Giving sight to the blind spots in your warehouse / Michael Demes, Co-Founder & CEO @Sentics / ep.05
Industry 4.0 digitized the machines. The humans on the factory floor remained invisible.Michael Demes, CEO of Sentics, explains how optical AI is closing the $1 trillion blind spot in intralogistics.
The global industrial AI manufacturing market is accelerating at a rapid pace, projected to reach $230.95 billion by 2034 with a staggering 44.2% CAGR. While the industry has successfully digitized stationary assets like CNC machines and robotic arms, an estimated $1 trillion in uncaptured value remains locked on the factory floor.
One core issue lies in the analog nature of intralogistics: the unpredictable, unscripted movement of manual laborers, forklifts, and third-party logistics providers. Historically, operators treated safety as a pure compliance cost and efficiency as a separate margin driver. Nowadays, these topics converge in warehouse environments, driven by labor shortages, macroeconomic pressures, and the unreliability of legacy sensor stacks.
As supply chain resilience becomes a sovereign mandate, deploying physical AI to track spatial data is a requirement to protect workers and guarantee operational continuity.
Today’s guest is Michael Demes, CEO of Sentics, who is solving this problem by building an Optical Real-Time Localization System (ORTLS) that creates a live, privacy-compliant digital twin of the warehouse floor. This technology eliminates life-threatening blind spots without requiring workers to wear tracking tags.
Physical AI Closes the Analog Gap in Intralogistics
With the explosive growth of e-commerce and persistent labor scarcity, warehouses are no longer static storage facilities: they have evolved into highly dynamic, high-throughput fulfillment hubs. To meet throughput demands amidst a chronic shortage of skilled operators, enterprises are densely packing their floor plans with a mix of human-driven forklifts, AGVs, and pedestrian workers. This intermingling of man and heavy machinery significantly elevates collision risks, with the Occupational Safety and Health Administration estimating between 35,000 to 62,000 injury-related forklift accidents annually in the United States alone. Despite massive capital expenditures on digitizing the machines, the actual manual workflows of the human operators have remained largely opaque to enterprise software.
There were already significant digital trends focused on machinery, but manual operations were completely absent. The human element was entirely missing. The idea was to utilize existing CCTV cameras to create a digital replica of the physical world, enabling us to implement various optimizations and improve the environment.
While telematics can track the fuel consumption or general location of a forklift, it fundamentally cannot contextualize the chaotic, unstructured physical processes occurring around the vehicle. Without deterministic, real-time spatial awareness of how human laborers and manual assets navigate the shop floor, supply chain executives are forced to rely on lagging indicators and post-incident manual audits rather than proactive, predictive intelligence.
Compounding this technological blind spot are rigorous legislative frameworks and a harsh reality of workplace hazards. In Europe, directives like the EU Framework Directive 89/391/EEC establish strict mandates for proactive risk mitigation, forcing enterprises to seek robust compliance mechanisms to avoid punitive damages and facility shutdowns. Yet, despite these regulations, the warehousing and storage sector continues to experience disproportionately high fatality rates, registering 4.4 deaths per 100,000 workers recently, with vehicle-related incidents standing as the leading cause of death for material-moving workers.
By anchoring the deployment of physical AI directly to this life-critical mandate, industrial technology providers effectively bypass traditional software procurement friction. Once the optical network is installed to execute real-time collision avoidance, it simultaneously generates an immutable, digitized map of all floor activity. This transforms the physical AI layer from a passive physical barrier into an active, intelligent nervous system capable of diagnosing process bottlenecks and driving systemic facility efficiency.
Safety as the Entry Point for Physical AI
The genesis of Sentics represents a direct response to the digital asymmetry affecting manufacturing hubs. This massive oversight in the Industry 4.0 roadmap became apparent to the Sentics founding team during their academic tenure.
It was born out of university research where my co-founders and I met. We actually did a PhD in a really close-to-industry environment next to VW in Germany. And there we realized that actually the complete automation which is there on the shop floor and new technologies like AI were not used in production so frequently. There were already a lot of digital trends of machines and everything, but which was not there was all the manual operations.
