Industrial Safety AI: Ailytics Raises $7.4M to Watch Cameras You Own

Industrial Safety AI: Ailytics Raises $7.4M to Watch Cameras You Own

Ailytics Bets Industrial Safety AI Can Run on the Cameras You Already Own

The Singapore startup raised $7.4 million to take monocular 3D vision and procedure verification into 15 countries. The hard part for industrial safety AI is no longer detecting a missing helmet.

Ailytics raised 7.4 million dollars in a Series A led by Tin Men Capital, and the most interesting thing about the round is not the amount. It is who else came in: Safie, which describes itself as Japan’s largest cloud camera company, and one of Japan’s largest construction companies, alongside SEEDS, part of SG Growth Capital, and family offices. A camera vendor and a construction group investing together in a Singapore software company tells you where the growth is expected to come from.

Ailytics turns existing industrial CCTV into a safety and operations layer. It began in 2021 on Singapore public housing projects, and now runs deployments at NatSteel, PaxOcean, DHL, Rome’s Leonardo da Vinci International Airport and the Changi Airport Terminal 5 build. Since a 2.7 million dollar pre-Series A led by Tin Men in April 2024, it has gone from 70 projects in four countries to more than 400 deployments across 15 countries. The company says client churn sits below 1 per cent.

The technical bet

The product, Ailyssa, does something that sounds modest and is not: it derives three-dimensional spatial understanding from a single two-dimensional camera feed. Depth normally needs a second lens, LiDAR or radar. Ailytics rebuilds the ground plane from one view, finds the worker and the suspended load in each frame, projects the drop zone where the load would land, and measures the distance between the two in metres. That is what turns a generic object detector into an exclusion-zone alert a lifting supervisor can act on. The company says the system holds 85 to 99 per cent detection accuracy in the field, works through rain, dust and low light on cameras that may be a decade old, and goes live in under two weeks on hardware the site already owns.

That last point carries the commercial argument. Ailytics cites the figure that around 95 per cent of industrial sites have no active or AI-enabled monitoring, which means the cameras are there and nobody watches them. A vendor selling new sensors must justify hardware, installation and downtime. A vendor selling software onto an installed base argues only about the subscription. The company’s own framing, no rip-and-replace and any IP camera from two megapixels over RTSP, reads as a deliberate attack on the cost of adoption rather than on the accuracy of competitors.

Singapore Housing Board Projects

The company’s origin explains a lot about the product. It started on Singapore Housing Board projects, which is an unusual place for a computer-vision startup to begin: dense, regulated, high-rise construction with tight safety enforcement and a client that expects documentation. That environment forces software to work in heat, glare and rain on cameras installed years earlier, and it teaches a vendor to produce evidence rather than alerts. Singapore has also become a credible base for industrial AI that sells outward. The company now describes itself as building a global industrial AI business from Singapore, and the round’s geography, with Japanese strategics and Gulf expansion, suggests the next phase is less about Southeast Asia than about the markets where heavy industry spends on safety and where labour costs make automation pay.

From detection to verification

The more consequential shift is in what the system checks. Detecting a missing hard hat is a solved problem, and dozens of vendors solve it. Ailytics has been moving toward sequences: a lorry crane may not lift until its outriggers deploy, a moving plant needs a banksman, and refuelling should not involve a phone. Verifying those means reasoning across a series of frames about whether the right things happened, in the right order, in the right window. The company’s reasoning engine, AilyssaX, does that, and in June it deployed NVIDIA’s Cosmos 3 across the platform to add natural-language scene search, periodic scene reasoning for slow-moving hazards such as a spill or a drifting barrier, and spatio-temporal procedure checks.

The architecture is layered on purpose: fast specialised models handle high-volume perception, and the reasoning model runs selectively where its richer output justifies the compute. That matters, because the economics of a safety system depend on how often you invoke the expensive model.

