Ask a warehouse operator in Singapore what keeps them up at night, and forklifts come up before fires, floods, or fraud.

They should. Between 2022 and 2023, vehicular incidents were the leading cause of fatal workplace injuries in Singapore, and one in four of those deaths involved a forklift, as per the Ministry of Manpower (MOM).

MOM didn’t wait for the data to accumulate further. In November 2024, it introduced enhanced forklift refresher training requirements, on top of a 2015 circular on the safe use of storage racks, following fatal rack-collapse cases.

That’s the backdrop on warehouse safety in Singapore.

But here’s the forecast – within five years, every warehouse operating in Singapore will run on some form of AI-based safety monitoring. Not because it’s trendy. Because three things are converging at once, and none of them is slowing down.

The rules are no longer satisfied by good intentions

Singapore’s Workplace Safety and Health Act asks employers to take “reasonably practicable” steps to protect workers. For years, that meant a folder of risk assessments, a monthly walk-through, and a rack inspection schedule with daily visual checks, weekly compiled reports, and annual professional audits.

On paper, it works. In practice, a blocked emergency exit gets cleared for an audit and drifts back within days. PPE compliance holds in the morning shift and slips by the afternoon. A bent upright from a forklift impact goes unnoticed until the next scheduled inspection, weeks later.

Non-compliance isn’t a soft cost anymore. Fines under WSH regulations can run up to SG$500,000 for corporate entities, and severe violations trigger a Stop Work Order — a warehouse shutdown overnight, mid-fulfilment cycle. The Workplace Safety and Health Council has already named warehousing an accident hotspot, specifically around forklift use and loading operations. Regulators are asking for continuous, demonstrable compliance now, not a clean paper trail collected once a quarter.

That’s a standard periodic inspection that was never built to meet — and it’s precisely the standard AI-based monitoring is built to meet instead, because it doesn’t inspect on a schedule. It watches continuously, which is the only way “reasonably practicable” starts to mean something real rather than something documented after the fact.

Also Read: Why your data warehouse is just a very expensive attic

The floor is shrinking while the volume grows

Singapore’s freight and logistics market is worth roughly US$26 billion this year and is on track to hit over US$35 billion by 2031, growing at more than 6 per cent annually, with warehousing itself among the fastest-growing segments, propelled by e-commerce and just-in-time stocking demand.

That growth is landing on a footprint that isn’t expanding at the same rate. Land is scarce and expensive. Warehouses are going vertical, consolidating operations, and running leaner headcounts than the volume suggests they need. A supervisor who once covered one aisle now effectively covers three.

More product moving through less space, watched by fewer people, is a formula periodic manual checks were never designed to handle, and it’s exactly the gap AI is being built to close. A camera system that already exists on-site doesn’t need more headcount to watch more aisles; it just needs to be given the job.

The technology stopped being the limitation

The biggest change, if we consider the last five years in warehouse safety technology, is not that the cameras have become better. It is that the purpose of the camera has evolved.

For most warehouses, CCTV has historically been a forensic tool. Footage becomes valuable after something has happened, like an injury, a collision, damaged stock or a disputed near miss. Someone identifies the approximate time, retrieves the recording and reconstructs the event.

Computer vision-based monitoring changes that sequence.

Instead of waiting for a supervisor to review footage, AI models can analyse visual conditions as operations unfold and identify predefined risk patterns. That distinction matters because many warehouse incidents develop over seconds rather than hours, leaving very little time for conventional supervision to intervene.

The technical barriers to doing this at operational scale have also fallen. The AI warehouse monitoring systems can increasingly work with existing surveillance infrastructure like CCTVs on site, while edge computing allows safety-critical processing to happen close to where footage is generated rather than depending entirely on cloud connectivity.

But detection itself may prove to be only the first stage.

Warehouses generate thousands of visual observations across shifts, aisles and loading areas. Analysed over time, those observations can reveal something more valuable than individual violations, for example, the patterns of exposure.

Also Read: Boardrooms to warehouses: How SEA leaders can build cyber resiliency from top-down

This is where newer technological developments like vision-language models (VLMs) and agentic AI systems could push warehouse safety further. Rather than simply classifying an event, these systems are beginning to interpret sequences of activity, retrieve relevant evidence and help safety teams identify recurring conditions across larger volumes of operational data.

That changes the role of collected footage on the floor again. It translates from evidence of what happened to detection of what is happening, and eventually to intelligence about what is likely to keep happening unless the underlying condition changes.

Warehouses generate an enormous volume of operational data every second, but most of it has traditionally gone unused because it couldn’t be analysed in real time. AI changes that by transforming visual information into measurable safety intelligence, allowing organisations to intervene before isolated events develop into systemic risks.

Five years is the generous estimate

None of this replaces a supervisor’s judgment or a good toolbox talk. What it removes is the lag between a hazard forming and someone catching it — where most warehouse incidents live.

Put the three forces together — regulators demanding continuous proof, a market outgrowing its floor space and headcount, and AI infrastructure finally cheap and local enough to run on cameras a warehouse already owns — and five years starts to look conservative, not ambitious.

The operators moving now aren’t betting on a trend. They’re the ones who read the regulatory notices, looked at the growth numbers, and did the math first.

—

The post Why every warehouse in Singapore will run on AI safety monitoring within five years appeared first on e27.