This fundamental realization dictated the startup’s core value proposition: digitizing the negative space of the factory floor. By establishing an Optical Real-Time Localization System (ORTLS), Sentics aims to map industrial environments without relying on the physical transponders that legacy systems require. However, introducing comprehensive optical mapping presents immediate privacy and procurement friction. To bypass this resistance, the company anchored its initial commercial mandate strictly to life-saving collision avoidance, utilizing “Mission Zero” as a high-urgency wedge into enterprise facilities.
The intention is, you go to work to make your life possible, to earn some things, to evolve, to love your life. And then you’re going to work and lose your life. And then we thought, this might be the one thing which we want to change. And then we identified a lot of accidents with manual vehicles, like forklifts and humans. And so the idea was born.
This life-critical focus neutralizes the standard organizational resistance to AI deployment. By addressing immediate regulatory and liability concerns, specifically the mandate to reduce the severe accident rates associated with mixed-traffic environments, the company secures the necessary physical placement to capture comprehensive facility data. The deployment of physical AI infrastructure establishes the foundational sensor layer required to monitor all dynamic floor activity, shifting the system’s positioning from a passive physical barrier into an active, intelligent network.
The basic idea was always the same, having this digital replica of the world to do everything. But we thought it would be maybe a good idea to start with safety because saving someone’s life could be better.
By overcoming the localized blindness of vehicle-mounted sensors and the scaling limitations of wearable tags, Sentics positions its technology as a superior, holistic intelligence layer. This strategic positioning allows the subsequent expansion of software capabilities into efficiency metrics, spatial utilization, and automated documentation without requiring secondary hardware installations or additional capital expenditures.
Edge Computing Resolves Privacy Latency
Incumbent intralogistics systems predominantly utilize 4 GHz Ultra-Wideband radio signals or vehicle-mounted radar and lidar to establish active warning zones around moving machinery. While these systems excel in measuring precise distances, they suffer from structural blind spots and severe scaling limitations. Furthermore, tag-based architectures require total human compliance; if a visiting contractor or temporary worker steps onto the floor without an active transponder, they are entirely invisible to the safety network.
Using a camera just from a forklift, which has inherent difference disadvantages because if you have a forklift only looking in one direction, you cannot see the other way around. But on a shop floor, what is better than having the multi-camera vision set up on a shop floor to get all the information at one place.
Rather than attempting to equip every moving entity with a transmitting device, the platform utilizes a network of standard IP cameras mounted to the facility ceiling to establish an occlusion-free map of the environment. This transponder-less architecture shifts the computational burden away from the worker and onto the hardware, utilizing deep learning algorithms to continuously recognize and classify objects based entirely on their visual signature.
However, deploying high-definition computer vision in live industrial environments introduces severe computational and regulatory friction. Streaming massive volumes of raw video data to a centralized cloud architecture introduces network latency that is unacceptable in life-critical scenarios, where a heavy vehicle moving at speed requires millisecond intervention times. Simultaneously, the introduction of continuous optical monitoring triggers intense scrutiny from European labor unions and works councils, who rightly view raw video storage as an enabler of punitive algorithmic management and a direct violation of stringent privacy mandates like the General Data Protection Regulation.
And for this purpose, we are processing the video streams on the edge, on the shop floor, and also for data privacy. And only the metadata, so the X and Y coordinates of the objects, and objects like forklifts, humans, pallets, AGVs, trucks, whatever you can think about, we are streaming to the cloud so that our customers can see a view of this environment, can do replays, can do analytics on it, whatever you want.
To resolve the dual mandates of instantaneous actuation and absolute anonymity, the architecture is heavily dependent on localized edge computing. By processing the video streams directly on-site through specialized edge devices, the system achieves the ultra-low latency required to execute physical speed-throttling commands to the forklift seamlessly.
Physical AI Delivers Dual-Use Operational Value
The industrial sector notoriously resists recurring software fees attached to physical hardware deployments. Traditional intralogistics procurement categorizes safety equipment as a one-time capital expenditure, making the transition to a hardware-enabled software-as-a-service model a significant commercial hurdle. Enterprise buyers demand clear justification for ongoing expenditures when purchasing camera networks and vehicle interfaces. However, introducing high-friction, deeply integrated hardware fundamentally alters the churn dynamics characteristic of modern software: once a sensor network is physically wired into the braking systems of a multi-million-dollar forklift fleet and woven into the daily compliance workflows of a facility, the cost and disruption of ripping out that infrastructure becomes prohibitive.