Why the installed base is the moat

Ailytics is not alone in industrial safety AI, and the field has a history worth recalling. A wave of safety-video startups arrived around 2018 promising to watch every site, and many struggled, because detection accuracy in a demonstration and detection accuracy on a muddy site at night are different products, and because safety managers had no budget for a system that produced more alerts than they could act on. The survivors differentiated on the unglamorous parts: camera compatibility, deployment time, and whether the output fits an existing safety process. Ailytics’ emphasis on legacy hardware, two-week deployment and audit-ready reporting reads as a direct response to those failures. Its differentiation now rests on the reasoning layer, which is newer and less proven ground, and on the fact that its deployments concentrate in heavy industry rather than in general workplace monitoring.

Adoption economics deserve a note, because safety software competes with a manual process rather than with another vendor. A site that installs cameras and hires inspectors has a known cost; a site that installs software has a subscription plus the cost of responding to what it finds. Ailytics’ pitch turns on the second number staying small and the first being avoidable, and on the claim that customers recover the cost within months by reducing stop-work orders, audit downtime and fines. That is a plausible argument and an unproven one at portfolio scale, because the company does not publish contract values, retention by cohort, or the share of alerts that lead to an action.

The numbers and their provenance

Ailytics’ industrial safety AI performance claims come from the company. It reports a 60 to 80 per cent reduction in safety non-compliance, a sevenfold increase in hazard detection and a 20 per cent improvement in operational efficiency, with payback measured in months. The deployment and country counts are consistent across three years of coverage, which is a good sign: 70 projects in four countries in April 2024, more than 300 projects in 10 countries by April 2026, and 400-plus deployments in 15 countries now. The customer names are specific and checkable in principle. The efficiency figures are not, and this article found no independent audit of them. A reader should treat them as the vendor’s own measurements of its own deployments.

The uncomfortable questions

Three issues deserve more attention than a funding announcement gives them. The first is surveillance. A system that watches workers all shift and flags their behaviour is a workplace monitoring technology, and Ailytics’ privacy controls, including face and body blurring, on-premise or regional hosting and ISO 27001 certification, address where the video goes rather than whether the monitoring is appropriate. In jurisdictions with works councils or strong data-protection regimes, that is a consultation question before it is a procurement one.

The second is liability. If a site installs a system that promises to catch exclusion-zone breaches and then misses one, the question of who carries the consequence becomes interesting, and no vendor in this category has answered it publicly. The third is alert fatigue. A safety team that receives too many false positives stops reading them, and an accuracy range of 85 to 99 per cent is wide enough to matter at volume.

Industrial Safety AI: Ailytics Raises $7.4M to Watch Cameras You Own

What to watch

Two things will decide whether this round marks a step change or simply a good Series A. The first is whether the Japanese strategic investors convert into distribution. Safie sells cloud cameras into Japanese construction and industrial sites, and a large construction group is both a channel and a reference customer in one of the hardest markets to enter. The second is whether procedure verification works at scale. Detection is a commodity; verifying that a lift followed its sequence is a different product, and the company has been candid that the Cosmos-based capabilities sat in beta with a select group of customers as of June. If those ship and hold accuracy, Ailytics moves from selling alerts to selling compliance evidence, which is a better business.

A third question is quieter and matters just as much: whether the company can hold its deployment cost down as it enters the United States and the Gulf, where site conditions, labour rules and data-residency expectations all differ from Singapore’s.

What it signals

The round is a useful marker of where industrial safety AI is actually being bought. Not in the data centre and not in the chatbot, but on construction sites, at ports and in steel plants, where the value is a stopped incident and the buyer is a safety manager with a budget and a regulator. Ailytics’ proposition is deliberately unglamorous: use the cameras you have, show a supervisor something true in the next thirty seconds, and produce the report the auditor wants. That is a smaller story than a frontier model and a more durable one, because it depends on the installed base of industrial cameras rather than on the next release of anything. The risk is equally unglamorous. This category will consolidate, and the camera vendors are already at the table, which is precisely why two of them just bought into this one.