skew is moderately unfavorable for growth-stage investors seeking 10x+ returns: the most probable outcome is a successful acquisition at 3-5x, with tail risk of margin erosion and competitive displacement if the company tries to remain independent in a category that attracts well-capitalized entrants.

ALPHA

NOISE

The Consensus

The market believes that industrial condition monitoring is a mature, well-served category dominated by vibration-based sensor systems. The consensus view is that predictive maintenance requires physical sensor installation on or near equipment, and that the primary innovation frontier is in the software/AI layer — better algorithms applied to existing sensor data, digital twins, and cloud-based analytics platforms. Large incumbents and well-funded startups compete primarily on software sophistication, and the assumption is that sensor hardware is a commodity input to the analytics stack. The market also broadly assumes that AI/ML models trained on process data or vibration data are sufficient for reliability insights, and that scaling predictive maintenance is principally a software scaling problem.

The market's logic chain is: install sensors on machines → collect vibration/process data → apply AI/ML analytics → predict failures → reduce downtime costs. The implicit causal assumption is that more data and better algorithms are the primary levers for improving predictive maintenance outcomes. Digital twins and autonomous factory concepts are seen as the next evolution, with the assumption that sensor data availability is a solved or readily solvable problem. Competitors focus on the software/AI layer because that is where venture capital flows and where scalability appears most achievable.

SIGNAL

The Variant

Jaggers believes the market has a fundamental blind spot: 30-40% of critical industrial machines operate in environments where physical sensor installation is impossible or economically prohibitive (extreme temperatures, submersion, corrosive environments, explosion-risk zones). This means the entire vibration-based monitoring paradigm structurally cannot address roughly a third of the market's most critical assets — and these hard-to-reach assets are disproportionately the most expensive to lose. His variant view is that the real bottleneck is not algorithmic sophistication but fit-for-purpose data generation, and that electrical signature analysis (ESA) — monitoring current and voltage sine waves at high frequency from a safe, remote installation point — constitutes a fundamentally different and superior approach for this underserved segment. He further believes the market underestimates how difficult the full-stack integration (proprietary hardware sensor + signal processing + AI + human-in-the-loop service) is to replicate, creating a durable competitive moat that software-only competitors cannot breach.

Jaggers' causal logic inverts the conventional stack: the binding constraint is not the algorithm but the data. Process data was developed for process optimization, not reliability monitoring — repurposing it produces mediocre reliability insights. Vibration data is excellent for reliability but physically cannot be collected on 30-40% of critical assets. Therefore, the correct causal chain is: solve the data generation problem first (proprietary high-frequency electrical sensor measuring at 20,000 Hz) → apply domain-specific signal processing to extract mechanical and electrical fault signatures from current/voltage waveforms → layer AI for anomaly detection and fault classification → add human-in-the-loop verification to compensate for early-stage algorithm limitations and build client trust → use scale deployments to generate training data flywheel → establish category dominance before competitors can replicate the full stack. Critically, Jaggers argues that the hardware-software integration requirement is what deters competitors — most startups only solve one piece — and that the human-in-the-loop service layer is not a stopgap but a structural differentiator that simultaneously improves product quality, builds client trust, and generates proprietary training data.

SOURCE OF THE EDGE

Jaggers' claimed edge rests on three pillars: (1) hands-on operational experience discovering that vibration sensors fail in harsh industrial environments, which gave him the problem insight before it was obvious to software-focused competitors; (2) the full-stack integration of proprietary hardware, signal processing, AI, and human expert services, which creates compound difficulty for replication; and (3) a data flywheel from tens of thousands of deployed machines generating proprietary high-frequency electrical data that no competitor possesses at comparable scale. The credibility assessment is mixed but tilts strongly positive. The operational experience edge is genuine and verifiable — the Heineken V1 failure (300 current-only sensors that needed voltage data) and the Dutch railway project origin story are specific, detailed, and carry the texture of lived experience rather than constructed narrative. The full-stack integration claim is structurally credible: hardware development is genuinely expensive, slow, and incompatible with typical venture-backed software startup timelines. His claim of approximately 50% market share in the hard-to-reach asset monitoring category is somewhat self-serving and unverifiable, but his specific observation that competitors lack comparable public client testimonials and scale deployment evidence is a reasonable inferential signal. The weakest element of the edge claim is the suggestion that the underlying science (ESA has existed since the 1970s) is somehow uniquely accessible to Samotics — in reality, the science is well-known, and the edge is in engineering execution and scale, not in proprietary scientific insight. However, execution-and-scale moats in industrial contexts are historically among the most durable. The edge appears genuine: it is a compounding operational advantage built through painful iteration, not a narrative overlay on commodity technology.

CONVICTION DETECTED

• We are the global leader in our field • We would be unique, globally unique • ESA is the most elegant, the most scalable way to get essential data about machines • We are the only ones to date that deliver this at massive scale • The problem that hasn't been solved for hundreds of years has now a solution • If this would fail the company would fail • Failure was not an option • We had no alternative going for us and there was no offramp • I would say 50% market share • We have 90% or more detection accuracy • 30 to 40% of machines in a factory operate in places where it's either impossible or prohibitively expensive to install a sensor

HEDGE DETECTED

• By chance to be fair • It could have gone gone wrong a number of times • To be fair, that is something where we use experience — more art than science • I don't want to misspeak, so let me stick to what I know which are the broad terms • Maybe I'm generous. Maybe it's more ambition than reality or I could be underselling ourselves • A lot of it was luck as well because you make those decisions based on incomplete information maybe even wrong information • There's also survivor bias in place • Was that wisdom? Well, a part of it obviously, but a lot of it was luck • We should stay vigilant • I think I have a lot to learn The ratio of conviction to hedging reveals a speaker who is genuinely certain about his company's technical differentiation and market position but intellectually honest about the role of contingency in reaching that position. The hedging concentrates on historical path dependency (luck, survivor bias, incomplete information at decision points) rather than on current product capability or competitive standing. This is the pattern of a founder who has earned conviction through painful iteration rather than one performing certainty for an audience — the hedges make the conviction markers more credible, not less. Substantial weight should be placed on the thesis, particularly the claims about technical moat and market position, while appropriately discounting the more expansive forward-looking claims about platform evolution and market share dominance.