From the product perspective, these days I’m really happy that we don’t have a pure software product because pure software product, like what is our disadvantage of course, it’s taking long to implement. But having all those processes always integrated in one customer, building trust, relationship with the customer. They are also not interested in opening this process and again and again. And so it’s pretty likely that the customers will just stick with us.
This entanglement provides the foundation for value-based pricing architectures. Rather than commoditizing the hardware layer through flat equipment fees, successful physical artificial intelligence providers capture a direct share of the yield they generate. Pricing scales dynamically based on the volume of actively managed assets, applying distinct license fees for vehicle braking modules or documentation agents operating across the floor. This structure aligns vendor revenue directly with the customer’s utilization rate, transforming the procurement conversation from a debate over hardware margins into a highly quantifiable return on investment calculation.
Securing that ROI purely through safety metrics presents a structural challenge for enterprise sales teams. While eliminating fatal collisions is a corporate imperative, calculating the precise financial return of an accident that did not occur relies on probabilistic models of avoided downtime and unfiled insurance claims. To unlock major operational expenditure budgets, the technology must exhibit dual-use characteristics, utilizing the exact same optical system installed for collision avoidance to simultaneously harvest hard efficiency gains. Plant managers require tools that directly improve overall equipment effectiveness and eliminate the hidden costs of supply chain disputes.
The customer liked it really much, which we implemented there. And they said like, but guys, if you can do it for the safety, why don’t we do it for efficiency? If you want to pay for it, why don’t we do it? So we went then for documentation processes for loading... They know which truck arrived with a license plate, everything documented, how many empty pallets, how many full pallets, safety straps installed, procedures analyzed, everything. And then also they’re using it for complaints in automotive industry especially... This is like pure money, which we can save them.
Converting compliance technology into a yield-generating asset radically accelerates the capital payback period. Automating the documentation of loading dock procedures eliminates massive inefficiencies associated with manual data entry and provides irrefutable visual evidence to reject costly logistics complaints. The digital twin generated by the physical AI layer serves as an omnipresent auditing tool, tracking pallet locations in real-time to prevent mis-shipments and identifying the precise moment inventory is damaged. By monetizing these friction points, the platform transitions from an environmental health and safety expense into core physical AI infrastructure, anchoring its valuation to the fundamental productivity of the facility.
Land and Expand through Partnerships
Penetrating brownfield industrial environments requires navigating a fragmented buying center. Historically, safety technology was procured almost exclusively by environmental, health, and safety managers primarily concerned with regulatory compliance and insurance audits. Because ORTLS systems also measure process sequences, space utilization, and bottlenecks, the modern sales cycle critically involves COOs focused on throughput and IT directors evaluating edge computing cybersecurity. This complexity frequently traps AI startups in “pilot purgatory”, where proofs-of-concept fail to scale into enterprise-wide rollouts due to integration friction and stakeholder misalignment.
First we’re starting most of the times with small pilot installation or with small systems only. Then after small systems, it’s scaling to bigger. Then on the path, right, you always have to convince works council, local managers, IT. But the good thing is if you have convinced them one shop floor, then it’s easier to go to the other shop floor.
Scaling within this sector demands a highly disciplined land-and-expand execution. By deploying a rapid proof-of-concept focused on a single high-traffic intersection, operators can demonstrate immediate efficacy and secure cross-departmental alignment. Once the initial facility proves the return on investment and passes the stringent data anonymization audits required by labor unions, the organizational friction of subsequent network expansions drastically decreases.
The primary bottleneck for scaling physical AI is the capital intensity and engineering friction of the physical installation. Enterprise buyers demand rapid time-to-value, yet retrofitting a massive legacy logistics hub with a novel sensor network traditionally involves downtime and unpredictable edge-case failures. Startups that can utilize standard commercial hardware and standardize their deployment playbooks possess a critical structural advantage in lowering customer acquisition costs and accelerating the path to recognized revenue.
Just yesterday, we commissioned a massive new shop floor covering over 50,000 square meters and operating 34 trucks. The installation for a facility of that size took only two months and went incredibly smoothly. We have now developed a highly effective product array, bringing us to the point where rapid, large-scale deployment is fully achievable.
Achieving this deployment velocity requires specialized implementation strategies that resemble forward-deployed engineering. By maintaining a highly proactive customer success model, where the vendor’s backend agents continuously monitor system health and proactively notify operators of obscured camera lenses before the client even detects an anomaly, startups build the trust necessary to secure long-term retention. This high-touch integration transforms the initial facility into a tangible sales asset, effectively utilizing active shop floors as live demonstration environments to convert highly skeptical enterprise prospects.
Despite these deployment efficiencies, direct enterprise sales remain inherently resource-intensive, placing a severe constraint on capital efficiency. To achieve exponential scale without depleting venture capital, vision platforms are pivoting to integrated ecosystem partnerships. By aligning with OEMs, systems integrators, and independent camera installers, the intelligence layer can be bundled natively into the procurement cycle of new forklift fleets or existing facility upgrades.
Yes, enterprise sales cycles are naturally long, but that is exactly why we are leveraging partnerships. We are currently establishing our first network of partners who will sell the solution on our behalf. As our product becomes more mature and versatile, the sales process becomes much smoother. We can simply propose that customers start by installing a few cameras, onboard their team, and then scale the network from there. This ability to ‘land and expand’ represents a massive opportunity for us.
This channel strategy effectively bypasses the gatekeepers of the intralogistics market. When a proprietary software platform is distributed and installed by the very entities that already own the customer relationship, the startup’s CAC plummets.
Funding Autonomous Fleet Orchestration
Industrial hardware deployments face a challenging growth trajectory: while the TAM for physical AI is massive, the capital intensity of scaling sensor networks severely restricts early-stage velocity. VC liquidity has recently rotated away from horizontal generative software toward applied industrial technologies, yet deep-tech startups still require patient capital to bridge the gap between pilot deployments and enterprise-wide agreements. For hardware-enabled software-as-a-service, surviving the extended commissioning phase is the primary existential threat, forcing operators to choose between slow, self-sustaining organic growth or raising aggressive rounds to capture the market before legacy incumbents adapt.
The biggest obstacle preventing us from scaling rapidly is simply capital. We raised a modest seed round two years ago, and we have managed it very effectively. If we continue on this path, we could achieve self-sustainability and grow slowly, which is acceptable. However, our current focus with our small core team is building the organizational structure to position ourselves for rapid expansion. I have a CRM full of leads that are on the verge of converting. The immediate goal is to demonstrate that we can accelerate our sales in a short timeframe and successfully fulfill every order.
The core technology has already crossed the proof-of-concept chasm; the immediate objective is injecting liquidity to lubricate the go-to-market engine, specifically by funding the creation of scalable sales collateral and expanding integration partnerships. Securing this growth capital enables Sentics to evolve its product scope over the next decade. Rather than remaining a passive safety monitoring layer, the platform accumulates the definitive spatial dataset required to orchestrate the movements of AGVs and human fleets.
Investment lens
The venture capital allocation framework for industrial automation is executing a hard pivot. The initial wave of Industry 4.0 investments saturated the market with horizontal predictive maintenance software and isolated robotics. However, the European intralogistics sector, constrained by structural labor shortages and stringent occupational safety directives, demands deterministic physical intelligence.
The AI software landscape is accelerating rapidly with the rise of autonomous agents. In a few months, someone might be able to build an AI bot to replicate basic software functions. However, what a software bot cannot do today is physically go to a customer’s shop floor and execute the hardware integration. Our ultimate vision is to become the operating system for the shop floor because we possess all the underlying data. We track the processes, we know when the lorries arrive, and we know exactly where the forklifts are.
Generative software applications face rapid commoditization and near-zero switching costs. Conversely, platforms that successfully endure the capital-intensive brownfield commissioning phase to embed edge-computing sensor networks into the mechanical and compliance workflows of an enterprise generate permanent retention. As heavy industry transitions toward fully autonomous fleets, legacy localization systems like ultra-wideband will prove insufficient for complex orchestration. The entities that capture the proprietary optical dataset of human and machine movement today are constructing the prerequisite infrastructure to direct the automated manufacturing ecosystems of the future.